<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Hbm on Korea Invest Insights</title><link>https://koreainvestinsights.com/series/hbm/</link><description>Recent content in Hbm on Korea Invest Insights</description><generator>Hugo -- gohugo.io</generator><language>en</language><copyright>koreainvestinsights.com · @korea_invest_insights</copyright><lastBuildDate>Wed, 22 Jul 2026 17:37:01 +0900</lastBuildDate><atom:link href="https://koreainvestinsights.com/series/hbm/feed.xml" rel="self" type="application/rss+xml"/><item><title>FADU's Q2 Surprise and Its Real Moat: Controller-Firmware Integration Matters More Than Gen6 Speed</title><link>https://koreainvestinsights.com/post/fadu-2q26-earnings-surprise-gen6-essd-controller-narrow-moat-2026-07-22/</link><pubDate>Wed, 22 Jul 2026 17:32:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/fadu-2q26-earnings-surprise-gen6-essd-controller-narrow-moat-2026-07-22/</guid><description>&lt;p&gt;FADU reported preliminary Q2 2026 revenue of KRW 73.155bn and operating profit of KRW 16.267bn. Revenue beat local consensus by 16.1%, while operating profit beat by 89.2%. The operating margin reached 22.2%, above the 20.5% upper bound of this site&amp;rsquo;s earlier bull case.&lt;/p&gt;
&lt;p&gt;The quarter removed much of the uncertainty around whether FADU can generate profit from its Gen5 controller business. The stock market immediately moved to a harder question. FADU closed down 9.3% at KRW 70,400 on July 22, while foreigners sold a net KRW 70.3bn over the latest ten sessions. Investors now want evidence for 2027 growth: Gen6 mass production, commercial QLC adoption and customer diversification.&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Related research
This report reviews the outcome of the &lt;a class="link" href="https://koreainvestinsights.com/post/fadu-2q26-earnings-preview-essd-controller-moderate-beat-2026-07-04/" &gt;FADU Q2 2026 earnings preview&lt;/a&gt; and follows the longer-duration framework in &lt;a class="link" href="https://koreainvestinsights.com/post/fadu-ai-infra-storage-bottleneck-p-q-new-segment-2026-06-02/" &gt;What FADU Must Prove on Price, Volume and New Markets&lt;/a&gt;. The first flow-based comparison with Sandisk is available in &lt;a class="link" href="https://koreainvestinsights.com/post/fadu-sandisk-ai-storage-korea-beta-jeju-semiconductor-2026-05-31/" &gt;Is FADU Korea&amp;rsquo;s Sandisk Beta?&lt;/a&gt;. See also the &lt;a class="link" href="https://koreainvestinsights.com/page/exclusive-analysis-hub/" &gt;Exclusive Analysis Hub&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/page/korea-semiconductor-equipment-ip-hub/" &gt;Korea Semiconductor Value Chain Hub&lt;/a&gt; and &lt;a class="link" href="https://koreainvestinsights.com/page/korea-semiconductor-hbm-kospi-hub/" &gt;AI HBM Hub&lt;/a&gt;.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;FADU&amp;rsquo;s core advantage is not &lt;code&gt;PCIe Gen6 speed&lt;/code&gt; by itself. It is the integration of controller ASIC and firmware that makes NAND behave with low power, consistent latency and high endurance under a customer&amp;rsquo;s workload.&lt;/li&gt;
&lt;li&gt;The current moat is &lt;strong&gt;narrow but expandable&lt;/strong&gt;. Gen5 commercialization is visible in revenue and OCP participation. Gen6 customer qualification and production, 64TB/128TB QLC design wins and customer diversification remain unproven.&lt;/li&gt;
&lt;li&gt;Q2 revenue of KRW 73.16bn beat consensus of KRW 63.0bn by 16.1%. Operating profit of KRW 16.27bn beat KRW 8.6bn by 89.2%. The 22.2% margin is the more important signal.&lt;/li&gt;
&lt;li&gt;The preliminary filing does not disclose net income, product mix, gross margin, cash flow or receivables. Revised 2026 EPS of roughly KRW 1,930 and a 36.5x P/E are analytical estimates, not reported or consensus figures.&lt;/li&gt;
&lt;li&gt;KRW 70,400 is close to the low end of the 2027 base-case range of KRW 72,000-86,000. The stock is not cheap on 2026 earnings alone, but modestly discounted if 2027 growth is achieved.&lt;/li&gt;
&lt;li&gt;Fundamentals improved while flows remain weak. Foreigners sold KRW 70.3bn over ten sessions, and program selling reached KRW 17.9bn during the three sessions after the earnings release.&lt;/li&gt;
&lt;/ul&gt;
&lt;hr&gt;
&lt;div class="thesis-callout"&gt;
 &lt;div class="thesis-callout__label"&gt;Bottom line&lt;/div&gt;
 &lt;div class="thesis-callout__body"&gt;
 FADU has proved that it can commercialize a high-performance Gen5 enterprise SSD controller. It has an option to become a broader platform in Gen6 and QLC, but that option requires customer qualification, production orders, lower customer concentration and independent performance evidence.
 &lt;/div&gt;
&lt;/div&gt;
&lt;hr&gt;
&lt;h2 id="1-q2-the-margin-beat-matters-more-than-the-revenue-beat"&gt;1. Q2: The Margin Beat Matters More Than the Revenue Beat
&lt;/h2&gt;&lt;p&gt;FADU filed preliminary consolidated results on July 20. These numbers have not completed external auditor review and can change in the half-year report.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Metric&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Q2 2026 preliminary&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Consensus&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Difference&lt;/th&gt;
 &lt;th style="text-align: right"&gt;QoQ&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 73.16bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 63.0bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+16.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+22.9%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Operating profit&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 16.27bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 8.6bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+89.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+111.6%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Operating margin&lt;/td&gt;
 &lt;td style="text-align: right"&gt;22.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;13.7%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+8.6pp&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+9.3pp&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;First-half revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 132.69bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+209.3% YoY&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;First-half operating profit&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 23.95bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Turned profitable YoY&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Sources: &lt;a class="link" href="https://dart.fss.or.kr/dsaf001/main.do?rcpNo=20260720900116" target="_blank" rel="noopener"
 &gt;DART preliminary earnings filing&lt;/a&gt;; local consensus snapshot based on Naver WiseReport as of July 22, 2026.&lt;/p&gt;
&lt;p&gt;The comparison with the July 4 preview is revealing.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Reference case&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Revenue&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Operating profit&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Operating margin&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Prior midpoint&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 70.5bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 11.8bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;16.7%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Prior bull-case upper bound&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 75.0bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 15.4bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;20.5%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Preliminary result&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 73.16bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 16.27bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;22.2%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Difference vs midpoint&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+3.8%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+37.9%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+5.5pp&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Revenue fell just below the prior KRW 75bn bull threshold, but operating profit exceeded the bull-case upper bound. The surprise is therefore not simply more units. It likely reflects a favorable controller mix, fixed-cost absorption and contract economics. Those drivers must still be confirmed in the half-year filing.&lt;/p&gt;
&lt;p&gt;Management said disclosed new orders exceeded KRW 340bn during the first half, a new server OEM customer was secured, and second-half revenue and profit should exceed the first half. That is useful guidance, but not audited evidence. Controller mix, gross margin, operating cash flow and receivables are the next checks.&lt;/p&gt;
&lt;h2 id="2-what-an-enterprise-ssd-controller-actually-does"&gt;2. What an Enterprise SSD Controller Actually Does
&lt;/h2&gt;&lt;p&gt;NAND is the storage medium. The controller combines traffic management, error correction and power management. It translates host commands into physical NAND operations, distributes workloads across channels, and manages performance, endurance and data integrity.&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"&gt;&lt;code class="language-text" data-lang="text"&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;CPU and GPU
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; ↓ PCIe/NVMe commands
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;FADU controller ASIC + firmware
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; ├─ maps logical addresses to physical NAND locations
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; ├─ schedules reads and writes across NAND channels
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; ├─ manages ECC, bad blocks and endurance
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; ├─ controls garbage collection and WAF
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; ├─ manages power, thermals and power-loss protection
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; └─ provides QoS, security and telemetry
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; ↓
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;NAND packages
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;Peak speed is not the only metric. Cloud customers care about performance consistency under congestion, throughput within a power envelope, and data integrity during power failure. Very slow outlier requests can damage service quality even when average latency looks good. This makes p99.99 tail latency an important benchmark.&lt;/p&gt;
&lt;h2 id="3-why-the-controller-matters-more-after-pcie-gen6"&gt;3. Why the Controller Matters More After PCIe Gen6
&lt;/h2&gt;&lt;p&gt;PCIe 6.0 introduced 64GT/s signaling, PAM4, FEC and FLIT mode, doubling Gen5 bandwidth. The practical ceiling for an x4 SSD is around 28GB/s. As the interface gets faster, the bottleneck shifts toward NAND parallelism, power, thermal control and firmware scheduling. &lt;a class="link" href="https://pcisig.com/pci-express-6.0-specification" target="_blank" rel="noopener"
 &gt;PCI-SIG PCIe 6.0&lt;/a&gt;&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Technology shift&lt;/th&gt;
 &lt;th&gt;Server requirement&lt;/th&gt;
 &lt;th&gt;Controller challenge&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;PCIe Gen5 to Gen6&lt;/td&gt;
 &lt;td&gt;2x bandwidth, less GPU idle time&lt;/td&gt;
 &lt;td&gt;PAM4/FEC handling and NAND scheduling&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AI inference and RAG&lt;/td&gt;
 &lt;td&gt;High random reads, low tail latency&lt;/td&gt;
 &lt;td&gt;QoS, queue management and caching&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;High-capacity QLC&lt;/td&gt;
 &lt;td&gt;122TB and above, lower cost per TB&lt;/td&gt;
 &lt;td&gt;ECC, write amplification and endurance&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FDP and ZNS&lt;/td&gt;
 &lt;td&gt;Host-directed data placement&lt;/td&gt;
 &lt;td&gt;Less garbage collection, consistent performance&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;E1.S and E3.S&lt;/td&gt;
 &lt;td&gt;Higher rack density and cooling efficiency&lt;/td&gt;
 &lt;td&gt;Power and thermal throttling&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Multi-tenant cloud&lt;/td&gt;
 &lt;td&gt;Per-customer isolation&lt;/td&gt;
 &lt;td&gt;Virtualization, multiple PFs, consistent latency&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Stronger security&lt;/td&gt;
 &lt;td&gt;Firmware and data protection&lt;/td&gt;
 &lt;td&gt;Root of Trust, encryption and attestation&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;AI enterprise SSD demand is splitting into two segments. High-performance TLC serves checkpoints, RAG, vector databases and AI data pipelines. High-capacity QLC serves data lakes, object storage and read-heavy inference data.&lt;/p&gt;
&lt;p&gt;QLC improves density and cost but has weaker write performance, endurance and error characteristics. This increases the value of ECC, garbage collection, wear leveling and FDP. Micron is already shipping a 245TB QLC SSD after launching 122TB-class products. &lt;a class="link" href="https://investors.micron.com/news-releases/news-release-details/industry-leading-245tb-micron-6600-ion-data-center-ssd-now" target="_blank" rel="noopener"
 &gt;Micron 245TB 6600 ION&lt;/a&gt;&lt;/p&gt;
&lt;h2 id="4-what-fadu-has-already-proved"&gt;4. What FADU Has Already Proved
&lt;/h2&gt;&lt;h3 id="41-architecture-and-firmware"&gt;4.1 Architecture and Firmware
&lt;/h3&gt;&lt;p&gt;FADU describes an architecture built around a high-speed point-to-point network, hardware offload, light cores and simpler firmware instead of a conventional many-core shared interconnect. The objective is to reduce software overhead and memory movement.&lt;/p&gt;
&lt;p&gt;The company&amp;rsquo;s Gen5 FC5161 specifications show sequential reads of 13.7GB/s and random reads of 3.3M IOPS. FADU also presents favorable write throughput and performance-per-watt comparisons, but these are company tests rather than independent, same-condition benchmarks. &lt;a class="link" href="https://kind.krx.co.kr/external/dst/irReference/18509/1Q26%20IR%20Book_KOR_260512.pdf" target="_blank" rel="noopener"
 &gt;FADU Q1 2026 IR&lt;/a&gt;&lt;/p&gt;
&lt;h3 id="42-fdp-and-the-ocp-ecosystem"&gt;4.2 FDP and the OCP Ecosystem
&lt;/h3&gt;&lt;p&gt;The FC5161-based Gen5 SSD is listed in the OCP Marketplace. OCP describes it as the first OCP SSD to support FDP, with 64 Physical Functions, power-loss protection, security, telemetry and thermal management. &lt;a class="link" href="https://www.opencompute.org/products/648/fadu-xdw223-pcie-gen5-nvme-u2-ssd" target="_blank" rel="noopener"
 &gt;OCP FADU XDW223&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;FDP places data more efficiently inside the SSD, reducing internal movement, garbage collection and write amplification. That can improve performance consistency, endurance and power consumption at the same time.&lt;/p&gt;
&lt;h3 id="43-commercial-gen5-revenue"&gt;4.3 Commercial Gen5 Revenue
&lt;/h3&gt;&lt;p&gt;Controller revenue reached approximately KRW 47.9bn in Q1 2026, or 80.4% of total revenue. This is the clearest evidence that FADU has moved from development into commercial supply. The Q2 preliminary filing does not provide product-level revenue, so the controller share must be checked in the half-year report.&lt;/p&gt;
&lt;h3 id="44-gen6-progress"&gt;4.4 Gen6 Progress
&lt;/h3&gt;&lt;p&gt;The FC6161 has completed tape-out. FADU targets 28.5GB/s sequential reads and writes and 6.9M random-read IOPS. The OCP contribution database also lists 2026 security-evaluation entries for the FC6161. &lt;a class="link" href="https://www.opencompute.org/contributions?contributions%5Bpage%5D=17" target="_blank" rel="noopener"
 &gt;OCP Contribution Database&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;This demonstrates design progress and participation in the security-evaluation ecosystem. It does not prove customer qualification, production orders or revenue.&lt;/p&gt;
&lt;h2 id="5-competitors-are-already-moving"&gt;5. Competitors Are Already Moving
&lt;/h2&gt;&lt;p&gt;FADU&amp;rsquo;s performance target is competitive, but Gen6 leadership has not been established.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Company and product&lt;/th&gt;
 &lt;th&gt;Public status&lt;/th&gt;
 &lt;th&gt;Implication&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron 9650&lt;/td&gt;
 &lt;td&gt;28GB/s, 5.5M random-read IOPS, high-volume production in 2026&lt;/td&gt;
 &lt;td&gt;Integrated NAND-controller supplier is already shipping Gen6&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung PM1763&lt;/td&gt;
 &lt;td&gt;Mass production announced in July 2026, ninth-generation V-NAND and 4nm controller&lt;/td&gt;
 &lt;td&gt;Gen6 optimized for AI and liquid-cooled infrastructure&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Silicon Motion SM8466&lt;/td&gt;
 &lt;td&gt;Gen5, TLC/QLC, FDP/ZNS, 128TB QLC reference design&lt;/td&gt;
 &lt;td&gt;Direct merchant-controller competition&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FADU FC6161&lt;/td&gt;
 &lt;td&gt;Tape-out, targets 28.5GB/s and 6.9M IOPS&lt;/td&gt;
 &lt;td&gt;Strong target, but customer qualification and revenue remain unconfirmed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Sources: &lt;a class="link" href="https://investors.micron.com/news-releases/news-release-details/micron-unveils-portfolio-industry-first-ssds-power-ai-revolution" target="_blank" rel="noopener"
 &gt;Micron AI SSD portfolio&lt;/a&gt;, &lt;a class="link" href="https://news.samsung.com/kr/%EC%82%BC%EC%84%B1%EC%A0%84%EC%9E%90-ai-%EC%9D%B8%ED%94%84%EB%9D%BC%EC%97%90-%EC%B5%9C%EC%A0%81%ED%99%94%EB%90%9C-%EA%B8%B0%EC%97%85%EC%9A%A9-ssd-pm1763-%EC%96%91%EC%82%B0" target="_blank" rel="noopener"
 &gt;Samsung PM1763&lt;/a&gt;, &lt;a class="link" href="https://www.siliconmotion.com/products/enterprise/detail" target="_blank" rel="noopener"
 &gt;Silicon Motion enterprise controller&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;FADU states that its controllers can support both TLC and QLC. The limitation is therefore not the absence of QLC support. The missing evidence is a public 64TB or 128TB QLC customer design win, qualification and production revenue. Silicon Motion already presents a 128TB QLC reference platform.&lt;/p&gt;
&lt;p&gt;Qualification is also a two-sided barrier. Enterprise SSD development and qualification can take more than 18 months, followed by three to five years of supply. That creates switching costs after entry. Meta&amp;rsquo;s open-source SSD qualification framework can reduce testing friction for FADU, but over time it may also reduce process advantages held by incumbents. &lt;a class="link" href="https://www.opencompute.org/blog/introducing-metas-open-source-testing-framework-revolutionizing-ssd-qualifications" target="_blank" rel="noopener"
 &gt;Meta SSD qualification framework&lt;/a&gt;&lt;/p&gt;
&lt;h2 id="6-customer-concentration-remains-a-discount"&gt;6. Customer Concentration Remains a Discount
&lt;/h2&gt;&lt;p&gt;The Q1 business report shows Customer 1 at 51.53% of revenue. A separate financial-note classification shows Customer A at 49.15% and Customer B at 31.87%. The classifications differ, but both show high concentration. &lt;a class="link" href="https://kind.krx.co.kr/external/2026/05/12/000812/20260512001819/11013.htm" target="_blank" rel="noopener"
 &gt;FADU Q1 2026 business report&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;Customer qualification can create recurring revenue once production begins, but it also makes quarterly results sensitive to a small number of order schedules. The new server OEM and KRW 340bn of orders are positive. Diversification is not proved until the largest customer falls from the 50% range toward 30% of reported revenue.&lt;/p&gt;
&lt;h2 id="7-moat-score-305-narrow-but-expandable"&gt;7. Moat Score: 3.0/5, Narrow but Expandable
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Area&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Score&lt;/th&gt;
 &lt;th&gt;Assessment&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Controller architecture and firmware&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4.0/5&lt;/td&gt;
 &lt;td&gt;Proprietary design and commercial Gen5 revenue&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Power efficiency and QoS&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3.5/5&lt;/td&gt;
 &lt;td&gt;Strong company and OCP evidence, limited independent comparison&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Custom development and qualification&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3.5/5&lt;/td&gt;
 &lt;td&gt;Long process creates barriers, concentration remains high&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;OCP and FDP ecosystem&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3.5/5&lt;/td&gt;
 &lt;td&gt;Public evidence for first FDP-enabled OCP SSD&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Customer and NAND partner scale&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2.5/5&lt;/td&gt;
 &lt;td&gt;Expanding, but still concentrated&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Gen6 mass-production proof&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2.5/5&lt;/td&gt;
 &lt;td&gt;Tape-out and security evaluation, no public production order&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Commercial QLC adoption&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2.0/5&lt;/td&gt;
 &lt;td&gt;Technical support claimed, no 64TB/128TB revenue evidence&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Overall&lt;/td&gt;
 &lt;td style="text-align: right"&gt;&lt;strong&gt;3.0/5&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;Narrow moat&lt;/strong&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;There is not enough evidence to call this a wide moat. There is also too much commercial and ecosystem evidence to dismiss FADU as a theme stock without technology. The company is expanding a real but narrow Gen5 position into the next generation.&lt;/p&gt;
&lt;h2 id="8-translating-the-preliminary-beat-into-eps"&gt;8. Translating the Preliminary Beat Into EPS
&lt;/h2&gt;&lt;p&gt;The preliminary filing does not disclose net income. The following is a mechanical bridge, not a revised broker consensus.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Step&lt;/th&gt;
 &lt;th&gt;Calculation&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Result&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Prior 2026 EPS&lt;/td&gt;
 &lt;td&gt;Naver WiseReport-based snapshot&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 1,817&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Operating-profit upside&lt;/td&gt;
 &lt;td&gt;KRW 16.27bn actual minus KRW 8.6bn consensus&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 7.67bn&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;After-tax conversion&lt;/td&gt;
 &lt;td&gt;Upside multiplied by 76%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 5.83bn&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Per-share increase&lt;/td&gt;
 &lt;td&gt;KRW 5.83bn divided by about 50.08m shares&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 116&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Mechanical revised EPS&lt;/td&gt;
 &lt;td&gt;KRW 1,817 plus KRW 116&lt;/td&gt;
 &lt;td style="text-align: right"&gt;&lt;strong&gt;About KRW 1,930&lt;/strong&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026 P/E&lt;/td&gt;
 &lt;td&gt;KRW 70,400 divided by KRW 1,930&lt;/td&gt;
 &lt;td style="text-align: right"&gt;&lt;strong&gt;36.5x&lt;/strong&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Adding the Q2 operating-profit upside to the prior 2026 operating-profit estimate of KRW 85.9bn, while leaving second-half forecasts unchanged, gives about KRW 93.6bn. Tax, currency, financial gains and tax credits can change the EPS conversion. The local snapshot also lacks a populated analyst-count field, so the estimate set should be treated as thin.&lt;/p&gt;
&lt;h2 id="9-a-scenario-range-not-a-single-fair-value"&gt;9. A Scenario Range, Not a Single Fair Value
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Scenario&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Earnings assumption&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Multiple&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Price range&lt;/th&gt;
 &lt;th&gt;Required conditions&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Bear&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2027 EPS KRW 3,400-3,800&lt;/td&gt;
 &lt;td style="text-align: right"&gt;16-18x&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 54,000-68,000&lt;/td&gt;
 &lt;td&gt;Slower growth, Gen6 delay, continued concentration&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Base&lt;/td&gt;
 &lt;td style="text-align: right"&gt;EPS KRW 4,000-4,300&lt;/td&gt;
 &lt;td style="text-align: right"&gt;18-20x&lt;/td&gt;
 &lt;td style="text-align: right"&gt;&lt;strong&gt;KRW 72,000-86,000&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;Gen5 growth, backlog conversion, Gen6 progress&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Bull&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2027-28 average EPS around KRW 4,600&lt;/td&gt;
 &lt;td style="text-align: right"&gt;24-28x&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 110,000-129,000&lt;/td&gt;
 &lt;td&gt;New hyperscaler, Gen6 production, QLC and diversification&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The July 22 close of KRW 70,400 is near the lower end of the base case. The stock remains expensive on mechanically revised 2026 EPS, but modestly discounted if 2027 EPS reaches KRW 4,000 or more. The KRW 130,000 market target is closer to a bull case than to a valuation supported by current earnings alone.&lt;/p&gt;
&lt;h2 id="10-why-the-stock-fell-after-strong-results"&gt;10. Why the Stock Fell After Strong Results
&lt;/h2&gt;&lt;p&gt;FADU traded as high as KRW 79,800 on the release date but closed at KRW 71,600. It rebounded to KRW 77,600 on July 21 and then fell to KRW 70,400 on July 22.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Period&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Foreigners&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Institutions&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Individuals&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Latest 3 sessions&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 20.79bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 9.67bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 8.20bn&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Latest 5 sessions&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 37.20bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 18.97bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 14.83bn&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Latest 10 sessions&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 70.28bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 49.28bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 15.24bn&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Investment trusts purchased about 494,000 shares over ten sessions. However, both foreigners and institutions sold on July 22, while program selling totaled KRW 17.88bn over the three post-earnings sessions.&lt;/p&gt;
&lt;p&gt;Short-sale value represented an average of roughly 4.9% of turnover over the latest five sessions and 3.0% on July 22. New short selling alone does not explain the 9.3% decline. Profit-taking and program flows matter more. Securities-lending balances rose from 1.08m shares on July 16 to 1.53m on July 21.&lt;/p&gt;
&lt;p&gt;Flow, program, short-sale and lending figures come from the Research OS local database and Kiwoom REST provider layer. They are useful directional data, but the short-sale and lending series have not been fully reconciled with official KRX source data. Short-interest balances are unavailable.&lt;/p&gt;
&lt;h2 id="11-execution-framework"&gt;11. Execution Framework
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Dimension&lt;/th&gt;
 &lt;th&gt;Assessment&lt;/th&gt;
 &lt;th&gt;Reason&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Fundamentals&lt;/td&gt;
 &lt;td&gt;Upgraded&lt;/td&gt;
 &lt;td&gt;Operating profit and margin exceeded the prior bull case&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Valuation&lt;/td&gt;
 &lt;td&gt;Neutral to modestly discounted&lt;/td&gt;
 &lt;td&gt;Expensive on 2026, near low end of 2027 base case&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Flows and chart&lt;/td&gt;
 &lt;td&gt;Weak&lt;/td&gt;
 &lt;td&gt;Foreign and program selling, higher lending balance&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Approach&lt;/td&gt;
 &lt;td&gt;Confirmation-based&lt;/td&gt;
 &lt;td&gt;Earnings are visible, flow reversal is not&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Three price zones can be used as confirmation levels rather than forecasts.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;KRW 70,000 support&lt;/strong&gt;: close to the July 22 intraday low of KRW 70,200. Stabilization and slower foreign selling would improve the setup.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;KRW 76,800-77,600 recovery&lt;/strong&gt;: regaining the post-earnings supply zone with lower foreign or program selling would provide stronger confirmation.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;KRW 63,900-65,000 risk zone&lt;/strong&gt;: a failure to recover after losing KRW 70,000 could open a retest.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;These levels are a public analytical framework, not personalized investment advice.&lt;/p&gt;
&lt;h2 id="12-four-pieces-of-evidence-that-would-prove-a-wider-moat"&gt;12. Four Pieces of Evidence That Would Prove a Wider Moat
&lt;/h2&gt;&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;FC6161 Gen6 customer qualification and production orders&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;A 64TB or 128TB QLC design win&lt;/strong&gt; that turns claimed support into customer revenue.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Largest-customer share falling toward 30%&lt;/strong&gt; as new OEMs and hyperscalers contribute.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Independent evidence&lt;/strong&gt; of p99.99 latency, WAF and performance-per-watt superiority under equal conditions.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;Two or more of these would materially improve confidence in the 2027 base case. A Gen6 delay beyond 2027, no QLC design win and persistent customer concentration would compress the multiple even if Gen5 remains profitable.&lt;/p&gt;
&lt;h2 id="13-red-team"&gt;13. Red Team
&lt;/h2&gt;&lt;p&gt;First, the 22.2% margin may not repeat. Product mix, revenue timing and lower one-time spending could have helped the quarter. The half-year report must confirm gross margin and controller mix.&lt;/p&gt;
&lt;p&gt;Second, integrated suppliers have scale advantages. Micron and Samsung can optimize NAND, controller, firmware and finished SSD together. Their position may strengthen as Gen6 and 122TB-245TB QLC products become mainstream.&lt;/p&gt;
&lt;p&gt;Third, standardization can lower qualification barriers. OCP and Meta&amp;rsquo;s open tools helped FADU enter the market, but they can also help future competitors.&lt;/p&gt;
&lt;p&gt;Fourth, orders are not cash. Purchase orders, delivery, acceptance, revenue recognition and collection occur at different times. Operating cash flow and receivables must keep pace with reported profit.&lt;/p&gt;
&lt;h2 id="14-conclusion"&gt;14. Conclusion
&lt;/h2&gt;&lt;p&gt;Q2 strongly demonstrated that FADU can earn money from its Gen5 controller business. The 22.2% operating margin is more important than KRW 73.16bn of revenue because it suggests a favorable controller mix and operating leverage.&lt;/p&gt;
&lt;p&gt;The quarter did not prove Gen6 production, QLC design wins or customer diversification. FADU&amp;rsquo;s current moat is 3.0 out of 5: narrow, commercially real and capable of expanding. The stock is not cheap on 2026 earnings alone. It enters the lower end of the base-case valuation only if 2027 EPS can exceed KRW 4,000.&lt;/p&gt;
&lt;p&gt;The most precise description is this:&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;FADU is a high-performance enterprise SSD controller supplier with proven Gen5 commercialization and an option to become a broader Gen6 and QLC platform. The Q2 surprise proved the profitability of the current business, not the success of that future option.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="evidence-map"&gt;Evidence Map
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Classification&lt;/th&gt;
 &lt;th&gt;Items&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Confirmed facts&lt;/td&gt;
 &lt;td&gt;Preliminary Q2 revenue and operating profit, Gen5 controller revenue, OCP FDP listing, FC6161 tape-out/security listing, Q1 customer concentration&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Analysis&lt;/td&gt;
 &lt;td&gt;Meaning of the margin beat, mechanical EPS bridge, 2027 scenario range, 3.0/5 moat score&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Unconfirmed&lt;/td&gt;
 &lt;td&gt;Q2 product gross margin, net income and cash flow; Gen6 qualification and orders; 64TB/128TB QLC adoption; independent performance comparison&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;&lt;em&gt;Disclaimer: For research and information purposes only. Not investment advice. Names cited are for analytical illustration; readers should perform their own due diligence and consult licensed advisors before any investment decision.&lt;/em&gt;&lt;/p&gt;</description></item><item><title>Big Tech AI Earnings: 2027 Demand Is Contracted, but 2028 Free Cash Flow Is the Real Test</title><link>https://koreainvestinsights.com/post/big-tech-ai-earnings-capex-roi-memory-2028-fcf-2026-07-22/</link><pubDate>Wed, 22 Jul 2026 17:30:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/big-tech-ai-earnings-capex-roi-memory-2028-fcf-2026-07-22/</guid><description>&lt;p&gt;Amazon, Microsoft, and Alphabet plan to spend roughly $570-580 billion on capital investment in 2026. Their latest reported cloud growth rates were 28%, 39% in constant currency, and 63%, respectively. The market is no longer debating whether current AI demand is strong. It is asking whether the data centers built now will be well utilized, whether revenue can outrun depreciation and power costs, and whether memory profits can survive the supply response expected from 2028.&lt;/p&gt;
&lt;p&gt;The last four quarters suggest that much of the demand through 2027 is already supported by contracts, reservations, and supply constraints. Beyond 2028, scarcity is no longer enough. Utilization, inference revenue, customer concentration, and cash collection become the decisive variables.&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Related context
This is a follow-up to the &lt;a class="link" href="https://koreainvestinsights.com/en/post/big-tech-july-earnings-call-memory-thesis-scenarios-2026-07-07/" &gt;July Big Tech earnings and memory thesis preview&lt;/a&gt;. The financing side is covered in &lt;a class="link" href="https://koreainvestinsights.com/en/post/hyperscaler-financing-race-ai-capex-memory-bottleneck-2026-07-07/" &gt;Hyperscaler financing and the memory bottleneck&lt;/a&gt;, while the semiconductor supply response is developed in the &lt;a class="link" href="https://koreainvestinsights.com/en/post/weekly-semiconductor-deep-dive-strong-earnings-lower-multiple-2028-pivot-2026-07-19/" &gt;weekly semiconductor deep dive&lt;/a&gt;.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="summary"&gt;Summary
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;AI demand is likely to remain stronger than expected through 2027. Backlogs, long-term commitments, supply shortages, and customer expansion point in the same direction.&lt;/li&gt;
&lt;li&gt;The most durable demand is likely to come from enterprise inference, agents, coding, search, and data services rather than one-off frontier training runs.&lt;/li&gt;
&lt;li&gt;The three companies plan approximately $570-580 billion of 2026 capital spending. Spending can grow again in 2027, although the growth rate should slow from an unusually high base.&lt;/li&gt;
&lt;li&gt;Return on investment should be measured through incremental cloud gross profit and first-party revenue or cost savings, less depreciation, power, and financing costs.&lt;/li&gt;
&lt;li&gt;Current evidence supports the highest probability of post-2028 free cash flow recovery at Alphabet, followed by Microsoft and Amazon. Alphabet shows the clearest combination of monetization and unit-cost improvement. Microsoft has the strongest contracted visibility but more concentration and lease complexity. Amazon has the largest rebound option but the greatest current cash-flow pressure.&lt;/li&gt;
&lt;li&gt;Memory purchases should continue across HBM, server DRAM, and enterprise SSDs. Custom accelerators change the product mix but do not remove the need for memory.&lt;/li&gt;
&lt;li&gt;The principal risk after 2028 is price rather than bit demand. New capacity can reduce average selling prices and margins even if memory shipments keep rising.&lt;/li&gt;
&lt;li&gt;Credit conditions are cautionary, not distressed. Bond books remain covered, but lower cover ratios and wider long-end spreads show that capital markets are beginning to impose discipline.&lt;/li&gt;
&lt;/ul&gt;
&lt;hr&gt;
&lt;div class="thesis-callout"&gt;
 &lt;div class="thesis-callout__label"&gt;Core judgment&lt;/div&gt;
 &lt;div class="thesis-callout__body"&gt;
 Through 2027, supply constraints and contracted demand support continued spending. From 2028, the decisive variables shift to utilization, cash collection, and returns after depreciation. For memory, strong volume and normalizing prices can coexist.
 &lt;/div&gt;
&lt;/div&gt;
&lt;hr&gt;
&lt;h2 id="1-the-three-earnings-calls-and-the-questions-that-matter"&gt;1. The three earnings calls and the questions that matter
&lt;/h2&gt;&lt;p&gt;Alphabet reports at 5:30 a.m. KST on July 23, Microsoft at 6:30 a.m. KST on July 30, and Amazon at 6:00 a.m. KST on July 31, based on company investor-relations announcements.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Company&lt;/th&gt;
 &lt;th&gt;Earnings time&lt;/th&gt;
 &lt;th&gt;Primary question&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Alphabet&lt;/td&gt;
 &lt;td&gt;July 23, 05:30 KST&lt;/td&gt;
 &lt;td&gt;How quickly do 63% cloud growth and a $462.3 billion backlog convert into revenue and cash?&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Microsoft&lt;/td&gt;
 &lt;td&gt;July 30, 06:30 KST&lt;/td&gt;
 &lt;td&gt;When does Azure supply improve, and does backlog growth reaccelerate outside OpenAI?&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Amazon&lt;/td&gt;
 &lt;td&gt;July 31, 06:00 KST&lt;/td&gt;
 &lt;td&gt;Can AWS protect margins and restore free cash flow while Amazon spends about $200 billion?&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The market is trying to answer three questions. How long can AI demand grow above expectations, and from which customers and workloads? How much longer can capital spending rise, and can the investments earn adequate returns? Will that spending translate into recurring purchases of HBM, server DRAM, and enterprise SSDs?&lt;/p&gt;
&lt;p&gt;The last four quarters provide a constructive answer to the first question. The second and third require company-specific analysis because a dollar spent on accelerators has a different return profile from a dollar spent on land, power, networking, or long-lived buildings.&lt;/p&gt;
&lt;h2 id="2-four-quarters-of-operating-evidence"&gt;2. Four quarters of operating evidence
&lt;/h2&gt;&lt;h3 id="21-amazon-aws-accelerated-while-cash-left-first"&gt;2.1 Amazon: AWS accelerated while cash left first
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Quarter&lt;/th&gt;
 &lt;th style="text-align: right"&gt;AWS revenue&lt;/th&gt;
 &lt;th style="text-align: right"&gt;YoY growth&lt;/th&gt;
 &lt;th style="text-align: right"&gt;AWS operating income&lt;/th&gt;
 &lt;th&gt;Cash-flow signal&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Q2 2025&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$30.9B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+17.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$10.2B&lt;/td&gt;
 &lt;td&gt;TTM FCF $18.2B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q3 2025&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$33.0B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+20%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$11.4B&lt;/td&gt;
 &lt;td&gt;AI infrastructure spending kept rising&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q4 2025&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$35.6B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+24%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$12.4B&lt;/td&gt;
 &lt;td&gt;Amazon signaled about $200B of 2026 investment&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q1 2026&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$37.6B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+28%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$14.2B&lt;/td&gt;
 &lt;td&gt;TTM FCF fell to $1.2B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;AWS growth accelerated for four consecutive quarters. Its Q1 operating margin reached 37.8%, showing that revenue and profit expanded despite supply constraints.&lt;/p&gt;
&lt;p&gt;The cash profile moved in the opposite direction. Amazon generated $148.5 billion of trailing operating cash flow, but net property and equipment spending reached $147.3 billion. AWS-centered long-term commitments stood at $364 billion with a weighted-average remaining life of 5.5 years. Revenue visibility is high, but spending occurs before the associated revenue arrives.&lt;/p&gt;
&lt;p&gt;Amazon&amp;rsquo;s 2025 shareholder letter stated that the company was not spending $200 billion on a hunch and had substantial customer commitments. Graviton, Trainium, and Nitro together exceeded a $20 billion annual revenue run rate and were growing at a triple-digit rate. Custom silicon is both a cost-control tool and a way to keep workloads inside AWS.&lt;/p&gt;
&lt;p&gt;Amazon can produce the largest cash-flow rebound if the new assets fill quickly, but it also has the widest gap to close. AWS needs to maintain growth around the high-20% range and defend margins in the mid-to-high 30s while 2026 investments begin monetizing in 2027 and 2028.&lt;/p&gt;
&lt;p&gt;Sources: &lt;a class="link" href="https://ir.aboutamazon.com/news-release/news-release-details/2026/Amazon-com-Announces-First-Quarter-Results/default.aspx" target="_blank" rel="noopener"
 &gt;Amazon Q1 2026 results&lt;/a&gt;, &lt;a class="link" href="https://www.aboutamazon.com/news/company-news/amazon-ceo-andy-jassy-2025-letter-to-shareholders" target="_blank" rel="noopener"
 &gt;Amazon 2025 shareholder letter&lt;/a&gt;, &lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/1018724/000101872426000014/amzn-20260331.htm" target="_blank" rel="noopener"
 &gt;Amazon Q1 2026 10-Q&lt;/a&gt;&lt;/p&gt;
&lt;h3 id="22-microsoft-the-strongest-contract-visibility-and-the-largest-concentration-question"&gt;2.2 Microsoft: the strongest contract visibility and the largest concentration question
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Quarter&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Revenue&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Azure growth, constant currency&lt;/th&gt;
 &lt;th style="text-align: right"&gt;RPO&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Capital spending&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Free cash flow&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;FY25 Q4&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$76.4B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about +39%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$368B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$24.2B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$25.6B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY26 Q1&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$77.7B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+39%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$392B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$34.9B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$25.7B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY26 Q2&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$81.3B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+38%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$625B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$37.5B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$5.9B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY26 Q3&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$82.9B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+39%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$627B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$31.9B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$15.8B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Azure grew 40% as reported and 39% in constant currency in the latest quarter. Microsoft&amp;rsquo;s AI business exceeded a $37 billion annual revenue run rate and grew 123%. Commercial RPO reached $627 billion with an average duration of 2.5 years, and roughly 25% is expected to convert to revenue over the next 12 months.&lt;/p&gt;
&lt;p&gt;Contract quality needs to be separated from contract size. OpenAI accounted for approximately 45% of commercial RPO in FY26 Q2, while RPO excluding OpenAI grew 26% in FY26 Q3. A major strategic customer and a broader enterprise expansion are both present.&lt;/p&gt;
&lt;p&gt;Microsoft also has the most complex capital-spending definition. Q3 cash property and equipment spending was $30.9 billion, while spending including finance leases reached $31.9 billion. Management said roughly two-thirds of capital spending went to shorter-lived assets such as GPUs and CPUs, with one-third allocated to land and buildings. Short-lived assets monetize earlier but also return to the replacement cycle sooner.&lt;/p&gt;
&lt;p&gt;Azure is expected to remain supply constrained through 2026, with modest growth acceleration possible in the second half of the calendar year. Q4 capital spending is expected to exceed $40 billion. Calendar 2026 investment is approximately $190 billion, including about $25 billion of component inflation. Microsoft expects double-digit revenue and operating-income growth in FY27.&lt;/p&gt;
&lt;p&gt;The company has the strongest contracted visibility, but post-2028 free cash flow depends on independent enterprise demand, Copilot and agent usage, lease-adjusted cash generation, and whether gross profit grows faster than depreciation.&lt;/p&gt;
&lt;p&gt;Sources: &lt;a class="link" href="https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q3" target="_blank" rel="noopener"
 &gt;Microsoft FY26 Q3 earnings call&lt;/a&gt;, &lt;a class="link" href="https://www.microsoft.com/en-us/investor/earnings/FY-2026-Q3/press-release-webcast" target="_blank" rel="noopener"
 &gt;Microsoft FY26 Q3 results&lt;/a&gt;&lt;/p&gt;
&lt;h3 id="23-alphabet-demand-margin-and-unit-cost-improvement-are-visible-together"&gt;2.3 Alphabet: demand, margin, and unit-cost improvement are visible together
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Quarter&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Google Cloud revenue&lt;/th&gt;
 &lt;th style="text-align: right"&gt;YoY growth&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Operating income&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Margin&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Backlog&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Capital spending&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Q2 2025&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$13.6B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+32%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$2.8B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;20.7%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$106B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$22.4B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q3 2025&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$15.2B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+34%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$3.6B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;23.7%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$155B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$24.0B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q4 2025&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$17.7B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+48%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$5.3B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;30.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$240B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$27.9B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q1 2026&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$20.0B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+63%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$6.6B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;32.9%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$462.3B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$35.7B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Alphabet delivered the clearest simultaneous improvement. Cloud growth accelerated from 32% to 63%, the operating margin rose from 20.7% to 32.9%, and backlog increased from $106 billion to $462.3 billion.&lt;/p&gt;
&lt;p&gt;Q1 generative AI product revenue grew 800%. New customers doubled, deals between $100 million and $1 billion doubled, and Alphabet signed multiple deals above $1 billion. Existing customers consumed 45% more than their original commitments. Gemini Enterprise paid monthly users rose 40% quarter over quarter.&lt;/p&gt;
&lt;p&gt;Unit economics also improved. Alphabet said the cost per AI response fell by more than 30%. Higher demand and lower unit cost are the best combination for infrastructure returns. The same infrastructure supports Search, YouTube, Workspace, and advertising, so external cloud revenue understates part of the return.&lt;/p&gt;
&lt;p&gt;Alphabet guides to $180-190 billion of 2026 capital spending and expects another significant increase in 2027. Q1 uncommenced asset commitments were approximately $108.6 billion, and uncommenced data-center leases were $75.6 billion, scheduled to begin between 2026 and 2031. Total RPO was $467.6 billion, with just over half expected to be recognized within 24 months.&lt;/p&gt;
&lt;p&gt;The main risk is the speed of spending itself. Q1 operating cash flow was $45.8 billion and capital spending was $35.7 billion, leaving quarterly free cash flow of $10.1 billion. Trailing free cash flow remained a substantial $64.4 billion, but another large increase in 2027 spending could delay the recovery.&lt;/p&gt;
&lt;p&gt;Current evidence nevertheless places Alphabet first in proving AI investment returns. The next call needs to clarify how much backlog growth came from a few large contracts and how quickly new facilities are filling.&lt;/p&gt;
&lt;p&gt;Sources: &lt;a class="link" href="https://abc.xyz/investor/events/event-details/2026/2026-Q1-Earnings-Call-2026-nW8kCrBAKS/default.aspx" target="_blank" rel="noopener"
 &gt;Alphabet Q1 2026 earnings call&lt;/a&gt;, &lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/1652044/000165204426000048/goog-20260331.htm" target="_blank" rel="noopener"
 &gt;Alphabet Q1 2026 10-Q&lt;/a&gt;&lt;/p&gt;
&lt;h2 id="3-where-ai-demand-can-persist"&gt;3. Where AI demand can persist
&lt;/h2&gt;&lt;p&gt;AI demand should be divided into five layers with different customers and durations.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Demand layer&lt;/th&gt;
 &lt;th&gt;Customers&lt;/th&gt;
 &lt;th&gt;Expected durability&lt;/th&gt;
 &lt;th&gt;Key evidence&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Frontier training and post-training&lt;/td&gt;
 &lt;td&gt;Leading model labs and first-party models&lt;/td&gt;
 &lt;td&gt;Strong but concentrated&lt;/td&gt;
 &lt;td&gt;Cluster reservations, long-term power contracts, model roadmaps&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Enterprise inference and agents&lt;/td&gt;
 &lt;td&gt;Finance, retail, manufacturing, healthcare, government&lt;/td&gt;
 &lt;td&gt;Most durable candidate&lt;/td&gt;
 &lt;td&gt;Paid users, calls, spend per customer, renewals&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Development and coding&lt;/td&gt;
 &lt;td&gt;Software companies and developers&lt;/td&gt;
 &lt;td&gt;Rapid adoption, strong price competition&lt;/td&gt;
 &lt;td&gt;Seats, agent runtime, generated code, retention&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Data, storage, and general cloud&lt;/td&gt;
 &lt;td&gt;Existing cloud customers&lt;/td&gt;
 &lt;td&gt;Stable and bundled with AI&lt;/td&gt;
 &lt;td&gt;Database, storage, networking, migrations&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;First-party services&lt;/td&gt;
 &lt;td&gt;Search, advertising, productivity, commerce, recommendations&lt;/td&gt;
 &lt;td&gt;Revenue and cost savings combined&lt;/td&gt;
 &lt;td&gt;Conversion, cost per response, productivity, gross margin&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h3 id="31-supply-should-remain-the-constraint-through-2027"&gt;3.1 Supply should remain the constraint through 2027
&lt;/h3&gt;&lt;p&gt;AWS, Azure, and Google Cloud have all described supply constraints. Alphabet customers used 45% more than their initial commitments. Microsoft expects constraints through the end of 2026. Amazon&amp;rsquo;s long-term commitments have an average 5.5 years remaining.&lt;/p&gt;
&lt;p&gt;This supports a capacity-led revenue model through at least 2027. As data centers open and power is connected, revenue should follow. Near-term spending reductions are also difficult because construction, leases, power, servers, and memory are already under contract.&lt;/p&gt;
&lt;h3 id="32-2028-is-more-likely-to-be-a-change-in-demand-quality-than-a-demand-cliff"&gt;3.2 2028 is more likely to be a change in demand quality than a demand cliff
&lt;/h3&gt;&lt;p&gt;More facilities and semiconductor supply arrive from 2028. Once scarcity eases, investors must test whether customers actually use reserved capacity, renew contracts, and increase usage fast enough to offset lower inference prices.&lt;/p&gt;
&lt;p&gt;Demand can remain high while profitability falls. Model efficiency and custom silicon reduce cost per task, while cloud competition can lower revenue per token. If call volume grows faster than pricing declines, utilization and revenue still rise.&lt;/p&gt;
&lt;p&gt;The useful relationship is:&lt;/p&gt;
&lt;p&gt;&lt;code&gt;AI revenue growth = workload growth + customer growth - price decline per workload&lt;/code&gt;&lt;/p&gt;
&lt;p&gt;Enterprise agents, coding, and embedded inference have the best chance of producing high-frequency, recurring use.&lt;/p&gt;
&lt;h3 id="33-independent-customers-are-the-most-important-proof"&gt;3.3 Independent customers are the most important proof
&lt;/h3&gt;&lt;p&gt;Strategic customer contracts support early demand, but reciprocal investment and revenue commitments can overstate final demand. Confidence after 2028 requires more usage from enterprises, governments, developers, and smaller customers without financing links to the infrastructure provider.&lt;/p&gt;
&lt;p&gt;Alphabet&amp;rsquo;s doubled new-customer count, doubled large-deal count, and commitment overages are the clearest current evidence. Microsoft&amp;rsquo;s 26% RPO growth excluding OpenAI is more important than total RPO alone. Amazon needs to disclose more about how much of AWS commitments comes from AI versus general cloud demand.&lt;/p&gt;
&lt;h2 id="4-how-much-longer-can-capital-spending-grow"&gt;4. How much longer can capital spending grow?
&lt;/h2&gt;&lt;h3 id="41-approximately-570-580-billion-in-2026"&gt;4.1 Approximately $570-580 billion in 2026
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Company&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2026 investment plan&lt;/th&gt;
 &lt;th&gt;Definition caveat&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Amazon&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about $200B&lt;/td&gt;
 &lt;td&gt;Primarily AI and AWS, with some spending elsewhere&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Microsoft&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about $190B&lt;/td&gt;
 &lt;td&gt;Cash spending and finance leases can both be included&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Alphabet&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$180-190B&lt;/td&gt;
 &lt;td&gt;Servers, data centers, networks, and custom silicon&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Total&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about $570-580B&lt;/td&gt;
 &lt;td&gt;Accounting definitions differ across companies&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The absolute amount can grow again, but the 60-100% growth rates seen in 2026 are difficult to repeat. The base is larger, while power, sites, transformers, networking equipment, and skilled labor remain constrained.&lt;/p&gt;
&lt;p&gt;The following 2027 ranges are analytical scenarios, not company guidance.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Scenario&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Combined 2027 spending&lt;/th&gt;
 &lt;th&gt;Conditions&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Downside&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$530-580B&lt;/td&gt;
 &lt;td&gt;Slower bookings, power delays, digestion of 2026 prebuys&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Base&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$630-665B&lt;/td&gt;
 &lt;td&gt;Continued constraints, slower growth rate, execution of existing commitments&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Upside&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Above $700B&lt;/td&gt;
 &lt;td&gt;Agent usage surge, additional sovereign demand, custom silicon and GPU expansion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;A slower capital-spending growth rate is not by itself a bearish signal. The warning comes when slower spending, cloud growth, and backlog appear together.&lt;/p&gt;
&lt;h3 id="42-a-simple-return-framework"&gt;4.2 A simple return framework
&lt;/h3&gt;&lt;p&gt;&lt;code&gt;AI investment return = incremental cloud gross profit + first-party revenue and savings - depreciation - power - financing cost&lt;/code&gt;&lt;/p&gt;
&lt;p&gt;Capital spending hits the cash-flow statement first. Depreciation enters the income statement after servers and data centers begin operating. Cash-flow troughs and accounting-margin troughs therefore occur at different times.&lt;/p&gt;
&lt;p&gt;Free cash flow is pressured first in 2026. Revenue and depreciation rise together in 2027-2029. Around 2030, accelerators purchased in 2026 can begin returning to the replacement cycle. Cash flow can recover before reported margins fully absorb depreciation.&lt;/p&gt;
&lt;h3 id="43-current-evidence-by-company"&gt;4.3 Current evidence by company
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Company&lt;/th&gt;
 &lt;th&gt;Positive evidence&lt;/th&gt;
 &lt;th&gt;Remaining risk&lt;/th&gt;
 &lt;th&gt;Confidence in post-2028 FCF recovery&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Alphabet&lt;/td&gt;
 &lt;td&gt;Cloud +63%, margin 32.9%, cost per response down more than 30%&lt;/td&gt;
 &lt;td&gt;Fast spending growth, leases, possible large-deal concentration&lt;/td&gt;
 &lt;td&gt;Highest&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Microsoft&lt;/td&gt;
 &lt;td&gt;Azure +39%, RPO $627B, AI ARR above $37B&lt;/td&gt;
 &lt;td&gt;OpenAI concentration, finance leases, short-lived assets&lt;/td&gt;
 &lt;td&gt;Medium-high&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Amazon&lt;/td&gt;
 &lt;td&gt;AWS +28%, margin 37.8%, commitments $364B&lt;/td&gt;
 &lt;td&gt;TTM FCF $1.2B, $200B spending, monetization lag&lt;/td&gt;
 &lt;td&gt;Medium&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="5-the-dashboard-for-2028-and-beyond"&gt;5. The dashboard for 2028 and beyond
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Metric&lt;/th&gt;
 &lt;th&gt;Constructive signal&lt;/th&gt;
 &lt;th&gt;Warning signal&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Independent-customer RPO&lt;/td&gt;
 &lt;td&gt;Growth outside strategically financed customers&lt;/td&gt;
 &lt;td&gt;A few customers drive most of the increase&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Next-12-month recognition&lt;/td&gt;
 &lt;td&gt;A larger share of RPO converts soon&lt;/td&gt;
 &lt;td&gt;Contract duration rises while near-term conversion slows&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cash collection&lt;/td&gt;
 &lt;td&gt;Cash receipts outgrow contract assets&lt;/td&gt;
 &lt;td&gt;Receivables and contract assets grow without deposits&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Operating cash flow&lt;/td&gt;
 &lt;td&gt;Four-quarter OCF growth exceeds capex growth by at least 10 points&lt;/td&gt;
 &lt;td&gt;OCF trails investment growth&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cloud economics&lt;/td&gt;
 &lt;td&gt;Gross profit growth exceeds depreciation and power growth&lt;/td&gt;
 &lt;td&gt;Revenue grows while margin falls&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Utilization&lt;/td&gt;
 &lt;td&gt;New capacity fills rapidly&lt;/td&gt;
 &lt;td&gt;Low utilization after power connection&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Lease-adjusted FCF&lt;/td&gt;
 &lt;td&gt;Recovery remains after finance leases&lt;/td&gt;
 &lt;td&gt;Reported FCF improves while lease obligations rise&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Custom silicon&lt;/td&gt;
 &lt;td&gt;Lower cost and higher customer use&lt;/td&gt;
 &lt;td&gt;R&amp;amp;D rises without adoption&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Memory contracts&lt;/td&gt;
 &lt;td&gt;Commitments convert to shipments and cash&lt;/td&gt;
 &lt;td&gt;Deferrals, renegotiation, or cancellations&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Supply normalization&lt;/td&gt;
 &lt;td&gt;Shipments and utilization hold despite lower prices&lt;/td&gt;
 &lt;td&gt;ASP and utilization fall together&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Replacement cycle&lt;/td&gt;
 &lt;td&gt;First-generation assets earn returns before replacement&lt;/td&gt;
 &lt;td&gt;Replacement begins before initial payback&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The strongest combination is independent RPO growth, new capacity going online, stable cloud margins, and OCF growth overtaking capital-spending growth. The most dangerous combination is easing supply constraints, slower growth, higher depreciation, rising customer concentration, and increasing debt or leases.&lt;/p&gt;
&lt;h2 id="6-credit-markets-are-imposing-discipline-not-signaling-a-crisis"&gt;6. Credit markets are imposing discipline, not signaling a crisis
&lt;/h2&gt;&lt;p&gt;Apollo reports that hyperscaler bond cover ratios fell from around 5 times in February 2026 to below 2 times in July. Demand still exceeds supply, but investors have become more selective.&lt;/p&gt;
&lt;p&gt;Amazon issued $25 billion of bonds on July 7. Orders peaked near $62 billion and ended around $41 billion after pricing. Its longest maturities offered an additional 18-21 basis points of yield. Funding remained available, but at a higher cost.&lt;/p&gt;
&lt;p&gt;The Bank of England&amp;rsquo;s July 2026 Financial Stability Report said AI-related issuance had not yet impaired broader investment-grade terms or availability. Barclays projected approximately $240 billion of 2026 investment-grade hyperscaler financing, comparable to major bank issuance.&lt;/p&gt;
&lt;p&gt;The signal is yellow, not red. Capital markets remain open but increasingly demand backlog quality, utilization, and cash conversion. This should delay weaker projects and widen capital-cost differences rather than end AI spending outright.&lt;/p&gt;
&lt;p&gt;Sources: &lt;a class="link" href="https://www.apollo.com/wealth/insights-news/insights/daily-spark/cover-ratios-for-hyperscaler-bonds-declining" target="_blank" rel="noopener"
 &gt;Apollo hyperscaler bond cover ratios&lt;/a&gt;, &lt;a class="link" href="https://www.bankofengland.co.uk/financial-stability-report/2026/july-2026" target="_blank" rel="noopener"
 &gt;Bank of England July 2026 Financial Stability Report&lt;/a&gt;, &lt;a class="link" href="https://fortune.com/2026/07/08/amazons-25-billion-surprise-bond-sale-dangled-extra-yield-to-lure-in-buyers/" target="_blank" rel="noopener"
 &gt;Amazon bond demand&lt;/a&gt;&lt;/p&gt;
&lt;h2 id="7-does-the-spending-become-recurring-memory-demand"&gt;7. Does the spending become recurring memory demand?
&lt;/h2&gt;&lt;h3 id="71-ai-systems-require-a-memory-hierarchy"&gt;7.1 AI systems require a memory hierarchy
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Memory layer&lt;/th&gt;
 &lt;th&gt;Function&lt;/th&gt;
 &lt;th&gt;Demand direction&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM&lt;/td&gt;
 &lt;td&gt;Bandwidth close to GPUs and ASICs&lt;/td&gt;
 &lt;td&gt;Strongest in HBM3E and HBM4&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Server DRAM&lt;/td&gt;
 &lt;td&gt;CPU work, agent orchestration, preprocessing, caches, metadata&lt;/td&gt;
 &lt;td&gt;Expands with agents and multi-model systems&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Enterprise SSD and NAND&lt;/td&gt;
 &lt;td&gt;Checkpoints, vector databases, documents, video, long-context storage&lt;/td&gt;
 &lt;td&gt;Expands with inference volume and retention&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Training is HBM intensive. Enterprise inference and agents expand server DRAM and SSD demand alongside HBM. Multiple agents repeat tasks and preserve longer context, increasing memory outside the accelerator.&lt;/p&gt;
&lt;p&gt;TPUs, Trainium, and Maia can displace some external GPUs but still require HBM, server DRAM, and storage. Custom silicon changes specifications, packaging, and bargaining power rather than eliminating memory demand.&lt;/p&gt;
&lt;h3 id="72-procurement-is-indirect-and-arrives-with-a-lag"&gt;7.2 Procurement is indirect and arrives with a lag
&lt;/h3&gt;&lt;p&gt;Hyperscalers do not purchase all HBM directly. Procurement runs through Nvidia, custom ASIC partners, server ODMs, and packaging suppliers. Capital spending therefore reaches memory-vendor revenue with a lag through design, allocation, wafer starts, packaging, server assembly, and installation.&lt;/p&gt;
&lt;p&gt;Micron&amp;rsquo;s strategic customer agreements illustrate the connection. It has 16 agreements, generally covering 2026-2030, representing roughly 20% of DRAM volume and one-third of NAND volume. Fourteen agreements contain approximately $100 billion of cumulative minimum-price revenue, supported by about $22 billion of customer deposits and financing commitments.&lt;/p&gt;
&lt;p&gt;Long-term agreements raise the earnings floor. Take-or-pay and minimum prices reduce cancellation and downside pricing. Ceiling prices can also limit upside when spot prices surge. These contracts reduce the severity of a downturn but do not permanently lock in today&amp;rsquo;s scarcity margins.&lt;/p&gt;
&lt;p&gt;Source: &lt;a class="link" href="https://investors.micron.com/static-files/631b1a32-5537-46ae-8f40-82e42fc79dfe" target="_blank" rel="noopener"
 &gt;Micron FY3Q26 prepared remarks&lt;/a&gt;&lt;/p&gt;
&lt;h3 id="73-the-2028-memory-risk-is-price-not-necessarily-demand"&gt;7.3 The 2028 memory risk is price, not necessarily demand
&lt;/h3&gt;&lt;p&gt;Additional DRAM, NAND, and HBM packaging capacity should arrive from 2027 and 2028. Bit shipments can keep growing while average selling prices and margins normalize.&lt;/p&gt;
&lt;p&gt;&lt;code&gt;Memory profit = bit shipments × price per bit - expansion, depreciation, and input costs&lt;/code&gt;&lt;/p&gt;
&lt;p&gt;In 2026 and 2027, demand is growing faster than supply, supporting both volume and price. From 2028, supply growth can reduce pricing even if bit demand stays high. Micron&amp;rsquo;s current gross margin above 80% reflects scarcity and should not be treated as a steady-state margin.&lt;/p&gt;
&lt;p&gt;The semiconductor conclusion therefore requires more than a direction for Big Tech capital spending. Utilization, memory supply growth, contract pricing, inventory, and producer capital discipline must be read together.&lt;/p&gt;
&lt;h2 id="8-company-specific-earnings-checklist"&gt;8. Company-specific earnings checklist
&lt;/h2&gt;&lt;h3 id="alphabet"&gt;Alphabet
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Cloud growth near 60%, with both new customers and usage increasing.&lt;/li&gt;
&lt;li&gt;Backlog rising from $462.3 billion without a deterioration in near-term recognition.&lt;/li&gt;
&lt;li&gt;Continued explanation for $180-190 billion of 2026 spending and the 2027 step-up.&lt;/li&gt;
&lt;li&gt;Cloud margin staying above 30% and cost per response falling further.&lt;/li&gt;
&lt;li&gt;Evidence that external TPU sales and internal usage both scale.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="microsoft"&gt;Microsoft
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Azure growth near 39-40% after additional supply comes online.&lt;/li&gt;
&lt;li&gt;RPO growth reaccelerating outside OpenAI.&lt;/li&gt;
&lt;li&gt;More than $40 billion of quarterly spending converting into revenue quickly.&lt;/li&gt;
&lt;li&gt;Paid Copilot and agent seats accompanied by actual usage and revenue.&lt;/li&gt;
&lt;li&gt;Operating cash flow growing faster than capital spending and finance leases.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="amazon"&gt;Amazon
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;AWS growth in the high 20s and margins in the mid-to-high 30s.&lt;/li&gt;
&lt;li&gt;Faster conversion of the $364 billion commitment base.&lt;/li&gt;
&lt;li&gt;Trainium adoption and revenue demonstrating a cost advantage over external GPUs.&lt;/li&gt;
&lt;li&gt;A clear 2027-2028 monetization schedule for roughly $200 billion of investment.&lt;/li&gt;
&lt;li&gt;Operating cash flow rising while net investment growth slows.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="9-three-paths-for-2028-2031"&gt;9. Three paths for 2028-2031
&lt;/h2&gt;&lt;p&gt;These probabilities are analytical judgments, not company guidance.&lt;/p&gt;
&lt;h3 id="base-case-55"&gt;Base case, 55%
&lt;/h3&gt;&lt;p&gt;Capital spending rises again in 2027 but at a slower rate. New data centers fill, and enterprise inference and agent revenue expand. Alphabet recovers free cash flow first, followed by Microsoft. Amazon recovers later because of a longer monetization lag.&lt;/p&gt;
&lt;p&gt;Semiconductor revenue remains high, but new supply normalizes memory prices and margins from 2028. HBM and server DRAM volumes remain strong while explosive spot-price increases fade.&lt;/p&gt;
&lt;h3 id="upside-case-20"&gt;Upside case, 20%
&lt;/h3&gt;&lt;p&gt;Agent, coding, search, and enterprise inference usage grows faster than unit-price declines. Sovereign and enterprise customers add demand. Combined 2027 capital spending exceeds $700 billion while operating cash flow also expands.&lt;/p&gt;
&lt;p&gt;HBM4, advanced packaging, server DRAM, and enterprise SSD shortages persist into 2028. Memory contract volumes and prices rise, producing another round of semiconductor earnings upgrades.&lt;/p&gt;
&lt;h3 id="downside-case-25"&gt;Downside case, 25%
&lt;/h3&gt;&lt;p&gt;Model efficiency and price competition reduce revenue per task quickly. Strategic contracts are renegotiated, and new data centers operate below expected utilization. Depreciation and power costs are already committed, pressuring margins and free cash flow.&lt;/p&gt;
&lt;p&gt;Memory supply added in 2027-2029 exceeds demand growth, pushing down ASP and utilization together. Long-term agreements soften the decline but do not protect uncontracted volume and spot pricing.&lt;/p&gt;
&lt;h2 id="10-conclusion"&gt;10. Conclusion
&lt;/h2&gt;&lt;p&gt;AI demand is likely to remain stronger than expected through 2027. Cloud growth, backlogs, supply constraints, and long-term infrastructure commitments support that view. The customer base is broadening from frontier labs to enterprise inference, coding, search, and general cloud workloads.&lt;/p&gt;
&lt;p&gt;Capital spending should remain elevated in 2027, although its growth rate can slow. A slowdown from a very high base is not a cycle break. The real warning would be simultaneous weakness in backlog, utilization, cloud margins, and operating cash flow.&lt;/p&gt;
&lt;p&gt;Current evidence gives Alphabet the best chance of post-2028 free cash flow recovery. It has shown demand growth, margin improvement, and lower unit costs together. Microsoft has stronger contract visibility but more OpenAI concentration and finance-lease complexity. Amazon has powerful AWS economics and significant rebound potential, but it must close the gap between $200 billion of investment and trailing free cash flow of only $1.2 billion.&lt;/p&gt;
&lt;p&gt;Memory purchases should continue, but the decisive post-2028 issue is pricing after supply expands. Custom silicon and inference efficiency change the mix across HBM, server DRAM, and enterprise SSDs rather than eliminating demand. Strategic customer agreements raise the floor but do not make current scarcity margins permanent.&lt;/p&gt;
&lt;p&gt;The most important figures in these earnings calls are not capital spending totals alone. Independent-customer backlog, next-12-month conversion, utilization, cloud margins, and lease-adjusted free cash flow must move together. If they improve, 2028 becomes the start of cash harvesting. If customer concentration rises while utilization and cash collection lag, today&amp;rsquo;s earnings strength can return as tomorrow&amp;rsquo;s depreciation burden.&lt;/p&gt;
&lt;hr&gt;

 &lt;blockquote&gt;
 &lt;p&gt;This article is based on public company investor-relations materials, SEC filings, and industry sources. Backlogs and long-term agreements improve visibility but remain subject to usage, recognition timing, pricing, and contract terms. This is informational analysis, not a recommendation to buy or sell securities.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;p&gt;&lt;em&gt;Disclaimer: For research and information purposes only. Not investment advice. Names cited are for analytical illustration; readers should perform their own due diligence and consult licensed advisors before any investment decision.&lt;/em&gt;&lt;/p&gt;</description></item><item><title>Kimi K3 Resets the AI Price Curve: From Kimi Linear to HBM and Big Tech Strategy</title><link>https://koreainvestinsights.com/post/kimi-k3-linear-api-pricing-semiconductor-big-tech-impact-2026-07-17/</link><pubDate>Fri, 17 Jul 2026 12:31:36 +0900</pubDate><guid>https://koreainvestinsights.com/post/kimi-k3-linear-api-pricing-semiconductor-big-tech-impact-2026-07-17/</guid><description>&lt;p&gt;Kimi K3 launched on July 16, 2026 at $3 per million uncached input tokens and $15 per million output tokens. That is not the familiar ultra-low-price positioning of a Chinese challenger. It matches Claude Sonnet 5&amp;rsquo;s standard price and sits 40% below GPT-5.6 Sol on input and 50% below it on output. Moonshot AI is positioning K3 as a primary enterprise model, not a budget fallback.&lt;/p&gt;
&lt;p&gt;The architecture is designed to support that claim. K3 has 2.8 trillion total parameters but effectively activates 16 of 896 experts per token. Kimi Delta Attention controls the cost of long context, Attention Residuals selectively retrieve earlier representations across depth, and quantization-aware training uses MXFP4 weights with MXFP8 activations. Moonshot recommends a supernode with at least 64 accelerators.&lt;/p&gt;
&lt;p&gt;That combination creates a two-sided semiconductor outcome. Memory and compute per token can fall while the number of institutions able to deploy a frontier-class model, and the amount of work they run, can rise. K3 is both an efficiency technology and an infrastructure workload.&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Related reading: &lt;a class="link" href="https://koreainvestinsights.com/post/ai-token-value-memory-value-added-2026-07-09/" &gt;AI token value and memory value capture&lt;/a&gt; / &lt;a class="link" href="https://koreainvestinsights.com/post/ai-token-futures-cost-per-token-korea-semiconductor-thesis-2026-05-30/" &gt;AI token futures and cost per token&lt;/a&gt; / &lt;a class="link" href="https://koreainvestinsights.com/post/us-china-agentic-inference-stack-sram-opportunity-2026-07-09/" &gt;US-China divergence in agentic inference infrastructure&lt;/a&gt; / &lt;a class="link" href="https://koreainvestinsights.com/post/hbm-2030-supply-demand-267eb-demand-model-crosscheck-2026-07-13/" &gt;Cross-checking the 2030 HBM shortage model&lt;/a&gt;&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="executive-summary"&gt;Executive Summary
&lt;/h2&gt;&lt;ol&gt;
&lt;li&gt;K3 combines 2.8T parameters, native vision and a 1M-token context window. Its products and API are live, but as of July 17 the full weights, technical report and license have not been released. The open-weight thesis must be verified after the July 27 deadline.&lt;/li&gt;
&lt;li&gt;Moonshot&amp;rsquo;s official GDPval-AA v2 score is 1,668, not 1,687. AA-Briefcase is 1,548. BrowseComp 91.2 uses context compaction; the no-compaction 1M-context result is 90.4. The results are strong, but mixed harnesses and company-run evaluations require independent replication.&lt;/li&gt;
&lt;li&gt;Kimi Linear&amp;rsquo;s claims of up to 75% lower KV cache and up to 6.3x theoretical decoding throughput at 1M context come from a 48B-total, 3B-active research model. They should not be presented as measured K3 API performance.&lt;/li&gt;
&lt;li&gt;K3 is not a low-cost model. It matches Sonnet 5&amp;rsquo;s standard price, is 50% more expensive than Sonnet&amp;rsquo;s temporary launch promotion, and is cheaper than GPT-5.6 Sol. Because only max reasoning is currently available and independent tests show high output-token use, cost per completed task may be less attractive than the rate card implies.&lt;/li&gt;
&lt;li&gt;The semiconductor impact pits lower compute and KV cache per request against more self-hosted deployments and higher total workload. Server DRAM and enterprise SSDs are the cleanest second-order beneficiaries because long-lived agent context spills below HBM into disaggregated cache tiers.&lt;/li&gt;
&lt;li&gt;The model and cloud layers experience opposite economics. OpenAI, Anthropic and Gemini API pricing face pressure, while Azure, AWS and Google Cloud can monetize K3 and other models through compute, storage and networking. Meta must defend US open-weight leadership.&lt;/li&gt;
&lt;li&gt;The July 27 proof points are the license, full weights, external evaluation, throughput on NVIDIA and AMD, support on Chinese accelerators, actual output tokens per task, vLLM compatibility and cloud catalog adoption.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;&lt;img alt="Kimi K3 pricing strategy and AI infrastructure impact map" loading="lazy" sizes="(max-width: 767px) calc(100vw - 30px), (max-width: 1023px) 700px, (max-width: 1279px) 950px, 1232px" src="https://koreainvestinsights.com/images/posts/kimi-k3-pricing-infrastructure-impact-2026-07-17.png"&gt;
&lt;/p&gt;
&lt;h2 id="1-correcting-the-numbers-before-drawing-conclusions"&gt;1. Correcting the Numbers Before Drawing Conclusions
&lt;/h2&gt;&lt;p&gt;Several numbers changed as the launch circulated through social media. The official values matter because some differences alter the investment interpretation.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Official value&lt;/th&gt;
 &lt;th&gt;Investor interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Total parameters&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2.8T&lt;/td&gt;
 &lt;td&gt;Total model size, not per-token active compute&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Expert routing&lt;/td&gt;
 &lt;td style="text-align: right"&gt;16 of 896&lt;/td&gt;
 &lt;td&gt;Extremely sparse MoE; routing and communication become first-order constraints&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Context&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1M tokens&lt;/td&gt;
 &lt;td&gt;Useful for repositories and research, but task cost depends on output length and cache hits&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;GDPval-AA v2&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1,668&lt;/td&gt;
 &lt;td&gt;The official table does not show 1,687&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AA-Briefcase&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1,548&lt;/td&gt;
 &lt;td&gt;Above GPT-5.6 Sol at 1,495, below Claude Fable 5 at 1,583&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;BrowseComp&lt;/td&gt;
 &lt;td style="text-align: right"&gt;91.2&lt;/td&gt;
 &lt;td&gt;Uses compaction starting at 300K tokens&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;BrowseComp without compaction&lt;/td&gt;
 &lt;td style="text-align: right"&gt;90.4&lt;/td&gt;
 &lt;td&gt;The cleaner result for the native 1M-context claim&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Product availability&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Web, Work, Code and API live&lt;/td&gt;
 &lt;td&gt;Commercial testing can begin now&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Weights&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Promised by July 27&lt;/td&gt;
 &lt;td&gt;License and complete artifacts remain unverified as of July 17&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Reasoning effort&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Max only&lt;/td&gt;
 &lt;td&gt;Low-cost modes are not yet available&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Moonshot explicitly states that K3 still trails Claude Fable 5 and GPT-5.6 Sol in overall user experience. At the same time, it reports frontier-level performance across coding, knowledge work and multimodal benchmarks. The credible interpretation is not that China has conclusively taken the lead. It is that a Chinese model has entered the performance band immediately below the strongest proprietary systems while promising a 1M context and downloadable weights.&lt;/p&gt;
&lt;p&gt;The benchmark table is not a single controlled tournament. K3 runs at maximum reasoning effort. Depending on the task, models use Kimi Code, Claude Code or Codex, and some competitor scores are the best results across harnesses or are imported from external leaderboards. Independent reproduction remains necessary.&lt;/p&gt;
&lt;p&gt;Still, Terminal-Bench 2.1 at 88.3, FrontierSWE at 81.2, SWE Marathon at 42.0, AutomationBench at 30.8, GPQA-Diamond at 93.5 and MMMU-Pro at 81.6 indicate that K3 is designed for long-horizon tool use and coding, not merely chat. The more important signal is the package: near-frontier capability, 1M context and promised open weights.&lt;/p&gt;
&lt;h2 id="2-how-a-28t-model-becomes-deployable"&gt;2. How a 2.8T Model Becomes Deployable
&lt;/h2&gt;&lt;h3 id="21-total-parameters-are-not-active-parameters"&gt;2.1 Total parameters are not active parameters
&lt;/h3&gt;&lt;p&gt;Computing all 2.8T parameters for every token would be prohibitively expensive. Stable LatentMoE effectively activates 16 of 896 experts, or about 1.8% of the expert pool. The technical report is needed to establish exact active parameter counts, but total parameters clearly do not equal per-token compute.&lt;/p&gt;
&lt;p&gt;Sparse MoE shifts rather than eliminates bottlenecks.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;What improves&lt;/th&gt;
 &lt;th&gt;What becomes harder&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Active compute per token&lt;/td&gt;
 &lt;td&gt;Router quality and expert balance&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Compute needed for a quality target&lt;/td&gt;
 &lt;td&gt;Communication across accelerators&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Some inference costs&lt;/td&gt;
 &lt;td&gt;Avoiding idle accelerators and hot experts&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Scaling model capacity&lt;/td&gt;
 &lt;td&gt;Keeping the full weight set accessible&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This is why Moonshot recommends a supernode with 64 or more accelerators. Even when only a small subset of experts is active, the selected expert is not known in advance and the full model must remain accessible. At four bits, 2.8T weights imply a theoretical minimum of about 1.4 TB before scales, metadata, buffers, cache and replication. Sparse activation reduces arithmetic but does not remove model-residency memory or fabric requirements.&lt;/p&gt;
&lt;h3 id="22-kimi-linear-addresses-long-context-cost"&gt;2.2 Kimi Linear addresses long-context cost
&lt;/h3&gt;&lt;p&gt;The Kimi Linear paper predates K3 and evaluates a 48B-total, 3B-active research model, not K3 itself. It combines Kimi Delta Attention with full Multi-head Latent Attention in a 3:1 ratio.&lt;/p&gt;
&lt;p&gt;Full attention is strong at exact copying and fine-grained retrieval, but KV cache grows with context. Linear attention compresses history into a fixed-size state, reducing sequence-length dependence, but can lose exact detail. Kimi Linear uses three KDA layers followed by one full-attention layer to balance efficiency and expressiveness.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Component&lt;/th&gt;
 &lt;th&gt;Role&lt;/th&gt;
 &lt;th&gt;Trade-off&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;KDA&lt;/td&gt;
 &lt;td&gt;Compress long history into fixed-size state&lt;/td&gt;
 &lt;td&gt;Can weaken exact copying and fine retrieval&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Full MLA&lt;/td&gt;
 &lt;td&gt;Restores precise token-to-token recall&lt;/td&gt;
 &lt;td&gt;KV cache still grows with context&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;3:1 hybrid&lt;/td&gt;
 &lt;td&gt;Balances efficiency and quality&lt;/td&gt;
 &lt;td&gt;Requires more complex kernels and serving software&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The paper reports up to 75% lower KV cache and up to 6.3x theoretical decoding throughput at 1M context. A measured comparison in the complexity analysis reports 2.3x acceleration. These results establish direction, not guaranteed K3 API performance.&lt;/p&gt;
&lt;p&gt;For investors, lower KV cache means more concurrent requests per accelerator. It also makes longer repositories, more documents and persistent agents economical. Memory per token can fall while total tokens rise.&lt;/p&gt;
&lt;h3 id="23-attention-residuals-improve-depth-efficiency"&gt;2.3 Attention Residuals improve depth efficiency
&lt;/h3&gt;&lt;p&gt;Standard residual connections keep adding earlier layer outputs. In very deep networks, useful representations can be diluted. Attention Residuals allow a layer to select the earlier representations it needs. Block AttnRes groups layers and retains block-level representations, reducing memory overhead.&lt;/p&gt;
&lt;p&gt;Moonshot&amp;rsquo;s research says roughly eight blocks recover most of the gains, and Block AttnRes can match a baseline trained with about 1.25x compute. This improves training capital efficiency, but it does not automatically reduce data-center demand. Better efficiency can be reinvested into larger models and longer tasks.&lt;/p&gt;
&lt;h3 id="24-quantization-is-a-hardware-portability-strategy"&gt;2.4 Quantization is a hardware-portability strategy
&lt;/h3&gt;&lt;p&gt;K3 applies quantization-aware training from supervised fine-tuning onward, using MXFP4 weights and MXFP8 activations. This should control accuracy loss better than aggressive post-training quantization and make deployment across multiple hardware platforms easier.&lt;/p&gt;
&lt;p&gt;Moonshot&amp;rsquo;s kernel arena included NVIDIA H200 and a general-purpose GPU from an alternative vendor. The official post does not name a Chinese chip or claim H200-equivalent economics. What is verified is the strategic intent to optimize outside NVIDIA as well as on NVIDIA.&lt;/p&gt;
&lt;p&gt;That matters for both China and global buyers. China needs frontier-level models that can survive constrained access to top NVIDIA systems. Other buyers want bargaining power across NVIDIA, AMD and custom accelerators.&lt;/p&gt;
&lt;h2 id="3-what-the-price-card-reveals"&gt;3. What the Price Card Reveals
&lt;/h2&gt;&lt;h3 id="31-k3-is-priced-as-a-primary-model"&gt;3.1 K3 is priced as a primary model
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Model&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Cached input per 1M&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Standard input&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Output&lt;/th&gt;
 &lt;th&gt;Current positioning&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Kimi K3&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$0.30&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$3&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$15&lt;/td&gt;
 &lt;td&gt;Frontier primary-model pricing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Claude Sonnet 5 launch promo&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Varies&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$2&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$10&lt;/td&gt;
 &lt;td&gt;Temporary through August 31&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Claude Sonnet 5 standard&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Varies&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$3&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$15&lt;/td&gt;
 &lt;td&gt;Same headline price as K3&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;GPT-5.6 Sol&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$0.50&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$5&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$30&lt;/td&gt;
 &lt;td&gt;67% higher input and 100% higher output than K3&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;K3 is more expensive than Sonnet 5&amp;rsquo;s current promotion. It is inaccurate to describe it as universally half-price. The clearer strategy is to match Sonnet&amp;rsquo;s standard tier while undercutting the most expensive frontier API.&lt;/p&gt;
&lt;p&gt;The price is both a confidence signal and a monetization decision. Moonshot is no longer accepting a deep discount simply to acquire usage. It is claiming a position in the enterprise default-model tier.&lt;/p&gt;
&lt;h3 id="32-cost-per-completed-task-matters-more-than-cost-per-token"&gt;3.2 Cost per completed task matters more than cost per token
&lt;/h3&gt;&lt;p&gt;Rate cards assume equal token use. Agents differ in planning length, tool calls, retries and verbosity. K3 currently exposes only max reasoning. Artificial Analysis reports that K3 used about 130 million output tokens in its Intelligence Index evaluation, more than twice the peer median of roughly 63 million. A cheaper output token can still lead to a costly completed task.&lt;/p&gt;
&lt;p&gt;&lt;code&gt;Task cost = input tokens × input rate + output tokens × output rate + tool and retry cost&lt;/code&gt;&lt;/p&gt;
&lt;p&gt;Enterprise buyers will monitor task success, output tokens, time to first token, throughput, tool reliability, cache hit rate and session stability. Moonshot says Mooncake delivers more than 90% cache hits on coding workloads, making cached input one-tenth the uncached price. If that rate holds on enterprise traffic, K3&amp;rsquo;s effective economics improve materially.&lt;/p&gt;
&lt;h3 id="33-mooncake-moves-memory-demand-down-the-hierarchy"&gt;3.3 Mooncake moves memory demand down the hierarchy
&lt;/h3&gt;&lt;p&gt;Mooncake separates prefill and decode clusters and disaggregates KV cache across CPU, DRAM and SSD rather than keeping everything in GPU HBM. Its paper reports up to 525% higher throughput in simulation and 75% more requests on production workloads.&lt;/p&gt;
&lt;p&gt;This explains why AI memory demand extends beyond HBM.&lt;/p&gt;
&lt;p&gt;&lt;code&gt;Accelerator HBM → server DRAM → enterprise SSD → remote cache tier&lt;/code&gt;&lt;/p&gt;
&lt;p&gt;KDA can reduce KV cache per request, yet persistent 1M-context agents increase total cache volume and retention time. Hot data remains in HBM and DRAM; colder context moves to SSD. Efficiency can soften an HBM-only thesis while strengthening the broader memory hierarchy.&lt;/p&gt;
&lt;h2 id="4-chinese-open-weights-move-up-the-enterprise-stack"&gt;4. Chinese Open Weights Move Up the Enterprise Stack
&lt;/h2&gt;&lt;p&gt;OpenRouter reports that Chinese models surpassed US models in token share on its platform in early June 2026, based on more than 450 trillion tokens from January through June 14. DeepSeek&amp;rsquo;s share rose from roughly 9% to 18% and it became the leading provider by mid-May.&lt;/p&gt;
&lt;p&gt;The adoption path has three stages:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Low-cost open models take classification, translation and testing workloads.&lt;/li&gt;
&lt;li&gt;Better coding and agent performance take repetitive enterprise workflows.&lt;/li&gt;
&lt;li&gt;Near-frontier quality and million-token context compete for primary routing.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;K3&amp;rsquo;s price is designed for the third stage. It is not following the most aggressive low-price strategy. It is charging a frontier-tier API price while promising weights that clouds, governments and enterprises can deploy themselves.&lt;/p&gt;
&lt;p&gt;Proprietary APIs and open weights accumulate different strategic assets.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Proprietary frontier API&lt;/th&gt;
 &lt;th&gt;Near-frontier open weights&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Controls the best UX and capability&lt;/td&gt;
 &lt;td&gt;Multiplies deployment routes and hardware options&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Centralizes usage data&lt;/td&gt;
 &lt;td&gt;Lets enterprises retain data and operations&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Changes pricing and policy centrally&lt;/td&gt;
 &lt;td&gt;Released files are difficult to withdraw&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Model provider captures gross margin&lt;/td&gt;
 &lt;td&gt;Cloud, chip and application vendors share value&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The strongest proprietary model can remain number one while losing paid token share. Enterprises can route repetitive work to K3-class models and reserve the most difficult legal, design and research tasks for the top closed model. Paid token mix and blended price matter more than leaderboard rank.&lt;/p&gt;
&lt;h2 id="5-semiconductor-demand-efficiency-and-diffusion-arrive-together"&gt;5. Semiconductor Demand: Efficiency and Diffusion Arrive Together
&lt;/h2&gt;&lt;h3 id="51-nvidia-near-term-efficiency-risk-medium-term-deployment-elasticity"&gt;5.1 NVIDIA: near-term efficiency risk, medium-term deployment elasticity
&lt;/h3&gt;&lt;p&gt;Sparse MoE, KDA, quantization and autonomous kernel optimization reduce GPU time per task. Hardware portability also weakens the assumption that every frontier workload must stay on CUDA. Those are negative valuation signals for NVIDIA.&lt;/p&gt;
&lt;p&gt;The positive side is equally material. Moonshot recommends at least 64 accelerators for self-hosted K3. Released weights could drive clouds, governments, laboratories and large enterprises to build their own clusters. Demand concentrated behind one API becomes hardware demand across many data centers.&lt;/p&gt;
&lt;p&gt;&lt;code&gt;Total accelerator demand = lower GPU time per task × higher total workload × more self-hosting institutions&lt;/code&gt;&lt;/p&gt;
&lt;p&gt;The first term is negative; the last two are positive. K3 alone is not enough to cut NVIDIA earnings estimates, but neither is it enough to assume that every efficiency gain creates more GPU demand. Workload elasticity is the deciding variable.&lt;/p&gt;
&lt;h3 id="52-amd-optionality-from-hardware-choice"&gt;5.2 AMD: optionality from hardware choice
&lt;/h3&gt;&lt;p&gt;AMD benefits if MXFP4, MXFP8 and vLLM portability allow enterprises to separate model choice from accelerator choice. A near-frontier open model gives buyers a realistic workload on which to test NVIDIA alternatives.&lt;/p&gt;
&lt;p&gt;The proof must come after the weights. ROCm kernels, expert-parallel communication, 1M-context throughput and 64-accelerator stability need to be measured. AMD&amp;rsquo;s upside grows only if K3 demonstrates superior cost per successful task on MI systems.&lt;/p&gt;
&lt;h3 id="53-hbm-lower-per-request-cache-larger-model-residency"&gt;5.3 HBM: lower per-request cache, larger model residency
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Demand layer&lt;/th&gt;
 &lt;th&gt;K3 efficiency impact&lt;/th&gt;
 &lt;th&gt;Direction&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Weight residency&lt;/td&gt;
 &lt;td&gt;A 2.8T model must remain quickly accessible&lt;/td&gt;
 &lt;td&gt;More accelerators and high-bandwidth memory&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Per-request KV cache&lt;/td&gt;
 &lt;td&gt;KDA lowers cache size and growth&lt;/td&gt;
 &lt;td&gt;Less HBM capacity per request&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Concurrent users and agents&lt;/td&gt;
 &lt;td&gt;Lower cost and open deployment can expand workloads&lt;/td&gt;
 &lt;td&gt;More total HBM and DRAM&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The headlines can be negative for SK hynix, Micron and Samsung because 75% lower KV cache and 6.3x throughput sound like lower memory intensity. Those numbers belong to a Kimi Linear research model, not measured K3 production. HBM also stores weights, activations, communication buffers and batches, not only KV cache.&lt;/p&gt;
&lt;p&gt;The medium-term balance is neutral to positive if self-hosted clusters and agent workloads grow faster than efficiency. It turns negative if workload elasticity is weak and compression improves faster than usage.&lt;/p&gt;
&lt;h3 id="54-server-dram-and-enterprise-ssds-are-the-clearest-second-order-beneficiaries"&gt;5.4 Server DRAM and enterprise SSDs are the clearest second-order beneficiaries
&lt;/h3&gt;&lt;p&gt;Long context and cache reuse cannot remain entirely in HBM. Once serving separates prefill and decode and spills cache across CPU, DRAM and SSD, server DRAM and enterprise SSD become operating assets for AI inference.&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Samsung can sell HBM, server DRAM, high-capacity SSDs, foundry and packaging.&lt;/li&gt;
&lt;li&gt;SK hynix combines HBM and server DRAM with Solidigm enterprise SSDs.&lt;/li&gt;
&lt;li&gt;Micron supplies HBM, data-center DRAM and SSDs.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;K3 does not show that AI needs less memory. It shows that AI memory is becoming hierarchical. The hottest data stays in HBM, retained context sits in server DRAM, and colder cache moves to enterprise SSD. Vendors with a full memory stack have greater resilience than a single-product HBM thesis.&lt;/p&gt;
&lt;h3 id="55-networking-and-custom-silicon-are-the-hidden-moe-bottlenecks"&gt;5.5 Networking and custom silicon are the hidden MoE bottlenecks
&lt;/h3&gt;&lt;p&gt;Spreading 896 experts across accelerators creates variable communication depending on token routing. This is why Moonshot emphasizes balanced expert-parallel training, static shapes and removal of host synchronization from the critical path. Efficient operation across 64 or more accelerators requires high-bandwidth scale-up and scale-out fabric.&lt;/p&gt;
&lt;p&gt;That is structurally positive for Broadcom, Marvell, NVIDIA networking, optical interconnect and switch silicon. It also encourages inference ASICs optimized for open models. K3&amp;rsquo;s 48-hour small-chip design exercise is not a commercial product, but it illustrates faster model-hardware co-design.&lt;/p&gt;
&lt;h2 id="6-us-big-tech-strategy-and-stock-transmission"&gt;6. US Big Tech Strategy and Stock Transmission
&lt;/h2&gt;&lt;h3 id="61-microsoft-pressure-on-openai-economics-more-azure-usage"&gt;6.1 Microsoft: pressure on OpenAI economics, more Azure usage
&lt;/h3&gt;&lt;p&gt;Microsoft owns exposure to OpenAI IP and economics as well as Azure infrastructure. The April 2026 partnership update keeps Microsoft as a primary cloud partner and extends a non-exclusive IP license through 2032.&lt;/p&gt;
&lt;p&gt;K3 pressures the model layer because repetitive work can move away from OpenAI APIs. It can support the cloud layer if Azure hosts K3 as managed or self-hosted infrastructure.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Microsoft layer&lt;/th&gt;
 &lt;th&gt;K3 impact&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;OpenAI revenue share&lt;/td&gt;
 &lt;td&gt;Negative through price and mix&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Copilot&lt;/td&gt;
 &lt;td&gt;Positive if routing costs fall&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Azure AI&lt;/td&gt;
 &lt;td&gt;Positive if multi-model usage rises&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Custom silicon and data centers&lt;/td&gt;
 &lt;td&gt;Positive if open-weight optimization expands&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The near-term stock impact is neutral. The medium-term question is whether Azure converts model price deflation into usage and Copilot margin.&lt;/p&gt;
&lt;h3 id="62-amazon-anthropic-and-aws-have-different-economics"&gt;6.2 Amazon: Anthropic and AWS have different economics
&lt;/h3&gt;&lt;p&gt;Amazon is a major Anthropic investor and the provider of AWS, Bedrock, Trainium and Inferentia. K3 can reduce Claude pricing power and the value of Amazon&amp;rsquo;s Anthropic stake. If enterprises run K3 on AWS, however, EC2, Bedrock, storage, networking and Trainium usage can rise.&lt;/p&gt;
&lt;p&gt;Amazon&amp;rsquo;s optimal strategy is to keep the workload on AWS regardless of which model wins. If K3 arrives with a permissive license and efficient Trainium support, AWS can monetize a competitor&amp;rsquo;s diffusion. The stock impact is neutral to modestly positive, although inference price competition could lower cloud margins even as revenue grows.&lt;/p&gt;
&lt;h3 id="63-alphabet-gemini-pricing-pressure-tpu-and-vertex-defense"&gt;6.3 Alphabet: Gemini pricing pressure, TPU and Vertex defense
&lt;/h3&gt;&lt;p&gt;Google controls a model, a chip, a deployment platform and final demand through Search, Ads and Workspace. Vertex Model Garden supports first-party, third-party and open models.&lt;/p&gt;
&lt;p&gt;K3 pressures Gemini API pricing but can create TPU and Google Cloud demand. Lower model cost also reduces the expense of AI Overviews, ad generation and Workspace agents. The stock impact is neutral to positive because Google is not solely a model vendor. The risk is that Gemini loses both performance and price leadership, raising cloud customer-acquisition cost.&lt;/p&gt;
&lt;h3 id="64-meta-from-open-weight-beneficiary-to-defender"&gt;6.4 Meta: from open-weight beneficiary to defender
&lt;/h3&gt;&lt;p&gt;Meta benefited from open-weight diffusion by commoditizing competitors&amp;rsquo; APIs and lowering its own recommendation, advertising and content costs. If Chinese labs release near-frontier capability and longer context first, Meta risks losing its position as the benchmark US open-weight ecosystem.&lt;/p&gt;
&lt;p&gt;K3 forces two responses: the next Llama must compete on long context, agent tools and deployment cost, and Meta must provide a trusted US alternative for enterprises and governments unwilling to deploy Chinese weights. The near-term earnings impact is limited because advertising drives cash flow. The strategic risk is lower returns on AI infrastructure and developer ecosystem investment if Llama falls behind.&lt;/p&gt;
&lt;h3 id="65-existing-market-positioning-matters-more-than-one-launch"&gt;6.5 Existing market positioning matters more than one launch
&lt;/h3&gt;&lt;p&gt;From June 22 through July 16, before most of the K3 evidence could affect trading, the US AI complex had already diverged sharply.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Company&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Return&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Drawdown from period high&lt;/th&gt;
 &lt;th&gt;Positioning signal&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Meta&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+17.9%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-3.1%&lt;/td&gt;
 &lt;td&gt;Strong ad cash flow and AI utilization expectations&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Microsoft&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+9.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.2%&lt;/td&gt;
 &lt;td&gt;Cloud and software resilience&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Amazon&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+7.3%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-3.2%&lt;/td&gt;
 &lt;td&gt;AWS and consumer recovery expectations&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Alphabet&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+1.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-5.5%&lt;/td&gt;
 &lt;td&gt;Mixed search and cloud positioning&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NVIDIA&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-0.6%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-3.1%&lt;/td&gt;
 &lt;td&gt;Relatively resilient&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Broadcom&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-4.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-9.7%&lt;/td&gt;
 &lt;td&gt;Custom AI optimism meets valuation pressure&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AMD&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-9.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-14.3%&lt;/td&gt;
 &lt;td&gt;Alternative accelerator optionality with high volatility&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-29.6%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-32.0%&lt;/td&gt;
 &lt;td&gt;Correction after elevated memory expectations&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Oracle&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-29.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-32.7%&lt;/td&gt;
 &lt;td&gt;Tension between AI infrastructure growth and financing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Marvell&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-38.8%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-40.1%&lt;/td&gt;
 &lt;td&gt;Severe derating in networking and custom silicon&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;These are not K3-caused returns. They show the positions into which K3 arrived. A relatively resilient NVIDIA may be more sensitive to efficiency headlines, while already-corrected Micron and Marvell could react more strongly if the released weights create verifiable infrastructure demand.&lt;/p&gt;
&lt;h2 id="7-company-impact-map"&gt;7. Company Impact Map
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Company or layer&lt;/th&gt;
 &lt;th&gt;Near-term signal&lt;/th&gt;
 &lt;th&gt;Medium-term path&lt;/th&gt;
 &lt;th&gt;What to do now&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;NVIDIA&lt;/td&gt;
 &lt;td&gt;Valuation pressure from efficiency and portability&lt;/td&gt;
 &lt;td&gt;Self-hosted 64+ accelerator clusters offset efficiency&lt;/td&gt;
 &lt;td&gt;Watch workload elasticity, do not change estimates on the launch alone&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AMD&lt;/td&gt;
 &lt;td&gt;Alternative deployment optionality&lt;/td&gt;
 &lt;td&gt;Share upside if ROCm economics are proven&lt;/td&gt;
 &lt;td&gt;Wait for post-July 27 throughput&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Broadcom and Marvell&lt;/td&gt;
 &lt;td&gt;Networking complex already derating&lt;/td&gt;
 &lt;td&gt;MoE expert parallelism raises fabric demand&lt;/td&gt;
 &lt;td&gt;Verify orders and actual K3 cluster topology&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK hynix&lt;/td&gt;
 &lt;td&gt;Sensitive to KV-cache efficiency headlines&lt;/td&gt;
 &lt;td&gt;HBM, server DRAM and Solidigm SSD hierarchy exposure&lt;/td&gt;
 &lt;td&gt;Value the full memory stack, not HBM alone&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td&gt;HBM efficiency risk plus catch-up position&lt;/td&gt;
 &lt;td&gt;HBM, server DRAM, SSD and foundry optionality&lt;/td&gt;
 &lt;td&gt;Broadest portfolio, execution still required&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron&lt;/td&gt;
 &lt;td&gt;High expectations already corrected&lt;/td&gt;
 &lt;td&gt;Integrated US AI memory and storage exposure&lt;/td&gt;
 &lt;td&gt;Watch deployment volume versus pricing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Microsoft&lt;/td&gt;
 &lt;td&gt;OpenAI price and mix pressure&lt;/td&gt;
 &lt;td&gt;Azure and Copilot cost leverage&lt;/td&gt;
 &lt;td&gt;Cloud usage matters more than model margin&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Amazon&lt;/td&gt;
 &lt;td&gt;Pressure on Anthropic asset value&lt;/td&gt;
 &lt;td&gt;AWS, Bedrock and Trainium benefit from model choice&lt;/td&gt;
 &lt;td&gt;The cleanest internal offset&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Alphabet&lt;/td&gt;
 &lt;td&gt;Gemini price pressure&lt;/td&gt;
 &lt;td&gt;TPU, Vertex and Search cost benefits&lt;/td&gt;
 &lt;td&gt;Application-layer defense remains strong&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Meta&lt;/td&gt;
 &lt;td&gt;Pressure on US open-weight leadership&lt;/td&gt;
 &lt;td&gt;Llama acceleration or broader open ecosystem&lt;/td&gt;
 &lt;td&gt;Strategic impact exceeds near-term earnings impact&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;There is not enough evidence for a new buy or sell call from this launch alone. Weights, license and external hardware throughput remain unavailable. The order of proof matters more than the direction of the narrative.&lt;/p&gt;
&lt;h2 id="8-three-scenarios"&gt;8. Three Scenarios
&lt;/h2&gt;&lt;h3 id="base-case-strong-model-limited-re-rating"&gt;Base case: strong model, limited re-rating
&lt;/h3&gt;&lt;p&gt;Subjective probability: 55%. Full weights and a reasonably permissive license arrive by July 27. External results reproduce 85% to 95% of official performance. K3 runs on NVIDIA and AMD, but 64+ accelerators, max reasoning and high output-token use keep deployment expensive.&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Proprietary APIs face more price pressure on repetitive work.&lt;/li&gt;
&lt;li&gt;Clouds gain K3 hosting and self-deployment demand.&lt;/li&gt;
&lt;li&gt;Per-token efficiency offsets higher workload.&lt;/li&gt;
&lt;li&gt;HBM is neutral to modestly positive.&lt;/li&gt;
&lt;li&gt;Server DRAM, enterprise SSD and networking are positive.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="bull-case-open-weights-enter-primary-enterprise-routing"&gt;Bull case: open weights enter primary enterprise routing
&lt;/h3&gt;&lt;p&gt;Subjective probability: 25%. The license is close to MIT, vLLM and SGLang support is stable, and NVIDIA, AMD and Chinese accelerators show strong throughput. External evaluation confirms performance immediately below the top proprietary models. Lower reasoning modes reduce task cost. Major clouds and enterprise gateways add K3 to default routing.&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Proprietary model blended price and token share fall.&lt;/li&gt;
&lt;li&gt;Hyperscaler multi-model usage rises.&lt;/li&gt;
&lt;li&gt;Self-serving clusters increase accelerator demand.&lt;/li&gt;
&lt;li&gt;HBM, networking, server DRAM and enterprise SSD benefit from deployment volume.&lt;/li&gt;
&lt;li&gt;Meta and US open-model programs accelerate investment.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="bear-case-weights-reveal-a-cost-and-quality-gap"&gt;Bear case: weights reveal a cost and quality gap
&lt;/h3&gt;&lt;p&gt;Subjective probability: 20%. Weights are delayed or restricted, external scores fall below the official table, preserved-thinking-history requirements and excessive proactiveness create enterprise failures, and task cost exceeds Sonnet 5 because of max reasoning and verbosity. The 64-accelerator requirement limits self-hosting.&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Moonshot may have to retreat from frontier pricing.&lt;/li&gt;
&lt;li&gt;Proprietary APIs retain a UX and reliability premium.&lt;/li&gt;
&lt;li&gt;Incremental semiconductor demand remains limited.&lt;/li&gt;
&lt;li&gt;Existing earnings and capex regain control of stock prices.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="9-the-july-27-verification-checklist"&gt;9. The July 27 Verification Checklist
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Proof point&lt;/th&gt;
 &lt;th&gt;Strong signal&lt;/th&gt;
 &lt;th&gt;Weak signal&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Weights and license&lt;/td&gt;
 &lt;td&gt;Full release on time with commercial modification rights&lt;/td&gt;
 &lt;td&gt;Delay, restrictions or missing components&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;External evaluation&lt;/td&gt;
 &lt;td&gt;Official scores broadly reproduced under one harness&lt;/td&gt;
 &lt;td&gt;Large drop versus company table&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cost per task&lt;/td&gt;
 &lt;td&gt;Lower reasoning modes and fewer output tokens&lt;/td&gt;
 &lt;td&gt;Max-only reasoning and persistent verbosity&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NVIDIA throughput&lt;/td&gt;
 &lt;td&gt;Stable expert parallelism on 64-accelerator nodes&lt;/td&gt;
 &lt;td&gt;Fabric bottlenecks and low utilization&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AMD throughput&lt;/td&gt;
 &lt;td&gt;Cost advantage under ROCm and vLLM&lt;/td&gt;
 &lt;td&gt;Immature kernels or accuracy loss&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Chinese accelerators&lt;/td&gt;
 &lt;td&gt;Named hardware and measured throughput&lt;/td&gt;
 &lt;td&gt;Only “alternative GPU” language&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Caching&lt;/td&gt;
 &lt;td&gt;Near-90% hits on enterprise traffic&lt;/td&gt;
 &lt;td&gt;Sharp decline outside coding&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cloud adoption&lt;/td&gt;
 &lt;td&gt;AWS, Azure, Google Cloud or Oracle listings&lt;/td&gt;
 &lt;td&gt;Limited to a few Chinese platforms&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Competitive pricing&lt;/td&gt;
 &lt;td&gt;Standard price cuts or wider cache discounts&lt;/td&gt;
 &lt;td&gt;Temporary promotions only&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Memory orders&lt;/td&gt;
 &lt;td&gt;Upward revisions to HBM, server DRAM and SSD volumes&lt;/td&gt;
 &lt;td&gt;Efficiency rises while volumes stagnate&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="10-the-strongest-counterargument"&gt;10. The Strongest Counterargument
&lt;/h2&gt;&lt;p&gt;The strongest bear case is straightforward: semiconductor demand falls if efficiency improves faster than usage. Model compression, linear attention, quantization, cache reuse and inference ASICs could process the same number of useful tasks with far fewer GPUs and less HBM. Open-weight competition can reduce API prices without producing enough incremental paid work to justify AI capex.&lt;/p&gt;
&lt;p&gt;Evidence for that outcome would include slower inference revenue despite falling prices, higher accelerator utilization but fewer new clusters, HBM bit growth below model-efficiency gains, weak conversion of cloud AI backlog into revenue, and enterprises using open models only to cut costs rather than create new workflows.&lt;/p&gt;
&lt;p&gt;The bull case requires price declines to unlock new work: repository-scale coding, research automation, video editing, chip design and workflows that were previously uneconomic. Workload elasticity relative to model efficiency is the common proof point for both accelerator and HBM investing.&lt;/p&gt;
&lt;h2 id="11-conclusion"&gt;11. Conclusion
&lt;/h2&gt;&lt;p&gt;Kimi K3 does not prove that China has surpassed the strongest US proprietary models. Moonshot acknowledges the remaining UX gap. As of July 17, the full weights and technical report are not available, and a mixed-harness benchmark table cannot declare a winner.&lt;/p&gt;
&lt;p&gt;It does change market structure. A 2.8T, 1M-context, near-frontier model is live at Sonnet&amp;rsquo;s standard price, with weights promised within ten days. Chinese open weights are moving from cheap second-tier substitutes toward primary enterprise workloads.&lt;/p&gt;
&lt;p&gt;For semiconductors, the event is more demand reallocation than demand destruction. KDA and quantization reduce HBM and GPU time per request. Sparse MoE and 64-accelerator supernodes increase model-residency memory and fabric. Mooncake pushes cache into server DRAM and enterprise SSDs. The HBM-only story becomes less simple, while the full memory and data-center hierarchy remains exposed to wider deployment.&lt;/p&gt;
&lt;p&gt;US Big Tech faces the same split. Model APIs absorb price pressure; clouds and applications absorb lower model cost. Microsoft, Amazon and Alphabet can host K3 even if their preferred models lose share. Meta must defend its strategic role as the trusted US open-weight standard.&lt;/p&gt;
&lt;p&gt;On July 27, the important evidence is not the existence of a weight file. It is a permissive license, reproducible evaluation, multi-hardware throughput and competitive cost per completed task. If those conditions hold, K3 becomes an event that changes both the AI price curve and the semiconductor demand path. If they do not, it remains an impressive product launch rather than an industry reset.&lt;/p&gt;
&lt;h2 id="sources-and-limitations"&gt;Sources and Limitations
&lt;/h2&gt;&lt;p&gt;Primary materials: &lt;a class="link" href="https://www.kimi.com/blog/kimi-k3" target="_blank" rel="noopener"
 &gt;Kimi K3 launch&lt;/a&gt;, &lt;a class="link" href="https://platform.kimi.ai/docs/pricing/chat-k3" target="_blank" rel="noopener"
 &gt;Kimi K3 API documentation&lt;/a&gt;, &lt;a class="link" href="https://arxiv.org/abs/2510.26692" target="_blank" rel="noopener"
 &gt;Kimi Linear paper&lt;/a&gt;, &lt;a class="link" href="https://github.com/MoonshotAI/Kimi-Linear" target="_blank" rel="noopener"
 &gt;Kimi Linear GitHub&lt;/a&gt;, &lt;a class="link" href="https://arxiv.org/abs/2603.15031" target="_blank" rel="noopener"
 &gt;Attention Residuals paper&lt;/a&gt;, &lt;a class="link" href="https://arxiv.org/abs/2407.00079" target="_blank" rel="noopener"
 &gt;Mooncake paper&lt;/a&gt;, &lt;a class="link" href="https://www.anthropic.com/news/claude-sonnet-5" target="_blank" rel="noopener"
 &gt;Claude Sonnet 5 pricing&lt;/a&gt;, &lt;a class="link" href="https://developers.openai.com/api/docs/models/gpt-5.6-sol" target="_blank" rel="noopener"
 &gt;GPT-5.6 Sol pricing&lt;/a&gt;, &lt;a class="link" href="https://openrouter.ai/blog/insights/deepseek-v4-adoption/" target="_blank" rel="noopener"
 &gt;OpenRouter model adoption analysis&lt;/a&gt;, &lt;a class="link" href="https://blogs.microsoft.com/blog/2026/04/27/the-next-phase-of-the-microsoft-openai-partnership/" target="_blank" rel="noopener"
 &gt;Microsoft-OpenAI partnership&lt;/a&gt;, &lt;a class="link" href="https://cloud.google.com/vertex-ai/generative-ai/docs/model-garden/explore-models" target="_blank" rel="noopener"
 &gt;Google Vertex Model Garden&lt;/a&gt;, and &lt;a class="link" href="https://www.aboutamazon.com/news/company-news/amazon-aws-anthropic-ai" target="_blank" rel="noopener"
 &gt;Amazon-Anthropic partnership&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;The semiconductor and stock transmission analysis is an inference from public architecture, serving and market data. Exact active parameters, weight size, license terms, AMD and Chinese-accelerator throughput, and enterprise cost per completed task remain blocked as of July 17. This article is for research and information purposes only and is not investment advice.&lt;/p&gt;</description></item><item><title>The US-China Split in Agentic Inference Infrastructure and the SRAM Opportunity</title><link>https://koreainvestinsights.com/post/us-china-agentic-inference-stack-sram-opportunity-2026-07-09/</link><pubDate>Thu, 09 Jul 2026 17:20:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/us-china-agentic-inference-stack-sram-opportunity-2026-07-09/</guid><description>
 &lt;blockquote&gt;
 &lt;p&gt;Related reading:
&lt;a class="link" href="https://koreainvestinsights.com/en/post/nvidia-vera-rubin-lpx-cmx-inference-stack-samsung-hbm-2026-05-28/" &gt;NVIDIA Vera Rubin inference stack&lt;/a&gt; /
&lt;a class="link" href="https://koreainvestinsights.com/en/post/ai-token-value-memory-value-added-2026-07-09/" &gt;AI token value and memory value capture&lt;/a&gt; /
&lt;a class="link" href="https://koreainvestinsights.com/en/post/ai-datacenter-power-bottleneck-korea-value-chain-screen-2026-07-04/" &gt;AI data center power bottleneck map&lt;/a&gt;&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;p&gt;The United States and China are both scaling AI inference, but they are solving different constraints. The US bottleneck is power, rack density, HBM, advanced packaging, and token-per-watt efficiency. China’s bottleneck is access to leading-edge logic and HBM, so it is more likely to push Ascend-based clusters, optical mesh, parallel scaling, and domestic memory hierarchy workarounds.&lt;/p&gt;
&lt;p&gt;For Korean listed equities, the investable landing zone is not China optical modules by default. The cleaner landing is the US-style AI factory bottleneck: HBM4/HBM4E, power equipment, HBM packaging equipment, and a less appreciated Samsung Electronics option around SRAM/LPU foundry, AI storage, and memory hierarchy.&lt;/p&gt;
&lt;p&gt;The business winner and the best stock setup are not the same thing. SK hynix has the cleanest HBM4 leadership but is already a crowded winner. HD Hyundai Electric has high purity to US data center power bottlenecks, but order momentum is widely recognized. Hanmi Semiconductor is a real HBM tool bottleneck but needs repeat orders and customer diversification. The most asymmetric candidate is Samsung Electronics, if HBM4E, SRAM/LPU foundry, and AI storage start to show up in numbers.&lt;/p&gt;
&lt;h2 id="1-what-is-being-verified"&gt;1. What Is Being Verified
&lt;/h2&gt;&lt;p&gt;This note starts from YS-VC’s article &lt;a href="https://www.ys-vc.com/blog/us-china-ai-inference-stack-divergence"&gt;“Power or Chips: Why US and Chinese AI Inference Stacks Are Diverging”&lt;/a&gt;. The core argument is directionally right: inference scaling is no longer a generic GPU story, and the US and China face different binding constraints.&lt;/p&gt;
&lt;p&gt;The adjustment is in the Korean equity landing. A China optical-mesh story does not automatically translate into Korean optical component revenue. China’s AI infrastructure supply chain is strategically closed and increasingly domestic. Korean equity exposure is more direct through components sold into the US-style AI infrastructure bottleneck.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Claim&lt;/th&gt;
 &lt;th&gt;Assessment&lt;/th&gt;
 &lt;th&gt;Investor read-through&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;US and China AI inference stacks are diverging&lt;/td&gt;
 &lt;td&gt;Mostly true&lt;/td&gt;
 &lt;td&gt;The US optimizes power and rack efficiency; China works around logic and HBM limits&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;The US stack is moving toward GPU/HBM/SRAM rack-scale inference&lt;/td&gt;
 &lt;td&gt;True&lt;/td&gt;
 &lt;td&gt;Vera Rubin, LPX, HBM4, and SRAM/LPU become one system&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;China is using Ascend, optical mesh, and memory hierarchy workarounds&lt;/td&gt;
 &lt;td&gt;Partly true&lt;/td&gt;
 &lt;td&gt;Directionally plausible, but CloudMatrix performance and reliability need independent evidence&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;US export controls limit China’s NVIDIA data-center compute access&lt;/td&gt;
 &lt;td&gt;True&lt;/td&gt;
 &lt;td&gt;China cannot count on steady access to top-end GPUs and HBM&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM is still a key control point&lt;/td&gt;
 &lt;td&gt;True&lt;/td&gt;
 &lt;td&gt;BIS has explicitly treated HBM as important for AI training and inference at scale&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Korean opportunity equals China optical modules&lt;/td&gt;
 &lt;td&gt;Too aggressive&lt;/td&gt;
 &lt;td&gt;The cleaner Korean landing is HBM, power, packaging, and Samsung foundry optionality&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="2-why-the-stacks-are-diverging"&gt;2. Why The Stacks Are Diverging
&lt;/h2&gt;&lt;p&gt;Agentic AI increases token volume. Coding agents, research agents, voice agents, and enterprise workflow agents do not just answer once. They read long context, call tools, interpret tool output, and generate new tokens repeatedly. That makes prefill, decode, KV cache, storage, network, and power all matter.&lt;/p&gt;
&lt;p&gt;The bottleneck differs by country.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Area&lt;/th&gt;
 &lt;th&gt;United States&lt;/th&gt;
 &lt;th&gt;China&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Binding constraint&lt;/td&gt;
 &lt;td&gt;Power, rack density, transformers, HBM4, advanced packaging&lt;/td&gt;
 &lt;td&gt;Leading-edge logic, HBM access, export controls, software optimization&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Architecture direction&lt;/td&gt;
 &lt;td&gt;Vera Rubin, LPX/LPU, HBM4, high-power AI racks&lt;/td&gt;
 &lt;td&gt;Huawei Ascend, optical mesh, parallel scaling, domestic memory hierarchy&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Strength&lt;/td&gt;
 &lt;td&gt;Access to the best GPUs, HBM, and packaging ecosystem&lt;/td&gt;
 &lt;td&gt;Faster power buildout, state-led infrastructure, domestic cloud demand&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Weakness&lt;/td&gt;
 &lt;td&gt;Time-to-power, grid interconnection, transformer lead times&lt;/td&gt;
 &lt;td&gt;No EUV-based leading logic, HBM limits, closed ecosystem&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Korean equity landing&lt;/td&gt;
 &lt;td&gt;HBM, power equipment, HBM tools, foundry optionality&lt;/td&gt;
 &lt;td&gt;Limited unless direct suppliers are verified&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The IEA expects data-center electricity demand to reach about 945 TWh by 2030. ([IEA][1]) Power is no longer background infrastructure. It is becoming an upper bound on AI capacity.&lt;/p&gt;
&lt;p&gt;BloombergNEF data cited by Energy Connects says China may add more than 3.4 TW of generation capacity over five years, almost six times the US. The same article says data centers could account for 38% of US power demand growth from 2024 to 2030, versus 6% in China. ([Energy Connects][2]) This is why the US is forced to care about token throughput per MW, while China can lean more heavily on parallel infrastructure buildout.&lt;/p&gt;
&lt;h2 id="3-the-us-stack-hbm-plus-sramlpu"&gt;3. The US Stack: HBM Plus SRAM/LPU
&lt;/h2&gt;&lt;p&gt;The key US signal is NVIDIA’s Vera Rubin and LPX architecture. NVIDIA describes LPX as an inference accelerator for Vera Rubin. It combines Rubin GPUs using HBM with Groq 3 LPX using SRAM-based LPUs. ([NVIDIA LPX][3])&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Metric&lt;/th&gt;
 &lt;th style="text-align: right"&gt;NVIDIA LPX disclosure&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Agentic token load&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Up to 15x more tokens&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Vera Rubin NVL72 + LPX&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Up to 35x higher throughput per MW&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SRAM per LPU accelerator&lt;/td&gt;
 &lt;td style="text-align: right"&gt;500MB&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SRAM bandwidth per LPU accelerator&lt;/td&gt;
 &lt;td style="text-align: right"&gt;150TB/s&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Scale-up bandwidth per LPU accelerator&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2.5TB/s&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;LPX rack&lt;/td&gt;
 &lt;td style="text-align: right"&gt;256 LPU chips&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SRAM per LPX rack&lt;/td&gt;
 &lt;td style="text-align: right"&gt;128GB&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DDR5 per LPX rack&lt;/td&gt;
 &lt;td style="text-align: right"&gt;12TB&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SRAM bandwidth per LPX rack&lt;/td&gt;
 &lt;td style="text-align: right"&gt;40PB/s&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This is not an HBM replacement story. HBM remains central to the Rubin GPU. SRAM/LPU complements HBM by handling latency-sensitive decode work closer to compute.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Inference stage&lt;/th&gt;
 &lt;th&gt;Simple description&lt;/th&gt;
 &lt;th&gt;Main bottleneck&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Prefill&lt;/td&gt;
 &lt;td&gt;Read the prompt and documents first&lt;/td&gt;
 &lt;td&gt;GPU compute, HBM bandwidth&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Decode&lt;/td&gt;
 &lt;td&gt;Generate answers token by token&lt;/td&gt;
 &lt;td&gt;Latency, SRAM, LPU&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KV cache&lt;/td&gt;
 &lt;td&gt;Reuse previous context&lt;/td&gt;
 &lt;td&gt;HBM, DDR, SSD, cache hierarchy&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Serving&lt;/td&gt;
 &lt;td&gt;Handle many users reliably&lt;/td&gt;
 &lt;td&gt;Rack network, scheduler, power&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This creates an underappreciated Samsung Electronics angle. Samsung is not only an HBM catch-up candidate. At GTC 2026, Samsung highlighted HBM4 mass production, HBM4E, SOCAMM2 mass production, PCIe 6.0 SSDs, and broader AI computing solutions. ([Samsung Newsroom][4]) If inference moves toward HBM plus SRAM/LPU plus storage hierarchy, Samsung’s combined memory, foundry, packaging, and storage position becomes more relevant.&lt;/p&gt;
&lt;h2 id="4-china-optical-mesh-is-a-competitive-signal-not-a-clean-korean-equity-trade"&gt;4. China: Optical Mesh Is A Competitive Signal, Not A Clean Korean Equity Trade
&lt;/h2&gt;&lt;p&gt;China’s workaround is logical. If access to the latest GPUs and HBM is constrained, the system must rely on more chips, better interconnect, optical mesh, and domestic software optimization. But that is not the same as a direct Korean revenue bridge.&lt;/p&gt;
&lt;p&gt;China’s AI supply chain is strategically local. US export controls also push the system further inward. CloudMatrix-style performance, failure rates, and total cost of ownership still need stronger independent evidence. Therefore, China optical mesh is an important competitive variable, but it is not the best first Korean equity landing.&lt;/p&gt;
&lt;h2 id="5-korean-listed-equity-map"&gt;5. Korean Listed Equity Map
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Layer&lt;/th&gt;
 &lt;th&gt;Bottleneck&lt;/th&gt;
 &lt;th&gt;Korean names&lt;/th&gt;
 &lt;th&gt;View&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM4/HBM4E&lt;/td&gt;
 &lt;td&gt;Vera Rubin and AI server memory&lt;/td&gt;
 &lt;td&gt;SK hynix, Samsung Electronics&lt;/td&gt;
 &lt;td&gt;Structural beneficiaries. SK hynix leads, Samsung is the catch-up option&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SRAM/LPU foundry&lt;/td&gt;
 &lt;td&gt;Low-latency decode accelerators&lt;/td&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td&gt;Still not fully visible in revenue, but the option is underappreciated&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AI storage and KV cache&lt;/td&gt;
 &lt;td&gt;Agent memory, eSSD, PCIe 6.0&lt;/td&gt;
 &lt;td&gt;Samsung Electronics, FADU&lt;/td&gt;
 &lt;td&gt;Extends the memory hierarchy below HBM&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM packaging tools&lt;/td&gt;
 &lt;td&gt;TC bonders, advanced packaging&lt;/td&gt;
 &lt;td&gt;Hanmi Semiconductor&lt;/td&gt;
 &lt;td&gt;Real bottleneck, but customer concentration and valuation matter&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Power equipment&lt;/td&gt;
 &lt;td&gt;Transformers, switchgear, grid connection&lt;/td&gt;
 &lt;td&gt;HD Hyundai Electric, LS ELECTRIC, Hyosung Heavy Industries&lt;/td&gt;
 &lt;td&gt;Direct US AI data-center power exposure&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;China optical mesh&lt;/td&gt;
 &lt;td&gt;Optical modules for Chinese clusters&lt;/td&gt;
 &lt;td&gt;Limited direct Korean exposure&lt;/td&gt;
 &lt;td&gt;Avoid unless supplier evidence appears&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="6-power-equipment-is-the-front-end-bottleneck"&gt;6. Power Equipment Is The Front-End Bottleneck
&lt;/h2&gt;&lt;p&gt;In the US, power is a binding constraint. Data centers need grid connection, transformers, switchgear, and high-voltage equipment before GPUs can be monetized.&lt;/p&gt;
&lt;p&gt;HD Hyundai Electric is one of the clearest Korean examples. It raised its 2026 order target by 23%, from $4.222 billion to $5.185 billion, citing North American power infrastructure, 765kV transformers, North American data centers, and onshore generator demand for data centers. ([Seoul Economic Daily][5])&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Factor&lt;/th&gt;
 &lt;th&gt;Investment read&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;P&lt;/td&gt;
 &lt;td&gt;Power-equipment pricing, high-voltage transformer margins, North America premium&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q&lt;/td&gt;
 &lt;td&gt;Data centers, transmission projects, 765kV demand, replacement cycle&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;C&lt;/td&gt;
 &lt;td&gt;Raw materials, capacity, delivery timing, quality costs&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;What to watch&lt;/td&gt;
 &lt;td&gt;Book-to-bill, North America orders, margins, 2027 capacity expansion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Failure mode&lt;/td&gt;
 &lt;td&gt;AI capex delay, grid permitting delay, local sourcing pressure, margin compression&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The company quality and the entry price should still be separated. The theme is visible now. Better setups come when AI capex fears pressure the stock but backlog and pricing remain intact.&lt;/p&gt;
&lt;h2 id="7-hbm4hbm4e-more-tied-to-the-us-stack-than-china"&gt;7. HBM4/HBM4E: More Tied To The US Stack Than China
&lt;/h2&gt;&lt;p&gt;BIS has treated HBM as important for AI training and inference at scale. ([BIS][6]) If China’s access is restricted, allied AI infrastructure becomes more dependent on Korean HBM.&lt;/p&gt;
&lt;p&gt;SK hynix is the clearest leader. Yonhap reported that SK hynix won more than two-thirds of NVIDIA’s HBM orders for a next-generation AI platform. ([Yonhap][7]) But the investment issue is crowding. The market already recognizes SK hynix as the HBM winner.&lt;/p&gt;
&lt;p&gt;Samsung is different. The market has long treated it as an HBM laggard. If HBM4E, SOCAMM2, PCIe 6.0 SSD, foundry, and packaging are repriced as one AI memory-hierarchy platform, the discount can narrow.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Name&lt;/th&gt;
 &lt;th&gt;Strength&lt;/th&gt;
 &lt;th&gt;Risk&lt;/th&gt;
 &lt;th&gt;Current view&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;SK hynix&lt;/td&gt;
 &lt;td&gt;HBM4 leadership, NVIDIA allocation, pure HBM exposure&lt;/td&gt;
 &lt;td&gt;Crowding, high expectations, 2027 supply risk&lt;/td&gt;
 &lt;td&gt;Wait&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td&gt;HBM4E catch-up, foundry, storage, packaging option&lt;/td&gt;
 &lt;td&gt;HBM qualification delays, foundry losses, customer trust&lt;/td&gt;
 &lt;td&gt;Conditional Buy candidate&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="8-samsungs-sramlpu-foundry-option"&gt;8. Samsung’s SRAM/LPU Foundry Option
&lt;/h2&gt;&lt;p&gt;The non-consensus point is Samsung. If the inference stack moves from GPU-only to HBM plus SRAM/LPU plus cache storage, Samsung is no longer just a memory catch-up story. It is also a foundry and memory-hierarchy option.&lt;/p&gt;
&lt;p&gt;The thesis strengthens if at least two of the following become visible:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Condition&lt;/th&gt;
 &lt;th&gt;Evidence to watch&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM4/HBM4E customer qualification and shipment&lt;/td&gt;
 &lt;td&gt;Earnings, guidance, HBM mix&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;LPU or AI ASIC foundry revenue or utilization improvement&lt;/td&gt;
 &lt;td&gt;Foundry loss reduction, HPC customers, AI accelerator production&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AI storage and memory-hierarchy revenue&lt;/td&gt;
 &lt;td&gt;SOCAMM2, PCIe 6.0 SSD, eSSD, KV-cache demand&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;It weakens if HBM4 qualification is delayed by two or more quarters, foundry losses do not narrow, or LPU/AI ASIC revenue does not accrue to Samsung in a material way.&lt;/p&gt;
&lt;h2 id="9-hanmi-semiconductor-real-bottleneck-price-discipline-needed"&gt;9. Hanmi Semiconductor: Real Bottleneck, Price Discipline Needed
&lt;/h2&gt;&lt;p&gt;HBM4 increases packaging complexity. Hanmi Semiconductor is a real TC bonder bottleneck. The Elec reported that Hanmi received its first HBM4 TC bonder order from SK hynix. ([The Elec][8])&lt;/p&gt;
&lt;p&gt;The thesis is sound, but equipment stocks have order-gap risk. Follow-on SK hynix orders, official Samsung or Micron orders, and margin durability matter more than the broad HBM narrative.&lt;/p&gt;
&lt;h2 id="10-practical-view"&gt;10. Practical View
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th style="text-align: right"&gt;Priority&lt;/th&gt;
 &lt;th&gt;Exposure&lt;/th&gt;
 &lt;th&gt;View&lt;/th&gt;
 &lt;th&gt;Why&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;1&lt;/td&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td&gt;Conditional Buy candidate&lt;/td&gt;
 &lt;td&gt;HBM4E, SRAM/LPU foundry, and AI storage options may not be fully priced&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;2&lt;/td&gt;
 &lt;td&gt;HD Hyundai Electric&lt;/td&gt;
 &lt;td&gt;Wait&lt;/td&gt;
 &lt;td&gt;Direct US data-center power exposure, but order momentum is visible&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;3&lt;/td&gt;
 &lt;td&gt;SK hynix&lt;/td&gt;
 &lt;td&gt;Wait&lt;/td&gt;
 &lt;td&gt;Best HBM4 business position, but crowded winner&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;4&lt;/td&gt;
 &lt;td&gt;Hanmi Semiconductor&lt;/td&gt;
 &lt;td&gt;Watchlist&lt;/td&gt;
 &lt;td&gt;HBM tool bottleneck, but needs repeat orders and customer diversification&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;5&lt;/td&gt;
 &lt;td&gt;China optical-mesh beneficiaries&lt;/td&gt;
 &lt;td&gt;Avoid&lt;/td&gt;
 &lt;td&gt;Direct Korean supplier evidence is weak&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="conclusion"&gt;Conclusion
&lt;/h2&gt;&lt;p&gt;The Korean opportunity is not “China optical mesh.” It is the US AI factory bottleneck: power, HBM, SRAM/LPU, advanced packaging, and storage hierarchy.&lt;/p&gt;
&lt;p&gt;SK hynix may have the strongest business position. HD Hyundai Electric may have the cleanest power-equipment purity. But the most asymmetric rerating candidate is Samsung Electronics if the market stops viewing it only as an HBM laggard and starts pricing HBM4E, SRAM/LPU foundry, and AI storage together.&lt;/p&gt;
&lt;h2 id="sources"&gt;Sources
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://www.ys-vc.com/blog/us-china-ai-inference-stack-divergence" target="_blank" rel="noopener"
 &gt;YS-VC, Power or Chips: Why US and Chinese AI inference stacks are diverging&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai" target="_blank" rel="noopener"
 &gt;IEA, Energy demand from AI&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.energyconnects.com/news/renewables/2026/february/china-ramps-up-energy-boom-flagged-by-musk-as-key-to-ai-race/" target="_blank" rel="noopener"
 &gt;Energy Connects, China ramps up energy boom flagged by Musk as key to AI race&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.nvidia.com/en-us/data-center/lpx/" target="_blank" rel="noopener"
 &gt;NVIDIA, NVIDIA LPX&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://news.samsung.com/global/samsung-unveils-hbm4e-showcasing-comprehensive-ai-solutions-nvidia-partnership-and-vision-at-nvidia-gtc-2026" target="_blank" rel="noopener"
 &gt;Samsung Newsroom, Samsung unveils HBM4E and AI solutions at NVIDIA GTC 2026&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.bis.gov/press-release/commerce-strengthens-export-controls-restrict-chinas-capability-produce-advanced-semiconductors-military" target="_blank" rel="noopener"
 &gt;BIS, Commerce strengthens export controls including HBM&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://en.yna.co.kr/view/AEN20260128002800320" target="_blank" rel="noopener"
 &gt;Yonhap, SK hynix reportedly wins over two-thirds of Nvidia HBM orders&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.thelec.net/news/articleView.html?idxno=11132" target="_blank" rel="noopener"
 &gt;The Elec, Hanmi Semiconductor receives first HBM4 TC bonder order&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://en.sedaily.com/finance/2026/07/06/hd-hyundai-electric-raises-2026-order-target-23-percent-on" target="_blank" rel="noopener"
 &gt;Seoul Economic Daily, HD Hyundai Electric raises 2026 order target&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;</description></item><item><title>Samsung and SK Hynix 2028E Profit Valuation: Cheap-Looking Numbers and Cycle Tests</title><link>https://koreainvestinsights.com/post/samsung-hynix-2028e-profit-valuation-cycle-scenarios-2026-07-09/</link><pubDate>Thu, 09 Jul 2026 14:20:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/samsung-hynix-2028e-profit-valuation-cycle-scenarios-2026-07-09/</guid><description>
 &lt;blockquote&gt;
 &lt;p&gt;Related context: this is a follow-up to &lt;a class="link" href="https://koreainvestinsights.com/post/ai-token-value-memory-value-added-2026-07-09/" &gt;AI token value and memory value added&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/samsung-record-quarter-profit-signals-noise-2026-07-08/" &gt;Samsung 2Q26 signals and noise&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/big-tech-july-earnings-call-memory-thesis-scenarios-2026-07-07/" &gt;late-July Big Tech earnings and the memory thesis&lt;/a&gt;, and &lt;a class="link" href="https://koreainvestinsights.com/post/micron-fy3q26-ai-memory-sca-fcf-hold-2026-06-25/" &gt;Micron FY3Q26 AI memory review&lt;/a&gt;. Relevant hubs are the &lt;a class="link" href="https://koreainvestinsights.com/page/exclusive-analysis-hub/" &gt;Exclusive Analysis Hub&lt;/a&gt; and the &lt;a class="link" href="https://koreainvestinsights.com/page/korea-semiconductor-hbm-kospi-hub/" &gt;AI HBM Hub&lt;/a&gt;.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;p&gt;Samsung Electronics and SK Hynix look very cheap on 2028E net income. If we use MarketScreener aggregate estimates, 2028E net income is about KRW 439.8tn for Samsung and KRW 346.6tn for SK Hynix. Applying a 9 to 12x P/E gives fair market caps of KRW 3,958tn to 5,277tn for Samsung and KRW 3,119tn to 4,159tn for SK Hynix.&lt;/p&gt;
&lt;p&gt;Against current market caps, that implies roughly +120 to +193% for Samsung and +104 to +172% for SK Hynix. The market is valuing them at only about 4.1x and 4.4x 2028E P/E, respectively.&lt;/p&gt;
&lt;p&gt;But the bottleneck is not the multiple. It is the 2028E profit number. If 2028 earnings are normalized, both stocks are cheap. If 2028 is the cycle peak, a 4 to 5x P/E may not be cheap.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Samsung Electronics&lt;/th&gt;
 &lt;th style="text-align: right"&gt;SK Hynix&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;MarketScreener 2028E net income&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 439.8tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 346.6tn&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;9 to 12x fair market cap&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 3,958tn to 5,277tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 3,119tn to 4,159tn&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Current implied 2028E P/E&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 4.1x&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 4.4x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Scenario-weighted 2028E net income&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 272tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 208tn&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Scenario-weighted 9 to 12x fair market cap&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 2,444tn to 3,258tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 1,876tn to 2,502tn&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Scenario-weighted upside&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About +36 to +81%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About +23 to +64%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The stronger relative setup is Samsung. SK Hynix has higher-quality HBM earnings, but more of that quality is already embedded in the price. Samsung has HBM catch-up, commodity DRAM/NAND leverage and a lower implied 2028E P/E.&lt;/p&gt;
&lt;h2 id="1-current-coordinates"&gt;1. Current Coordinates
&lt;/h2&gt;&lt;p&gt;Samsung&amp;rsquo;s net income belongs economically to both common and preferred shares, so the right denominator is total equity value, not common-share market cap alone. Depending on the public screen, Samsung market cap appears around KRW 1,753tn, while a common-plus-preferred equity value framework is closer to KRW 1,800tn. SK Hynix is around KRW 1,530tn.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Samsung Electronics&lt;/th&gt;
 &lt;th style="text-align: right"&gt;SK Hynix&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Reference share price&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 274,000 to 274,500&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 2,151,000 to 2,160,000&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Analysis market cap&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 1,800tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 1,530tn&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026E OP reference&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 371tn to 381tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 280tn to 291tn&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2027E OP reference&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 500tn to 574tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 379tn to 432tn&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Samsung guided to 2Q26 revenue of KRW 171tn and operating profit of KRW 89.4tn. SK Hynix reported 1Q26 revenue of KRW 52.576tn, operating profit of KRW 37.610tn and net income of KRW 40.346tn, with operating margin of 72%.&lt;/p&gt;
&lt;p&gt;Using net income as roughly 75% of operating profit, current prices imply 2026E P/E near 6 to 7x and 2027E P/E near 4 to 5x. The market is not treating these companies as secular growth stocks. It is treating them as memory companies at or near peak earnings.&lt;/p&gt;
&lt;h2 id="2-the-point-estimate-valuation"&gt;2. The Point-Estimate Valuation
&lt;/h2&gt;&lt;p&gt;MarketScreener aggregate estimates put 2028E net income at KRW 439.781tn for Samsung and KRW 346.562tn for SK Hynix. The same dataset puts 2028E EBIT at KRW 543.635tn and KRW 423.866tn, respectively.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;KRW tn&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2028E net income&lt;/th&gt;
 &lt;th style="text-align: right"&gt;9x fair cap&lt;/th&gt;
 &lt;th style="text-align: right"&gt;10x fair cap&lt;/th&gt;
 &lt;th style="text-align: right"&gt;12x fair cap&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Current cap&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Upside&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td style="text-align: right"&gt;439.8&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3,958&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4,398&lt;/td&gt;
 &lt;td style="text-align: right"&gt;5,277&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 1,800&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+120% / +144% / +193%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td style="text-align: right"&gt;346.6&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3,119&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3,466&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4,159&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 1,530&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+104% / +127% / +172%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This simple calculation says both stocks are cheap. It also says Samsung is slightly more asymmetric because its implied 2028E P/E is lower.&lt;/p&gt;
&lt;p&gt;The danger is that 2028E is still far away. Memory cycles have repeatedly produced large peak profits that later collapsed. Therefore, the real question is whether 2028E is a normalized profit base or a peak.&lt;/p&gt;
&lt;h2 id="3-how-to-estimate-2028e"&gt;3. How to Estimate 2028E
&lt;/h2&gt;&lt;p&gt;Extending 2027E mechanically is not enough. A top-down “AI capex keeps rising, so memory profit rises” approach is also too thin. Memory profits are often set by supply, not demand. The 2028 supply base is already being shaped by 2026 to 2027 capex.&lt;/p&gt;
&lt;p&gt;The better process is:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Step&lt;/th&gt;
 &lt;th&gt;What to do&lt;/th&gt;
 &lt;th&gt;Why it matters&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;1&lt;/td&gt;
 &lt;td&gt;Split profits into contracted HBM, non-contracted HBM/server DRAM, and spot DRAM/NAND&lt;/td&gt;
 &lt;td&gt;Each bucket has a different earnings floor&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2&lt;/td&gt;
 &lt;td&gt;Track supply instead of forecasting it abstractly&lt;/td&gt;
 &lt;td&gt;2027 capex determines 2028 bit supply&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;3&lt;/td&gt;
 &lt;td&gt;Segment demand into AI servers, general servers and consumer devices&lt;/td&gt;
 &lt;td&gt;Price elasticity differs sharply&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;4&lt;/td&gt;
 &lt;td&gt;Force memory-cycle base rates into the model&lt;/td&gt;
 &lt;td&gt;Peak-year earnings historically fall hard&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;5&lt;/td&gt;
 &lt;td&gt;Use scenario distributions and reverse implied earnings&lt;/td&gt;
 &lt;td&gt;Current price already embeds a view&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The single most useful future disclosure would be the 2028 HBM long-term agreement coverage ratio. If a large share of 2028 shipment is already contracted, the 2028 profit floor is much higher than in historical cycles.&lt;/p&gt;
&lt;p&gt;The main supply variables are 2027 capex guidance, equipment orders from ASML/AMAT/TEL, Hynix&amp;rsquo;s Yongin and Cheongju investments, Samsung&amp;rsquo;s HBM and packaging catch-up spending, CXMT and YMTC expansion, and whether HBM bottlenecks remain in stacking and packaging rather than wafers alone.&lt;/p&gt;
&lt;p&gt;Demand must also be segmented. AI server demand is relatively inelastic. General server demand is cyclical. PC and smartphone memory demand is price-sensitive and can already show demand destruction when memory prices rise too quickly.&lt;/p&gt;
&lt;p&gt;Historical base rates are harsh.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Cycle&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Operating profit change&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung consolidated 2018 to 2019&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-53%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung DS 2018 to 2019&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-69%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix 2018 to 2019&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-87%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2023 memory downcycle&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Loss-making episodes&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This cycle deserves an upward adjustment because of HBM contracts, oligopoly structure, wafer cannibalization by HBM and low AI-server price elasticity. But it also deserves caution because of Chinese supply, consumer demand destruction, policy-driven capacity expansion and the statistical extremity of 70%+ operating margins.&lt;/p&gt;
&lt;h2 id="4-scenario-valuation"&gt;4. Scenario Valuation
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Scenario&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Probability&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Samsung 2028E NI&lt;/th&gt;
 &lt;th style="text-align: right"&gt;9x&lt;/th&gt;
 &lt;th style="text-align: right"&gt;12x&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Upside&lt;/th&gt;
 &lt;th style="text-align: right"&gt;SK Hynix 2028E NI&lt;/th&gt;
 &lt;th style="text-align: right"&gt;9x&lt;/th&gt;
 &lt;th style="text-align: right"&gt;12x&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Upside&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;S1 shortage persists&lt;/td&gt;
 &lt;td style="text-align: right"&gt;25%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 420tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3,780&lt;/td&gt;
 &lt;td style="text-align: right"&gt;5,040&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+110 to +180%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 315tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2,835&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3,780&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+85 to +147%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;S2 soft landing&lt;/td&gt;
 &lt;td style="text-align: right"&gt;45%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 280tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2,520&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3,360&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+40 to +87%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 215tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1,935&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2,580&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+26 to +69%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;S3 hard landing&lt;/td&gt;
 &lt;td style="text-align: right"&gt;30%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 135tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1,215&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1,620&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-32 to -10%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 110tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;990&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1,320&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-35 to -14%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Probability-weighted&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 272tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2,444&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3,258&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+36 to +81%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 208tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1,876&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2,502&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+23 to +64%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;MarketScreener&amp;rsquo;s estimate sits closer to S1. The probability-weighted model blends S2 and S3 risks. These two approaches do not conflict. One shows how cheap the stocks are if consensus is right. The other shows how much risk discount should be applied to that consensus.&lt;/p&gt;
&lt;h2 id="5-what-the-current-price-already-embeds"&gt;5. What the Current Price Already Embeds
&lt;/h2&gt;&lt;p&gt;Reverse the current market caps by a 9 to 12x P/E and we get the 2028E net income implied by today&amp;rsquo;s prices.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Current cap&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Implied 2028E NI at 9x&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Implied 2028E NI at 12x&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 1,800tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 200tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 150tn&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 1,530tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 170tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 127tn&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Against rough 2027E net income estimates, current prices already imply a meaningful drop.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Rough 2027E NI&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Implied 2028E NI&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Implied decline&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 400tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 150tn to 200tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-50 to -63%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 310tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 127tn to 170tn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-45 to -59%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This is the key reframing. The current bet is not necessarily that 2028 earnings will be spectacular. It may simply be that the decline from 2027 will be shallower than the 45 to 63% decline already embedded in prices.&lt;/p&gt;
&lt;h2 id="6-samsung-versus-sk-hynix"&gt;6. Samsung Versus SK Hynix
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Company&lt;/th&gt;
 &lt;th&gt;Core thesis&lt;/th&gt;
 &lt;th&gt;Risk&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td&gt;Lower implied 2028E P/E, HBM catch-up, commodity DRAM/NAND leverage&lt;/td&gt;
 &lt;td&gt;HBM4 delay, foundry/system LSI losses, weaker reported-to-core profit conversion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td&gt;Highest-quality HBM profit pool, customer reference, strong margin&lt;/td&gt;
 &lt;td&gt;More optimism embedded, ADR/event flow, need to prove 2028 LTA coverage&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Samsung is not the HBM leader, but that is partly the point. Expectations are lower. If HBM4 qualification, yield and allocation improve, the stock gets both earnings revision and multiple repair. Samsung also has more leverage to conventional DRAM and NAND pricing.&lt;/p&gt;
&lt;p&gt;SK Hynix has the cleaner HBM exposure and better current profit quality. But the market already gives it more credit for structural HBM change. The stock needs clearer evidence that 2028 profit is contracted and not just a peak-cycle extrapolation.&lt;/p&gt;
&lt;h2 id="7-practical-stance"&gt;7. Practical Stance
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Stock&lt;/th&gt;
 &lt;th&gt;Stance&lt;/th&gt;
 &lt;th&gt;Reason&lt;/th&gt;
 &lt;th&gt;What to confirm&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td&gt;Hold-biased, pullback candidate&lt;/td&gt;
 &lt;td&gt;More asymmetric 2028E valuation and catch-up optionality&lt;/td&gt;
 &lt;td&gt;July 30 segment data, HBM4 progress, NAND/DRAM pricing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td&gt;Watchlist, no event-chasing&lt;/td&gt;
 &lt;td&gt;Best HBM quality, but more expectation embedded&lt;/td&gt;
 &lt;td&gt;ADR absorption, 2Q call LTA coverage, HBM4 premium&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;For Samsung, selling aggressively here requires assigning a very high probability to a 2028 hard landing. The data does not yet justify that. For SK Hynix, the company quality is clear, but ADR, new-share absorption and 2028 contract visibility matter before chasing.&lt;/p&gt;
&lt;h2 id="8-invalidation"&gt;8. Invalidation
&lt;/h2&gt;&lt;p&gt;The thesis weakens if two or more of the following happen.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Invalidation signal&lt;/th&gt;
 &lt;th&gt;Meaning&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;2027 capex across the three major memory firms rises more than 50% YoY&lt;/td&gt;
 &lt;td&gt;Supply discipline breaks&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CXMT qualifies high-end server DDR5 with major customers&lt;/td&gt;
 &lt;td&gt;Commodity DRAM ceiling falls&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Chinese suppliers make visible HBM progress&lt;/td&gt;
 &lt;td&gt;Premium durability weakens&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Big Tech capex guidance slows sharply&lt;/td&gt;
 &lt;td&gt;AI server demand downside&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;No disclosure or evidence of 2028 HBM LTA coverage&lt;/td&gt;
 &lt;td&gt;Peak-profit discount remains&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;YMTC NAND ramp accelerates&lt;/td&gt;
 &lt;td&gt;NAND pricing risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="final-view"&gt;Final View
&lt;/h2&gt;&lt;p&gt;Samsung and SK Hynix both look cheap on 2028E earnings. But cheap-looking is not the same as cheap.&lt;/p&gt;
&lt;p&gt;Using MarketScreener point estimates, both have roughly double-like potential. Using a supply, contract and base-rate scenario model, the upside narrows to +36 to +81% for Samsung and +23 to +64% for SK Hynix. That is still attractive, but the margin of safety is smaller.&lt;/p&gt;
&lt;p&gt;The key is 2028E. It cannot be built from demand stories alone. It depends on HBM LTA coverage, 2027 capex discipline, DRAM/NAND pricing, Chinese supply and Big Tech capex.&lt;/p&gt;
&lt;p&gt;The relative preference is Samsung. SK Hynix has better HBM earnings, but requires more belief. Samsung has catch-up optionality, broader DRAM/NAND leverage and a lower implied 2028E P/E. For both names, the next checks are late-July Big Tech capex, Samsung&amp;rsquo;s July 30 detailed results, SK Hynix ADR absorption and the 2Q call.&lt;/p&gt;
&lt;h2 id="evidence-and-limits"&gt;Evidence and Limits
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Classification&lt;/th&gt;
 &lt;th&gt;Content&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Fact&lt;/td&gt;
 &lt;td&gt;Samsung 2Q26 preliminary revenue KRW 171tn and OP KRW 89.4tn. SK Hynix 1Q26 revenue KRW 52.576tn, OP KRW 37.610tn and net income KRW 40.346tn. MarketScreener 2028E net income of KRW 439.781tn for Samsung and KRW 346.562tn for SK Hynix&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Inference&lt;/td&gt;
 &lt;td&gt;Net income approximated at 75% of OP. Current prices imply 2028E net income of KRW 150-200tn for Samsung and KRW 127-170tn for SK Hynix&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Speculation&lt;/td&gt;
 &lt;td&gt;Scenario probabilities and 2028E net income assumptions: S1 25%, S2 45%, S3 30%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Blocked&lt;/td&gt;
 &lt;td&gt;2028 HBM LTA coverage, customer allocations, actual contract prices and detailed analyst-by-analyst 2028E assumptions are not fully available from public data&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h3 id="sources"&gt;Sources
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://news.samsung.com/global/samsung-electronics-announces-earnings-guidance-for-second-quarter-2026" target="_blank" rel="noopener"
 &gt;Samsung Global Newsroom, 2Q26 earnings guidance&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.prnewswire.com/news-releases/sk-hynix-announces-1q26-financial-results-302750959.html" target="_blank" rel="noopener"
 &gt;SK hynix 1Q26 financial results&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.marketscreener.com/quote/stock/SAMSUNG-ELECTRONICS-CO-LT-6494906/finances/" target="_blank" rel="noopener"
 &gt;MarketScreener, Samsung Electronics financial forecasts&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.reuters.com/world/asia-pacific/samsung-estimates-19-fold-rise-q2-operating-profit-beating-expectations-2026-07-06/" target="_blank" rel="noopener"
 &gt;Reuters, Samsung 2Q26 profit reaction&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://en.yna.co.kr/view/AEN20260701005500320" target="_blank" rel="noopener"
 &gt;Yonhap, SK hynix F-1/A and HBM share reference&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;</description></item><item><title>AI Token Value Today and Tomorrow: Value Added for Memory Companies</title><link>https://koreainvestinsights.com/post/ai-token-value-memory-value-added-2026-07-09/</link><pubDate>Thu, 09 Jul 2026 09:00:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/ai-token-value-memory-value-added-2026-07-09/</guid><description>
 &lt;blockquote&gt;
 &lt;p&gt;Related context: this article follows our work on &lt;a class="link" href="https://koreainvestinsights.com/post/big-tech-july-earnings-call-memory-thesis-scenarios-2026-07-07/" &gt;late July big tech earnings and the memory thesis&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/micron-fy3q26-ai-memory-sca-fcf-hold-2026-06-25/" &gt;Micron&amp;rsquo;s FY3Q26 AI memory earnings&lt;/a&gt;, and &lt;a class="link" href="https://koreainvestinsights.com/post/h1-2026-ai-infra-bottleneck-korea-narrow-market-postmortem-2026-06-30/" &gt;the 1H26 AI infrastructure bottleneck review&lt;/a&gt;. Related hubs are the &lt;a class="link" href="https://koreainvestinsights.com/page/exclusive-analysis-hub/" &gt;Exclusive Analysis Hub&lt;/a&gt; and the &lt;a class="link" href="https://koreainvestinsights.com/page/korea-semiconductor-hbm-kospi-hub/" &gt;AI HBM Hub&lt;/a&gt;. The question is not simply whether AI demand is strong. The question is who keeps the economic value created by AI tokens.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;p&gt;If we decompose one current dollar of AI token revenue, the largest share of value added still sits at the model and application layer. The equity market, however, is being driven by data center investment that is being pulled forward before final application revenue has fully arrived. That is why today’s AI value chain looks less like a normal pyramid and more like an inverted one.&lt;/p&gt;
&lt;p&gt;The current value-added split of one dollar of AI token revenue is roughly as follows.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Layer&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Current value-added estimate&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;AI model company&lt;/td&gt;
 &lt;td style="text-align: right"&gt;45 to 55 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cloud and data center&lt;/td&gt;
 &lt;td style="text-align: right"&gt;10 to 16 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;GPU supplier&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 13 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM, server DRAM, SSD&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 2.3 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Other infrastructure software and storage&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3 to 5 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Foundry&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 1 cent&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Power&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 1.3 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The market capitalization and capex cycle cannot be explained by this table alone. Current AI service revenue is estimated around 60 to 100 billion dollars per year, while the leading GPU supplier’s data center revenue run rate has reached around 300 billion dollars, and the big four cloud companies’ 2026 capex is estimated near 725 billion dollars. The industry is building future demand before current revenue catches up.&lt;/p&gt;
&lt;p&gt;Memory is a real bottleneck in this structure. But if memory captures only about 2 to 5 cents of value added per token dollar, the central question for memory stocks is not “is the P/E low?” The real question is whether the industry can sustain mid-cycle net income of roughly 100 to 140 trillion won.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Question&lt;/th&gt;
 &lt;th&gt;View&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Where does AI token value accrue?&lt;/td&gt;
 &lt;td&gt;Near term: models and cloud. Long term: workflow applications and physical bottlenecks&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Is memory a structural beneficiary?&lt;/td&gt;
 &lt;td&gt;Yes, but its value-added share is small and the stocks already price in a strong mid-cycle profit base&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;What should investors monitor?&lt;/td&gt;
 &lt;td&gt;HBM4 contract pricing, customer allocation, blended memory ASP, cloud backlog, and application revenue conversion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="1-the-real-question-who-keeps-the-dollar"&gt;1. The Real Question: Who Keeps the Dollar?
&lt;/h2&gt;&lt;p&gt;The most common mistake in AI investing is mixing up revenue attribution with value-added attribution. If a customer pays one dollar for an AI service, that dollar may first appear as revenue at a model company. But inside that dollar are cloud costs, GPU depreciation, memory, foundry capacity, power, networking, servers, cooling, labor, software, taxes, and financing costs.&lt;/p&gt;
&lt;p&gt;So “who books the revenue?” is not the same question as “who keeps the surplus?”&lt;/p&gt;
&lt;p&gt;This article uses five rules.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Rule&lt;/th&gt;
 &lt;th&gt;Meaning&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Revenue attribution&lt;/td&gt;
 &lt;td&gt;The first seller that invoices the customer&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Value-added attribution&lt;/td&gt;
 &lt;td&gt;The economic value left after removing double counting&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Token dollar&lt;/td&gt;
 &lt;td&gt;A dollar of API or model usage revenue&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Application dollar&lt;/td&gt;
 &lt;td&gt;A dollar of final workflow or agent software spend&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Depreciation basis&lt;/td&gt;
 &lt;td&gt;Data center capex is annualized over a five-year asset life&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The capex treatment matters. AI data center spending does not disappear as an immediate cost. It is capitalized into GPUs, servers, power gear, buildings, cooling, and networks, then depreciated over several years. Therefore, the right cost in a token-dollar model is annual depreciation, not total capex.&lt;/p&gt;
&lt;p&gt;Time also matters. Inference cost in 2025 is not the same as inference cost in 2028. Cost per unit of performance is falling quickly.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Period&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Inference cost as share of revenue&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Model gross margin&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;2025&lt;/td&gt;
 &lt;td style="text-align: right"&gt;50 to 67%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;33 to 50%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026&lt;/td&gt;
 &lt;td style="text-align: right"&gt;40 to 55%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;45 to 60%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2027 to 2028&lt;/td&gt;
 &lt;td style="text-align: right"&gt;30 to 40%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;60 to 70%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This leads to an important point. Token prices can fall while model gross margins rise, if inference costs fall faster than prices. Much of that margin is then reinvested into training, compute reservations, and customer acquisition. The value does not simply stay on the income statement. It flows back into GPUs, memory, power, and foundry capacity.&lt;/p&gt;
&lt;h2 id="2-the-current-value-waterfall"&gt;2. The Current Value Waterfall
&lt;/h2&gt;&lt;p&gt;The table below decomposes one current AI token dollar. Revenue attribution is the amount a layer effectively invoices or absorbs. Value added is the economic share after removing double counting.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Layer&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Revenue attribution&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Value-added estimate&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;AI model company&lt;/td&gt;
 &lt;td style="text-align: right"&gt;100 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;45 to 55 cents&lt;/td&gt;
 &lt;td&gt;Direct customer layer. Falling inference cost is the largest lever&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cloud and data center&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 50 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;10 to 16 cents&lt;/td&gt;
 &lt;td&gt;Operating layer that bundles GPUs, networks, power, and buildings&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;GPU supplier&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 18 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 13 cents&lt;/td&gt;
 &lt;td&gt;Main physical bottleneck today&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM, server DRAM, SSD&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 3 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 2.3 cents&lt;/td&gt;
 &lt;td&gt;Critical to system performance, but small as a token-dollar share&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Infrastructure software, storage, security, observability&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 5 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 3 cents&lt;/td&gt;
 &lt;td&gt;More important as operations become more complex&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Foundry&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 1.5 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 1 cent&lt;/td&gt;
 &lt;td&gt;Common bottleneck across AI chip designs&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Server OEM and ODM&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 3 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 1.4 cents&lt;/td&gt;
 &lt;td&gt;High throughput, limited margins&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Third-party networking&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 1.3 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 0.7 cents&lt;/td&gt;
 &lt;td&gt;Important at cluster scale, but supplier power varies&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Power&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 2.5 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 1.3 cents&lt;/td&gt;
 &lt;td&gt;Small cost line, large site bottleneck&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CPU and IP&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 0.5 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 0.3 cents&lt;/td&gt;
 &lt;td&gt;Control and support compute&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Construction, components, materials, residual&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 9 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Distributed&lt;/td&gt;
 &lt;td&gt;Lower layers of the data center stack&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Two observations matter.&lt;/p&gt;
&lt;p&gt;First, power is not the largest cost item inside a token dollar. But it is one of the largest physical bottlenecks. Power matters through grid access, substations, transmission, permits, and long-term power contracts.&lt;/p&gt;
&lt;p&gt;Second, memory’s value-added share per token dollar is still small. This does not mean HBM is unimportant. HBM is indispensable. But the economic share retained by memory companies is far smaller than the share retained by GPUs or model/application layers. If the equity market ignores that difference, valuation errors appear.&lt;/p&gt;
&lt;h2 id="3-the-inverted-pyramid"&gt;3. The Inverted Pyramid
&lt;/h2&gt;&lt;p&gt;The current AI stack looks like this.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Estimated scale&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;AI model and application revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;60 to 100 billion dollars per year&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;GPU data center revenue run rate&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 300 billion dollars&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026 big four cloud capex&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 725 billion dollars&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AI revenue needed to normalize 700 billion dollars of capex&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 2.8 to 3.0 trillion dollars&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;If AI service revenue is currently 60 to 100 billion dollars, and memory revenue is 3 to 7 cents per token dollar, AI memory flow revenue should be only about 2 to 7 billion dollars per year. Yet the leading memory companies are producing revenue and profit numbers far above what current token revenue alone can explain.&lt;/p&gt;
&lt;p&gt;For example, SK Hynix reported 97.1 trillion won of revenue in 2025 and 52.6 trillion won of revenue in the first quarter of 2026. That scale is hard to explain only from current AI token revenue. Much of today’s memory revenue is therefore tied to capex being pulled forward for future AI demand.&lt;/p&gt;
&lt;p&gt;This is not automatically wrong. Data centers cannot be built instantly after demand appears. Power, land, substations, GPUs, servers, cooling, and networking must be secured years in advance. Pre-investment is necessary.&lt;/p&gt;
&lt;p&gt;The question is speed. If application and cloud revenue catch up, today’s capex is normal growth investment. If final demand lags, the correction begins upstream: memory and server components first, then GPUs, then cloud capex plans.&lt;/p&gt;
&lt;h2 id="4-three-convergence-paths"&gt;4. Three Convergence Paths
&lt;/h2&gt;&lt;p&gt;AI revenue is not determined by token price alone.&lt;/p&gt;
&lt;p&gt;AI revenue = token volume × token price&lt;/p&gt;
&lt;p&gt;Token price per unit of performance is falling quickly. It may decline by roughly 10x per year through the early phase, then slow to 1.5 to 2x per year after 2027. But volume can explode at the same time. A simple question may use a few hundred tokens. A workflow, document process, coding task, or agent job can use tens of thousands, hundreds of thousands, or even more than a million tokens.&lt;/p&gt;
&lt;p&gt;That means price declines do not automatically mean revenue declines. If price falls by 90% but volume rises 100x, revenue still rises 10x.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Scenario&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Probability sense&lt;/th&gt;
 &lt;th&gt;Description&lt;/th&gt;
 &lt;th&gt;Market implication&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Demand convergence&lt;/td&gt;
 &lt;td style="text-align: right"&gt;45%&lt;/td&gt;
 &lt;td&gt;Application revenue scales quickly and 1.0 to 1.6 trillion dollars of capex normalizes&lt;/td&gt;
 &lt;td&gt;Limited upstream drawdown, bottleneck premium remains&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Digestion then reacceleration&lt;/td&gt;
 &lt;td style="text-align: right"&gt;40%&lt;/td&gt;
 &lt;td&gt;2027 to 2028 capex pauses by 20 to 30%, then workflow AI demand returns&lt;/td&gt;
 &lt;td&gt;Memory ASP could correct 40 to 60%, then recover&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Structural oversupply&lt;/td&gt;
 &lt;td style="text-align: right"&gt;15%&lt;/td&gt;
 &lt;td&gt;Usage elasticity disappoints and data center buildout proves excessive&lt;/td&gt;
 &lt;td&gt;Capex cuts of 50% or more and upstream recession&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;On-device AI changes the distribution, not the whole demand picture. More inference on phones, PCs, cars, robots, and industrial devices can reduce some cloud token revenue. But silicon demand does not disappear. LPDDR, mobile SoCs, edge NPUs, embedded storage, and local servers may all benefit.&lt;/p&gt;
&lt;h2 id="5-long-run-equilibrium-workflows-capture-the-dollar"&gt;5. Long-Run Equilibrium: Workflows Capture the Dollar
&lt;/h2&gt;&lt;p&gt;Over the long run, tokens are unlikely to remain scarce. As token prices fall and model performance gaps narrow, value moves from token production toward workflow ownership.&lt;/p&gt;
&lt;p&gt;On a pure API dollar, the long-run value split may look like this.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Layer&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Long-run value-added estimate&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Model and API company&lt;/td&gt;
 &lt;td style="text-align: right"&gt;45 to 55 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cloud and data center&lt;/td&gt;
 &lt;td style="text-align: right"&gt;12 to 15 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;GPU supplier&lt;/td&gt;
 &lt;td style="text-align: right"&gt;8 to 10 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Custom ASIC&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4 to 5 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Memory&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4 to 5 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Foundry&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3 to 4 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Power and data center infrastructure&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4 to 5 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CPU and IP&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 1.5 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Server OEM&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 1.5 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Other infrastructure software&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4 to 6 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Residual&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 5 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;But customers do not really buy tokens. They buy outcomes: customer support, coding productivity, claims processing, contract review, factory quality control, and logistics automation. On a final application dollar, the distribution changes.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Layer&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Long-run revenue attribution&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Long-run value added&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Workflow applications, agents, distribution&lt;/td&gt;
 &lt;td style="text-align: right"&gt;100 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;45 to 50 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Model and API&lt;/td&gt;
 &lt;td style="text-align: right"&gt;35 to 40 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;18 to 24 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cloud and data center&lt;/td&gt;
 &lt;td style="text-align: right"&gt;13 to 15 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4 to 6 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Total silicon&lt;/td&gt;
 &lt;td style="text-align: right"&gt;9 to 12 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;6 to 8 cents&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Power, land, cooling&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1.5 to 2 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 1 cent&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Server OEM and generic infrastructure&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 1 cent&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Low&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Other software&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3 to 5 cents&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Medium&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The long-run conclusion is simple: tokens commoditize, while workflow ownership and physical bottlenecks retain value.&lt;/p&gt;
&lt;h2 id="6-capture-tiers"&gt;6. Capture Tiers
&lt;/h2&gt;&lt;p&gt;The best AI businesses are not simply the ones with the most revenue. The best businesses have three traits.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Customers cannot switch easily.&lt;/li&gt;
&lt;li&gt;Cost declines become margin expansion.&lt;/li&gt;
&lt;li&gt;Supply cannot expand quickly when demand rises.&lt;/li&gt;
&lt;/ol&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Tier&lt;/th&gt;
 &lt;th&gt;Layer&lt;/th&gt;
 &lt;th&gt;Why it matters&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Tier 1&lt;/td&gt;
 &lt;td&gt;Foundry, workflow applications, integrated cloud platforms&lt;/td&gt;
 &lt;td&gt;Bottleneck supply, customer lock-in, full-stack control&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Tier 2&lt;/td&gt;
 &lt;td&gt;GPU, platform AI suppliers, HBM leaders, custom ASIC&lt;/td&gt;
 &lt;td&gt;Current bottlenecks and high growth, but with price and competition risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Tier 3&lt;/td&gt;
 &lt;td&gt;Server OEM, neocloud, commodity memory, regulated utilities&lt;/td&gt;
 &lt;td&gt;Volume grows, but rent capture is weaker&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Foundry belongs in Tier 1 because AI wafer demand exists whether the winning chip is a GPU or a custom ASIC. Workflow applications belong there because model costs can fall while end-user value remains tied to business outcomes.&lt;/p&gt;
&lt;p&gt;Integrated cloud platforms need case-by-case judgment. Pure GPU rental can become a capital-heavy utility. But a company that owns search, ads, productivity software, cloud, chips, models, and data centers can turn capex into margin.&lt;/p&gt;
&lt;h2 id="7-multiples-cheap-looking-is-not-the-same-as-cheap"&gt;7. Multiples: Cheap Looking Is Not the Same as Cheap
&lt;/h2&gt;&lt;p&gt;AI value-chain multiples should differ by layer.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Layer&lt;/th&gt;
 &lt;th&gt;Appropriate valuation lens&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Foundry&lt;/td&gt;
 &lt;td&gt;EV/S 8 to 11x, P/E 20 to 26x if growth persists&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;GPU supplier&lt;/td&gt;
 &lt;td&gt;Normalized EV/S 9 to 13x, scarcity 15 to 18x, P/E 18 to 24x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Integrated cloud&lt;/td&gt;
 &lt;td&gt;EV/S 5 to 8x, full-stack 8 to 10x, P/E 20 to 28x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Platform AI supplier&lt;/td&gt;
 &lt;td&gt;Hypergrowth EV/ARR 10 to 18x, deceleration 5 to 8x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Workflow AI software&lt;/td&gt;
 &lt;td&gt;Simple tools 2 to 4x, workflow owners 8 to 12x, regulated data owners 12 to 20x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM and memory&lt;/td&gt;
 &lt;td&gt;EV/S 3 to 5x, leaders 5 to 7x, mid-cycle P/E 8 to 14x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Storage&lt;/td&gt;
 &lt;td&gt;EV/S 2 to 3.5x, platform 4 to 7x, normalized P/E 8 to 14x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CPU and IP&lt;/td&gt;
 &lt;td&gt;CPU 12 to 18x, IP conditionally 30 to 50x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Server OEM&lt;/td&gt;
 &lt;td&gt;EV/S 0.5 to 1.5x, P/E 8 to 15x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Power and data center infrastructure&lt;/td&gt;
 &lt;td&gt;Contracted assets at EV/EBITDA 10 to 18x or project IRR&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Memory is the layer where investors should be most careful. A 5 to 7x P/E can signal cheapness. At a cycle peak, it can also simply mean that the earnings denominator is temporarily inflated.&lt;/p&gt;
&lt;p&gt;Useful valuation formulas include:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Formula&lt;/th&gt;
 &lt;th&gt;Use&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;EV/NOPAT = (1 - g/ROIC) / (WACC - g)&lt;/td&gt;
 &lt;td&gt;Normalized multiple with growth and capital efficiency&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Justified PBR = (sustainable ROE - g) / (COE - g)&lt;/td&gt;
 &lt;td&gt;Long-run capital-intensive industry PBR&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Mid-cycle EPS = (peak + trough + 2×normal) / 4&lt;/td&gt;
 &lt;td&gt;Cycle-aware earnings&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Pre-profit value = terminal NOPAT × terminal multiple / (1 + discount rate)^t&lt;/td&gt;
 &lt;td&gt;Early-stage or loss-making firms&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Capital costs also differ.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Layer&lt;/th&gt;
 &lt;th style="text-align: right"&gt;WACC or discount rate&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Hyperscalers&lt;/td&gt;
 &lt;td style="text-align: right"&gt;8.5 to 9.5%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;GPU and foundry&lt;/td&gt;
 &lt;td style="text-align: right"&gt;9 to 10.5%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Memory&lt;/td&gt;
 &lt;td style="text-align: right"&gt;10.5 to 12%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Vertical AI software&lt;/td&gt;
 &lt;td style="text-align: right"&gt;12 to 18%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Loss-making model layer&lt;/td&gt;
 &lt;td style="text-align: right"&gt;14 to 20%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="8-the-memory-question-can-mid-cycle-net-income-reach-100-to-140-trillion-won"&gt;8. The Memory Question: Can Mid-Cycle Net Income Reach 100 to 140 Trillion Won?
&lt;/h2&gt;&lt;p&gt;For memory companies, the key question is no longer whether HBM is good. The market already knows that. The key question is whether the current market cap requires a realistic or unrealistic mid-cycle profit base.&lt;/p&gt;
&lt;p&gt;Suppose a leading memory company is valued near 1,400 trillion won. To sustain that value after one or two peak years, mid-cycle net income must be very large.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Assumption&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Peak net income&lt;/td&gt;
 &lt;td style="text-align: right"&gt;240 trillion won&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Trough net income&lt;/td&gt;
 &lt;td style="text-align: right"&gt;50 trillion won&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Normalized net income&lt;/td&gt;
 &lt;td style="text-align: right"&gt;60 trillion won&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Mid-cycle net income&lt;/td&gt;
 &lt;td style="text-align: right"&gt;(240 + 50 + 2×60) / 4 = about 102 trillion won&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Applying an 8 to 14x mid-cycle multiple gives a fair-value range of about 820 to 1,430 trillion won. If the current market cap is already near the top of that range, the stock is not automatically expensive, but the margin of safety is thin.&lt;/p&gt;
&lt;p&gt;Further upside requires one or more of the following.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Requirement&lt;/th&gt;
 &lt;th&gt;Meaning&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Evidence that mid-cycle net income can exceed 140 trillion won&lt;/td&gt;
 &lt;td&gt;HBM premium and high-margin mix remain after the peak&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM4 custom premium is defended&lt;/td&gt;
 &lt;td&gt;Product customization protects pricing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;ASP holds after Samsung’s supply entry&lt;/td&gt;
 &lt;td&gt;More supply does not crush pricing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;General DRAM and NAND pricing also improves&lt;/td&gt;
 &lt;td&gt;The cycle broadens beyond HBM&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Long-term customer agreements and volume commitments&lt;/td&gt;
 &lt;td&gt;Downcycle earnings risk falls&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="9-samsung-sk-hynix-and-micron-are-not-the-same-trade"&gt;9. Samsung, SK Hynix, and Micron Are Not the Same Trade
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Company&lt;/th&gt;
 &lt;th&gt;Core logic&lt;/th&gt;
 &lt;th&gt;Risk&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td&gt;HBM share recovery and memory mix improvement. Expectations are lower than for the leader&lt;/td&gt;
 &lt;td&gt;HBM qualification, yield, foundry and system LSI losses&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td&gt;Highest HBM profit quality and customer reference&lt;/td&gt;
 &lt;td&gt;Much of the good news is already in the price&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron&lt;/td&gt;
 &lt;td&gt;Leading indicator for AI memory contracts, pricing, FCF, and capex&lt;/td&gt;
 &lt;td&gt;Volatility, short interest, and supply-cycle sensitivity&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Samsung has more catch-up optionality. SK Hynix has the highest-quality memory profit pool, but expectations are embedded. Micron is the canary because its pricing, long-term agreements, data center revenue, and capex guide influence Korean memory estimates directly.&lt;/p&gt;
&lt;h2 id="10-cloud-power-and-vertical-ai"&gt;10. Cloud, Power, and Vertical AI
&lt;/h2&gt;&lt;p&gt;Cloud companies split into two groups: integrated platforms and capital-heavy rental platforms. The most important numbers are backlog, revenue conversion, operating margin, internal silicon mix, free cash flow, and bond issuance.&lt;/p&gt;
&lt;p&gt;Power is a small token-dollar cost line but a large site bottleneck. Grid access, substations, transmission, PPAs, gas turbines, nuclear, SMRs, water, land, cooling, and permits matter more than the simple power bill.&lt;/p&gt;
&lt;p&gt;Vertical AI companies can capture the largest long-run share if they own the workflow. The winners will have regulated or physical domains, proprietary data, action authority, measurable ROI, model replaceability, a path to 70%+ gross margin, and compliance moats. Generic chat wrappers, basic retrieval integration, API resale, data-less automation, GPU brokerage, and server assembly should be treated with caution.&lt;/p&gt;
&lt;h2 id="11-what-to-monitor"&gt;11. What to Monitor
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Metric&lt;/th&gt;
 &lt;th&gt;Why it matters&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Combined cloud backlog&lt;/td&gt;
 &lt;td&gt;Whether capex is tied to future revenue&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Self-built capex ratio&lt;/td&gt;
 &lt;td&gt;How much demand is moving outside public cloud&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;On-device inference share&lt;/td&gt;
 &lt;td&gt;Cloud token revenue substitution risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM4 contract pricing&lt;/td&gt;
 &lt;td&gt;Durability of memory premium&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Customer allocation by HBM supplier&lt;/td&gt;
 &lt;td&gt;Supplier bargaining power&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Blended memory ASP&lt;/td&gt;
 &lt;td&gt;Whether the cycle is broadening&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;GPU lead times&lt;/td&gt;
 &lt;td&gt;Persistence of scarcity&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cloud operating margin&lt;/td&gt;
 &lt;td&gt;Whether capex turns into profit&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AI application revenue growth&lt;/td&gt;
 &lt;td&gt;Whether final demand catches infrastructure spending&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The thesis weakens if open-weight systems compress model pricing faster than expected, on-device AI reduces cloud demand more quickly, 2027 cloud capex slows sharply, HBM4 premiums compress, or AI application revenue fails to scale.&lt;/p&gt;
&lt;h2 id="final-view"&gt;Final View
&lt;/h2&gt;&lt;p&gt;The AI token economy can be summarized in one sentence:&lt;/p&gt;
&lt;p&gt;Tokens get cheaper, while value moves to workflow ownership and hard-to-expand physical bottlenecks.&lt;/p&gt;
&lt;p&gt;Memory is one of those bottlenecks. HBM, high-performance DRAM, and SSDs are essential to AI servers. But memory’s value-added share per token dollar remains limited. That is why memory stocks require more than the statement that HBM is strong. Current prices already demand a strong mid-cycle profit base.&lt;/p&gt;
&lt;p&gt;The four key questions are:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Does HBM4 preserve custom pricing?&lt;/li&gt;
&lt;li&gt;Do leading suppliers hold ASP and margins after Samsung’s entry?&lt;/li&gt;
&lt;li&gt;Do cloud backlog and AI application revenue catch up to capex?&lt;/li&gt;
&lt;li&gt;Can memory companies sustain mid-cycle net income of 100 to 140 trillion won?&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;If those four conditions are met, today’s large memory market caps can be justified. If not, low P/E ratios may prove to be a cycle-peak illusion rather than a margin of safety.&lt;/p&gt;</description></item><item><title>Samsung Electronics recorded the world's largest quarterly operating profit. Signals and noise poured in today.</title><link>https://koreainvestinsights.com/post/samsung-record-quarter-profit-signals-noise-2026-07-08/</link><pubDate>Wed, 08 Jul 2026 22:35:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/samsung-record-quarter-profit-signals-noise-2026-07-08/</guid><description>
 &lt;blockquote&gt;
 &lt;p&gt;Context
This is a follow-up to the Samsung 2Q26 preview, Micron FY3Q26 review, NVIDIA earnings analogue work, late-July Big Tech earnings-call scenario map, and Big Tech financing relay note. Related hubs are the &lt;a class="link" href="https://koreainvestinsights.com/page/exclusive-analysis-hub/" &gt;Exclusive Analysis Hub&lt;/a&gt; and the &lt;a class="link" href="https://koreainvestinsights.com/page/korea-semiconductor-hbm-kospi-hub/" &gt;AI HBM Hub&lt;/a&gt;.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;p&gt;Samsung Electronics presented provisional sales of 171 trillion won and operating profit of 89.4 trillion won for the second quarter of 2026. Based on Samsung Electronics&amp;rsquo; official newsroom, this is a very large number compared to the operating profit of 57.23 trillion won in the previous quarter and 4.68 trillion won in the same period last year. Looking at memory prices, AI server demand, hyperscaler investment, and HBM4 ramp-up together, the quality of Samsung Electronics&amp;rsquo; investment argument has actually become stronger.&lt;/p&gt;
&lt;p&gt;However, the stock price moved in the opposite direction. Based on local DB, Samsung Electronics closed at 296,000 won on July 7 and 277,500 won on July 8. On July 7, foreigners net sold about 1.82 trillion won, and on July 8, they sold about 874.1 billion won. Individuals made net purchases of approximately 2.32 trillion won and 584.7 billion won, respectively. Although good performance was achieved, foreign and programmatic supply and demand saw peak-out, interest rates, exchange rates, and overcrowding of positions before profit levels.&lt;/p&gt;
&lt;p&gt;The conclusion of this article is as follows.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Judgment&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics Investment Argument&lt;/td&gt;
 &lt;td&gt;enforce. DRAM/NAND prices, AI servers, and hyperscaler CapEx are all favorable.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Short term timing&lt;/td&gt;
 &lt;td&gt;damage. Despite good performance, there was selling pressure for two days in a row.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;supply and demand&lt;/td&gt;
 &lt;td&gt;danger. Foreign selling and private absorption are conflicting, and the Korean market is weak.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Sell ​​decision&lt;/td&gt;
 &lt;td&gt;no. The current data is closer to digesting expectations and supply and demand pressures than to damaging the thesis.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Additional purchase decision&lt;/td&gt;
 &lt;td&gt;Not yet. You must check ASML, TSMC, Samsung Electronics IR, foreign supply and demand, and exchange rates and interest rates.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;action&lt;/td&gt;
 &lt;td&gt;Strong holding, freeze on additional purchases, make a decision after confirming the next gateway.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;To sum it up in one line, the information released today is not “Samsung Electronics has worsened,” but is closer to “the market has already had too many good numbers, and now demands the next number and the quality of supply and demand.”&lt;/p&gt;
&lt;h2 id="1-why-should-we-separate-signal-and-noise"&gt;1. Why should we separate signal and noise?
&lt;/h2&gt;&lt;p&gt;Today&amp;rsquo;s data is not read in only one direction. Numbers are good. However, stock prices and supply and demand are bad. Memory price comments are strong. However, spot trading takes a breather on some items. Hyperscaler CapEx continues to grow. However, long-term interest rates in the United States are also high.&lt;/p&gt;
&lt;p&gt;On days like this, looking at only one side is wrong. “It’s the world’s largest quarterly operating profit, so you must buy it” is also hasty, and “the cycle is over because it fell behind despite good performance” is also hasty.&lt;/p&gt;
&lt;p&gt;Therefore, this article divides the data into four layers.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;floor&lt;/th&gt;
 &lt;th&gt;Content&lt;/th&gt;
 &lt;th&gt;Role in Investment Decisions&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Official Performance&lt;/td&gt;
 &lt;td&gt;Samsung Electronics 2Q26 provisional sales and operating profit&lt;/td&gt;
 &lt;td&gt;Check profit level&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;industrial signals&lt;/td&gt;
 &lt;td&gt;DRAM·NAND price, AI server, hyperscaler CapEx&lt;/td&gt;
 &lt;td&gt;Check the quality of the thesis&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;market noise&lt;/td&gt;
 &lt;td&gt;Sluggish spot trading, Morgan Stanley peak logic, ASML·TSMC high bar&lt;/td&gt;
 &lt;td&gt;Check short-term timing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Supply/Demand/Macro&lt;/td&gt;
 &lt;td&gt;Foreign selling, exchange rate, interest rate, Korean market breadth&lt;/td&gt;
 &lt;td&gt;Position Sizing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The key is to separate the profit direction and the stock price direction. Samsung Electronics’ main business profit direction has improved. However, the drivers of short-term stock prices were not profits, but supply and demand and the discount rate.&lt;/p&gt;
&lt;h2 id="2-officially-confirmed-values-sales-171-trillion-won-operating-profit-894-trillion-won"&gt;2. Officially confirmed values: Sales 171 trillion won, operating profit 89.4 trillion won
&lt;/h2&gt;&lt;p&gt;[Fact] Samsung Electronics announced its second quarter provisional results through its official newsroom on July 7, 2026. Consolidated sales are approximately 171 trillion won and operating profit is approximately 89.4 trillion won. According to Korean disclosure regulations, the company presented a median value rather than a range, and explained that the actual estimated range was KRW 170 trillion to KRW 172 trillion in sales and KRW 89.3 trillion to KRW 89.5 trillion in operating profit. Source: &lt;a class="link" href="https://news.samsung.com/global/samsung-electronics-announces-earnings-guidance-for-second-quarter-2026" target="_blank" rel="noopener"
 &gt;Samsung Electronics Official Newsroom&lt;/a&gt;&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2Q26 provisional figures&lt;/th&gt;
 &lt;th style="text-align: right"&gt;1Q26&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2Q25&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Compared to previous quarter&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Year-on-year comparison&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Sales&lt;/td&gt;
 &lt;td style="text-align: right"&gt;171.0 trillion won&lt;/td&gt;
 &lt;td style="text-align: right"&gt;133.87 trillion won&lt;/td&gt;
 &lt;td style="text-align: right"&gt;74.57 trillion won&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+27.7%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+129.3%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;operating profit&lt;/td&gt;
 &lt;td style="text-align: right"&gt;89.4 trillion won&lt;/td&gt;
 &lt;td style="text-align: right"&gt;57.23 trillion won&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4.68 trillion won&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+56.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+1,810.3%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This number is extremely unusual for a single quarter&amp;rsquo;s operating profit. The expression “world&amp;rsquo;s largest” may be subject to consideration depending on the comparison target and accounting standards, but it is clear that it is an overwhelmingly large number, at least compared to the normal quarterly profits of global mega-caps.&lt;/p&gt;
&lt;p&gt;What is important is not the operating profit itself of 89.4 trillion won, but the quality within it. The interpretation presented in the attached research is as follows.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;reported operating profit&lt;/td&gt;
 &lt;td&gt;89.4 trillion won&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Adjustment of one-time provisions such as DS performance bonus&lt;/td&gt;
 &lt;td&gt;The logic that profits from main business should be considered higher&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Core OP estimation&lt;/td&gt;
 &lt;td&gt;Possibility of over 100 trillion won excluding provisions&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM ASP&lt;/td&gt;
 &lt;td&gt;+47% compared to previous quarter&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NAND ASP&lt;/td&gt;
 &lt;td&gt;+66% compared to previous quarter&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM4&lt;/td&gt;
 &lt;td&gt;Maintain ramp-up expectations&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;[Inference] Just looking at the reported OP of KRW 89.4 trillion is strong, but if DRAM and NAND prices rose significantly at the same time and there was an effect of provisioning, the strength of the main business may be stronger than the reported number. This point is the most important signal of Samsung Electronics&amp;rsquo; thesis.&lt;/p&gt;
&lt;h2 id="3-stock-price-and-supply-and-demand-foreigners-sold-after-good-numbers"&gt;3. Stock price and supply and demand: Foreigners sold after good numbers
&lt;/h2&gt;&lt;p&gt;Despite good performance, the stock price fell. The prices and supply and demand based on local DB are as follows.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;date&lt;/th&gt;
 &lt;th style="text-align: right"&gt;closing price&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Daily fluctuations&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Foreign net purchase estimates&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Estimated institutional net purchase&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Individual net purchase estimate&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-07-03&lt;/td&gt;
 &lt;td style="text-align: right"&gt;309,500 won&lt;/td&gt;
 &lt;td style="text-align: right"&gt;&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 387.2 billion&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+1.3063 trillion won&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 918.4 billion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-07-06&lt;/td&gt;
 &lt;td style="text-align: right"&gt;318,000 won&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+2.75%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-640.9 billion won&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+5.8 billion won&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+596.4 billion won&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-07-07&lt;/td&gt;
 &lt;td style="text-align: right"&gt;296,000 won&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-6.92%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 1.8207 trillion&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 547.5 billion&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+2.3232 trillion won&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-07-08&lt;/td&gt;
 &lt;td style="text-align: right"&gt;277,500 won&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-6.25%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 874.1 billion&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+282.7 billion won&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+584.7 billion won&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The source is the local Kiwoom-based &lt;code&gt;prices_daily&lt;/code&gt; and &lt;code&gt;investor_flow_daily&lt;/code&gt; tables. On July 8, the foreign ownership rate was recorded at 46.58%.&lt;/p&gt;
&lt;p&gt;The message of this table is clear. The market did not believe that Samsung Electronics&amp;rsquo; profits had weakened, but that it was a structure in which foreigners reduced their risk and individuals accepted it in stocks that had already risen. On July 7, foreigners and institutions sold together, and individuals bought significantly. Institutions received some on July 8, but foreign selling continued.&lt;/p&gt;
&lt;p&gt;Therefore, today’s judgment is “performance is good, but supply and demand are bad.” It is not a basis for selling, but a basis for freezing additional purchases.&lt;/p&gt;
&lt;h2 id="4-four-axes-of-samsung-electronics-thesis"&gt;4. Four axes of Samsung Electronics thesis
&lt;/h2&gt;&lt;p&gt;The argument for investing in Samsung Electronics can be organized into four axes.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;axis&lt;/th&gt;
 &lt;th&gt;Content&lt;/th&gt;
 &lt;th&gt;Current judgment&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;AI server DRAM·HBM&lt;/td&gt;
 &lt;td&gt;Demand for high-bandwidth and high-capacity memory for AI servers and GPU/ASIC systems&lt;/td&gt;
 &lt;td&gt;Strengthening&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;eSSD·NAND mix&lt;/td&gt;
 &lt;td&gt;AI server storage and enterprise SSD demand&lt;/td&gt;
 &lt;td&gt;Strengthening&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Hyperscaler CapEx&lt;/td&gt;
 &lt;td&gt;Long-term investment in cloud and AI data centers&lt;/td&gt;
 &lt;td&gt;Strengthening&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Foundry turnaround&lt;/td&gt;
 &lt;td&gt;Edge logic·packaging·customer recovery options&lt;/td&gt;
 &lt;td&gt;Not yet an option&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Among these, the key ones are numbers 1 to 3. Foundry is a valuation upside option. For the stock price to hold up now, the foundry does not need to improve immediately. Investments in memory, eSSD, and hyperscaler alone explain the strength of profits. However, additional rerating requires reducing the foundry deficit and confirming HBM4 and HBM4E customers.&lt;/p&gt;
&lt;h2 id="5-signal-1-legacy-dram-and-nand-prices-are-stronger-than-expected"&gt;5. Signal 1: Legacy DRAM and NAND prices are stronger than expected
&lt;/h2&gt;&lt;p&gt;The most important bullish signal in the attached research is the memory price trend summarized in the comments of Meritz Securities Commissioner Kim Seon-woo. Since the full text of the original data was not directly compared, this article classifies it as “market comment-based interpretation.” Still, numbers and direction are important in interpreting Samsung Electronics&amp;rsquo; investment argument.&lt;/p&gt;
&lt;p&gt;The key point is that seasonal demand for the second half of the year began to arrive in July, and while B2B server demand was absorbing supply, B2C smartphone and PC customers also began to feel pressure to secure supplies.&lt;/p&gt;
&lt;h3 id="legacy-dram"&gt;Legacy DRAM
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Figures and Content&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Nanya 3Q26 DRAM price proposal&lt;/td&gt;
 &lt;td&gt;DDR4·LPDDR4 +50~90% QoQ&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Consumer DRAM price&lt;/td&gt;
 &lt;td&gt;Above $3/Gb, upward acceleration&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;supply expansion&lt;/td&gt;
 &lt;td&gt;Judgment that meaningful supply expansion will be difficult before 2027&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cause&lt;/td&gt;
 &lt;td&gt;Decrease in investment between 2023 and 2025 after loss in 2023, lack of cleanroom&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h3 id="legacy-nand"&gt;Legacy NAND
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Figures and Content&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Windbond·MXIC SLC NAND proposal&lt;/td&gt;
 &lt;td&gt;3Q26 +70~100% QoQ&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;price range&lt;/td&gt;
 &lt;td&gt;$1.9~2.0/Gb&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Demand changes&lt;/td&gt;
 &lt;td&gt;SLC NAND orders increase due to MLC NAND shortage&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Interpretation&lt;/td&gt;
 &lt;td&gt;Price pressure spreads not only to high-performance AI NAND but also to legacy NAND&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h3 id="supply-structure"&gt;Supply Structure
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Content&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM total supply growth rate in 2027&lt;/td&gt;
 &lt;td&gt;Expected 20~25%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Server demand growth rate next year&lt;/td&gt;
 &lt;td&gt;50~70% expected&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Bottleneck causes&lt;/td&gt;
 &lt;td&gt;Low-end/legacy shortage occurs due to concentration on high-end memory&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;[Inference] If this data is correct, Samsung Electronics&amp;rsquo; bull case is wider than SK Hynix&amp;rsquo;s HBM pure exposure. Samsung Electronics has a discount factor for being behind in HBM, but it also receives price increases for general DRAM, server DRAM, eSSD, NAND, and legacy products. In other words, “the frame of disparaging Samsung Electronics based solely on HBM” is weakened.&lt;/p&gt;
&lt;h2 id="6-signal-2-it-is-difficult-to-say-that-ai-capex-has-decreased-yet"&gt;6. Signal 2: It is difficult to say that AI CapEx has decreased yet
&lt;/h2&gt;&lt;p&gt;Hyperscaler investment also supports Samsung Electronics&amp;rsquo; thesis. According to numbers shared by Goldman Sachs charts, hyperscaler CapEx grows as follows:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Year&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Hyperscaler CapEx&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;2024&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$227 billion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2025&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$406 billion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026E&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$740 billion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2027E&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$996 billion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The forecast for 2027 is about 4.4 times that of 2024. This number is also in line with the bond and capital increase flows covered in &lt;a class="link" href="https://koreainvestinsights.com/ko/post/hyperscaler-financing-race-ai-capex-memory-bottleneck-2026-07-07/" &gt;Big Tech Financing Relay&lt;/a&gt;. Amazon, Alphabet, and Meta are even using corporate bonds and stock issuances to invest in AI infrastructure at a level that is difficult to cover with internal cash flow alone.&lt;/p&gt;
&lt;p&gt;What is important for Samsung Electronics is that “AI CapEx does not end at GPU.” Investing in an AI data center is not just about purchasing GPUs. Server DRAM, HBM, eSSD, NAND, substrate, power, network, and cooling will also increase.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;AI CapEx Items&lt;/th&gt;
 &lt;th&gt;Connection with Samsung Electronics&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;GPU·ASIC server&lt;/td&gt;
 &lt;td&gt;HBM, server DRAM&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;data center storage device&lt;/td&gt;
 &lt;td&gt;eSSD, NAND&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;High-density server proliferation&lt;/td&gt;
 &lt;td&gt;High capacity DDR5, LPDDR series, storage&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Rack-level AI factory&lt;/td&gt;
 &lt;td&gt;Increased memory payload&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Long-term investment cycle&lt;/td&gt;
 &lt;td&gt;Strengthening visibility in 2027 rather than short-term performance in 2026&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;[Inference] If hyperscaler CapEx increases by 2027, it is difficult to conclude that Samsung Electronics&amp;rsquo; memory profits will immediately decline from the peak in one quarter. The problem is not whether there is no demand, but how much the market has already reflected that demand in the price.&lt;/p&gt;
&lt;h2 id="7-signal-3-ai-server-supply-chain-is-not-yet-broken"&gt;7. Signal 3: AI server supply chain is not yet broken
&lt;/h2&gt;&lt;p&gt;Taiwan AI server supply chain data included in the attached research is also important.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Enterprise&lt;/th&gt;
 &lt;th&gt;Latest Signals&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Wistron&lt;/td&gt;
 &lt;td&gt;June sales +10.9% MoM, +53.8% YoY&lt;/td&gt;
 &lt;td&gt;GB300 NVL72, B300 shipment increases&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Wiwynn&lt;/td&gt;
 &lt;td&gt;June sales +32.5% MoM, +29.8% YoY&lt;/td&gt;
 &lt;td&gt;AMD Helios enters, prepares for mass shipment of 3Q ASIC AI servers&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Unimicron&lt;/td&gt;
 &lt;td&gt;ABF·HDI substrate revenue record high&lt;/td&gt;
 &lt;td&gt;ABF expansion CapEx increase mentioned&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This data does not fit the claim that “demand for AI servers has declined.” If server OEMs, ODMs, and substrate companies still show strong shipments and sales, memory demand is unlikely to disappear easily.&lt;/p&gt;
&lt;p&gt;물론 이 데이터만으로 삼성전자 실적을 바로 산출할 수는 없다. But the direction is clear. AI 서버 공급망은 아직 멈추지 않았다. 오히려 GPU, ASIC, 서버, 기판, 메모리, storage가 함께 움직이는 구조다.&lt;/p&gt;
&lt;h2 id="8-소음-1-trendforce-spot-데이터는-약하지만-붕괴는-아니다"&gt;8. 소음 1: TrendForce spot 데이터는 약하지만 붕괴는 아니다
&lt;/h2&gt;&lt;p&gt;There are also opposing signs. TrendForce spot market comments were summarized as an atmosphere where trading was quiet and buyers were waiting.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;spot change&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;DDR5 16Gb&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+0.50%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DDR5 eTT&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0.00%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DDR4 16Gb 3200&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+3.42%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DDR4 eTT&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.51%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DDR4 8Gb 3200&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+2.88%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Looking at this table, it is excessive to say, “Memory prices have fallen.” 대부분은 상승 또는 보합이고, 일부 eTT 품목만 하락했다. 더 정확한 표현은 “거래는 한산하고, spot 시장은 가격 발견 중”이다.&lt;/p&gt;
&lt;p&gt;The important distinction is spot and contract.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Category&lt;/th&gt;
 &lt;th&gt;meaning&lt;/th&gt;
 &lt;th&gt;Relevance to Samsung Electronics thesis&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Spot&lt;/td&gt;
 &lt;td&gt;Short-term transaction prices in distribution and module markets&lt;/td&gt;
 &lt;td&gt;Quickly reflects sentiment and short-term inventory&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Contract&lt;/td&gt;
 &lt;td&gt;Large customer long-term contract price&lt;/td&gt;
 &lt;td&gt;Server DRAM, HBM, eSSD more important to profits&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;삼성전자 thesis의 핵심은 spot이 아니다. 핵심은 contract 가격, server DRAM, HBM, eSSD 주문, hyperscaler 수요다. 진짜 위험 신호는 spot 거래 부진 자체가 아니라 다음 네 가지가 함께 나오는 경우다.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Contract price cut&lt;/li&gt;
&lt;li&gt;Slowdown in server DRAM orders&lt;/li&gt;
&lt;li&gt;HBM·eSSD order slowdown&lt;/li&gt;
&lt;li&gt;주요 고객의 inventory correction 코멘트&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;Currently, all four have not been confirmed. Therefore, it is correct to view TrendForce data not as bear confirmation but as a signal of anticipation digestion and price discovery.&lt;/p&gt;
&lt;h2 id="9-noise-2-morgan-stanley-peak-logic-is-only-partially-correct"&gt;9. Noise 2: Morgan Stanley peak logic is only partially correct
&lt;/h2&gt;&lt;p&gt;Morgan Stanley-type peak thesis should not be ignored either. The key point is that the speed of improvement in memory indicators is approaching its peak, and inventory normalization and price increases may slow down. AI trade 안에서는 반도체보다 Alphabet, Amazon 같은 AI cloud 쪽을 선호한다는 관점도 제시됐다.&lt;/p&gt;
&lt;p&gt;This claim is partly correct. This is especially important for short-term stock prices.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Morgan Stanley-style warning&lt;/th&gt;
 &lt;th&gt;Areas to acknowledge&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Improvement speed peak possibility&lt;/td&gt;
 &lt;td&gt;Stock prices are more sensitive to slope than level.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Price increases slowing&lt;/td&gt;
 &lt;td&gt;3Q 이후 상승률이 낮아지면 멀티플 압박 가능&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Semiconductor overcrowding position&lt;/td&gt;
 &lt;td&gt;Sell-the-news is possible even with good performance&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AI cloud preferred&lt;/td&gt;
 &lt;td&gt;Possibility of rotation towards consumers rather than semiconductors&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;However, extending this to the logic of structural selling is weak. There is too much opposing data.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Opposite logic&lt;/th&gt;
 &lt;th&gt;Evidence&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Continued expansion of AI investment&lt;/td&gt;
 &lt;td&gt;Hyperscaler CapEx expected to increase by 2027&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Prolonged supply shortage&lt;/td&gt;
 &lt;td&gt;Legacy shortage intensifies due to transition to cleanroom and high value-added products&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Server demand growth&lt;/td&gt;
 &lt;td&gt;Server demand expected to increase by 50~70% next year&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Domestic securities firms continue to buy&lt;/td&gt;
 &lt;td&gt;Target prices maintained or raised, including KB 600,000 won, Daol 585,000 won, NH 530,000 won, and Hana/iM 480,000 won&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;[Inference] The Morgan Stanley peak thesis is valid as a warning, “Don’t chase now.” However, the basis for “discarding Samsung Electronics’ thesis” is insufficient.&lt;/p&gt;
&lt;h2 id="10-noise-3-asml-and-tsmc-are-high-bar-events"&gt;10. Noise 3: ASML and TSMC are high bar events
&lt;/h2&gt;&lt;p&gt;The next gateway is ASML and TSMC. ASML&amp;rsquo;s official financial calendar displays Q2 2026 financial results on July 15, 2026. TSMC&amp;rsquo;s official IR page announces the 2Q26 earnings conference on July 16, 2026. Source: &lt;a class="link" href="https://www.asml.com/investors/financial-calendar" target="_blank" rel="noopener"
 &gt;ASML financial calendar&lt;/a&gt;, &lt;a class="link" href="https://investor.tsmc.com/english/quarterly-results/2026/q2" target="_blank" rel="noopener"
 &gt;TSMC 2Q26 quarterly results&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;The problem is that even if good performance is achieved, it is impossible to know whether the stock price will rise. Samsung Electronics&amp;rsquo; 2Q26 provisional results are already a preview. If the high bar is high, even good numbers seem lacking.&lt;/p&gt;
&lt;p&gt;Here&amp;rsquo;s what to look for in ASML and TSMC:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Event&lt;/th&gt;
 &lt;th&gt;checkpoint&lt;/th&gt;
 &lt;th&gt;Connection with Samsung Electronics&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;ASML 2Q26&lt;/td&gt;
 &lt;td&gt;bookings, EUV·High-NA demand, Chinese regulatory comments&lt;/td&gt;
 &lt;td&gt;Confirmation of semiconductor capex expectations and equipment demand&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;TSMC 2Q26&lt;/td&gt;
 &lt;td&gt;AI/HPC revenue, CoWoS·advanced packaging, CapEx guidance&lt;/td&gt;
 &lt;td&gt;Identifying AI accelerator and HBM packaging bottlenecks&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Market reaction&lt;/td&gt;
 &lt;td&gt;SOXX, NVDA, MU, TSM, ASML stock price reaction&lt;/td&gt;
 &lt;td&gt;Check if it is sold or relieved with good performance&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;As Mizuho commented, the fact that Samsung Electronics&amp;rsquo; stock price fell despite its good preliminary results shows the possibility of a global IT momentum shift. So ASML and TSMC are events that see “how the market digests good numbers” rather than “confirming the numbers.”&lt;/p&gt;
&lt;h2 id="11-noise-4-press-multiple-for-interest-rates-and-exchange-rates"&gt;11. Noise 4: Press multiple for interest rates and exchange rates
&lt;/h2&gt;&lt;p&gt;Even if Samsung Electronics&amp;rsquo; thesis is good, the stock price may be depressed if the discount rate increases. The interest rate environment based on the attached research is as follows.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;level&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;US 10-year note&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 4.58%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;US 30-year note&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 5.07%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Japanese 10-year note&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 2.88%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Japanese 40-year note&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 4.01%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;These numbers do not undermine the demand thesis. However, PER can be lowered. A typical section where stock prices fall despite strong profits is “the section where EPS rises but the multiple falls faster.”&lt;/p&gt;
&lt;p&gt;The same goes for exchange rates. A weak won may be beneficial to Samsung Electronics&amp;rsquo; performance. However, from a foreigner&amp;rsquo;s perspective, it increases foreign exchange losses, foreign exchange hedging costs, and the risk of Korea&amp;rsquo;s share. The foreign ownership rate of 46.58% and continuous selling on July 8 are signals that this part needs to be checked.&lt;/p&gt;
&lt;h2 id="12-noise-5-foreign-outflow-and-leverage-piping"&gt;12. Noise 5: Foreign outflow and leverage piping
&lt;/h2&gt;&lt;p&gt;The attached research includes comments on foreign outflow and leverage piping based on Bloomberg and Telegram. The figures presented here, such as foreign outflow in the Korean market since the beginning of the year -$100.16 billion, foreign securities outflow in March -$36.55 billion, and USD/KRW 1,554 won, need to be compared with official statistics by collection range. Therefore, in this article, only directional signals are used.&lt;/p&gt;
&lt;p&gt;Still, the message is important.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Observation&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Foreigners reduced Korea&lt;/td&gt;
 &lt;td&gt;Rather than avoiding Korea, it may be a memory peak-out bet.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Domestic institutions and individuals absorbed the rebound&lt;/td&gt;
 &lt;td&gt;It is a domestic liquidity absorption market rather than a foreign-led rally.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Single stock leverage/short gamma structure&lt;/td&gt;
 &lt;td&gt;Forced selling and hedging flows during declines can exaggerate prices.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Exchange rate rise&lt;/td&gt;
 &lt;td&gt;It is a barrier to additional purchases by foreigners.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;[Inference] The short-term plunge is more likely to be the result of memory peak-out betting, liquidation of leverage positions, and reduction of foreign risks, rather than the collapse of Samsung Electronics&amp;rsquo; main business. However, we cannot conclude that this is not a structural risk. We need to check whether foreign spot sales stop.&lt;/p&gt;
&lt;h2 id="13-korean-market-performance-is-also-not-good"&gt;13. Korean market performance is also not good
&lt;/h2&gt;&lt;p&gt;This is not just a problem for Samsung Electronics. As of 06:10 on July 8, the local KR integrated screener classified the Korean market as NEUTRAL.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;figures&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Proportion of stocks above the 50-day line&lt;/td&gt;
 &lt;td style="text-align: right"&gt;15.6%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Proportion of stocks above the 200-day line&lt;/td&gt;
 &lt;td style="text-align: right"&gt;21.5%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Passed items&lt;/td&gt;
 &lt;td style="text-align: right"&gt;17&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;New Passed&lt;/td&gt;
 &lt;td style="text-align: right"&gt;5&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Breakaway&lt;/td&gt;
 &lt;td style="text-align: right"&gt;7&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Breakout&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Newly passed stocks were APR, Handsome, GS Retail, Hana Financial Group, and KT&amp;amp;G. Stocks that left included TSE, YG-1, and Samsung Electronics.&lt;/p&gt;
&lt;p&gt;13 passed the quality compounder screen, and at the top were Gigabis, SK Hynix, VM, AP, and PSK. Only three companies, APR, Seegene, and Kolmar Korea, passed the smart money + quality condition.&lt;/p&gt;
&lt;p&gt;The meaning of this data is simple. Market breadth is weak. Even if Samsung Electronics&amp;rsquo; thesis is good, it is not widely accepted by the entire Korean market. In this environment, additional purchases should not be made by reporting only profit data. We need to see whether supply and demand and market width return.&lt;/p&gt;
&lt;h2 id="14-then-what-should-i-do"&gt;14. Then what should I do?
&lt;/h2&gt;&lt;p&gt;The current conclusion is “strong holding, freeze on additional purchases.”&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;action&lt;/th&gt;
 &lt;th&gt;Judgment&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;holding&lt;/td&gt;
 &lt;td&gt;possible. The thesis is intact.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Additional Purchase&lt;/td&gt;
 &lt;td&gt;Still waiting. Need to confirm supply/demand, ASML, TSMC, IR&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Sell ​​&lt;/td&gt;
 &lt;td&gt;It&amp;rsquo;s not just data now. The profit direction actually improved.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Short term trading&lt;/td&gt;
 &lt;td&gt;After confirming the slowdown in foreign selling and the global semiconductor response from July 15th to 16th&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;There are four additional purchase conditions:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Foreign spot sales of Samsung Electronics are slowing down, and programmatic selling pressure is decreasing.&lt;/li&gt;
&lt;li&gt;ASML bookings and TSMC AI/HPC·CoWoS·CapEx comments do not indicate a slowdown in AI investment.&lt;/li&gt;
&lt;li&gt;DS core profit, DRAM·NAND contract price, HBM4·HBM4E customer progress, and eSSD orders are confirmed in Samsung Electronics IR.&lt;/li&gt;
&lt;li&gt;The sluggish spot market does not spread to contract price cuts or a slowdown in server DRAM orders.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;The conditions for invalidation are as follows:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Invalidation Conditions&lt;/th&gt;
 &lt;th&gt;meaning&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Contract DRAM/NAND price decline&lt;/td&gt;
 &lt;td&gt;price thesis undermined&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Server DRAM·HBM·eSSD order slowdown&lt;/td&gt;
 &lt;td&gt;AI memory demand thesis undermined&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;ASML·TSMC strongly talks about CapEx slowdown&lt;/td&gt;
 &lt;td&gt;Hyperscaler investment thesis weakened&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Foreign selling and won weakening intensify simultaneously&lt;/td&gt;
 &lt;td&gt;Pressure to reduce Korea&amp;rsquo;s proportion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Market breadth further worsens&lt;/td&gt;
 &lt;td&gt;Even if Samsung Electronics is good, supply and demand absorption is weakened&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="15-conclusion"&gt;15. Conclusion
&lt;/h2&gt;&lt;p&gt;Samsung Electronics&amp;rsquo; 2Q26 provisional performance is not a number that weakens the thesis. Rather, it is the opposite. Operating profit of KRW 89.4 trillion, DRAM ASP +47%, NAND ASP +66%, soaring legacy memory prices, hyperscaler CapEx continuing until 2027, and AI server supply chain sales all strengthen the case for investing in Samsung Electronics.&lt;/p&gt;
&lt;p&gt;However, today&amp;rsquo;s stock prices and supply and demand have damaged the timing. Foreigners sold, and individuals received. Interest rates are high, exchange rates are a burden, and high bar events such as ASML and TSMC remain. Market width is also weak.&lt;/p&gt;
&lt;p&gt;So the conclusion is clear.&lt;/p&gt;
&lt;p&gt;Samsung Electronics has not become worse. However, today’s market called for “the next good number and supply and demand to believe in that number” rather than “a good number.” Now is not the time to abandon the thesis, but rather to protect the position, stop making additional purchases, and check for the next gateway.&lt;/p&gt;
&lt;h2 id="basis-classification"&gt;Basis classification
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Category&lt;/th&gt;
 &lt;th&gt;Content&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Fact&lt;/td&gt;
 &lt;td&gt;Samsung Electronics&amp;rsquo; official 2Q26 provisional sales of 171 trillion won and operating profit of 89.4 trillion won. Comparative figures for 1Q26 and 2Q25. ASML July 15th, TSMC July 16th earnings schedule. Samsung Electronics price, supply and demand, and foreign ownership rate in local DB.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Inference&lt;/td&gt;
 &lt;td&gt;Samsung Electronics&amp;rsquo; thesis was strengthened, but the short-term timing was judged to be damaged. TrendForce spot data is interpreted as price discovery, not bear confirmation. Morgan Stanley&amp;rsquo;s peak thesis is a catch-up warning and not a basis for structural selling.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Speculation&lt;/td&gt;
 &lt;td&gt;A scenario in which hyperscaler CapEx increases to approximately $1 trillion by 2027 and can support Samsung Electronics&amp;rsquo; memory demand for a longer period of time.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Blocked&lt;/td&gt;
 &lt;td&gt;Comparison of Bloomberg foreign outflow figures with official statistics, confirmation of the full text report by Meritz·TrendForce, 2Q26 actual profit by Samsung Electronics DS division, and volume and margin by HBM4·HBM4E customer.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="evidence-ledger"&gt;Evidence Ledger
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Source&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics 2Q26 provisional performance&lt;/td&gt;
 &lt;td&gt;&lt;a class="link" href="https://news.samsung.com/global/samsung-electronics-announces-earnings-guidance-for-second-quarter-2026" target="_blank" rel="noopener"
 &gt;Samsung Electronics Official Newsroom&lt;/a&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;ASML 2Q26 Schedule&lt;/td&gt;
 &lt;td&gt;&lt;a class="link" href="https://www.asml.com/investors/financial-calendar" target="_blank" rel="noopener"
 &gt;ASML financial calendar&lt;/a&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;TSMC 2Q26 Schedule&lt;/td&gt;
 &lt;td&gt;&lt;a class="link" href="https://investor.tsmc.com/english/quarterly-results/2026/q2" target="_blank" rel="noopener"
 &gt;TSMC 2Q26 quarterly results&lt;/a&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics price/supply&lt;/td&gt;
 &lt;td&gt;Local Research DB: &lt;code&gt;prices_daily&lt;/code&gt;, &lt;code&gt;investor_flow_daily&lt;/code&gt;, &lt;code&gt;foreign_ownership_daily&lt;/code&gt;, as of 2026-07-08&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Memory price·AI server·foreign outflow comments&lt;/td&gt;
 &lt;td&gt;Attached research packet. Uncollated items in the full text are separated from the main text as market comments or directional signals&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Related prior analysis&lt;/td&gt;
 &lt;td&gt;Korea Invest Insights&amp;rsquo; Samsung Electronics 2Q26 preview, Micron FY3Q26, Big Tech earnings call, Big Tech financing, NVIDIA similar quarter analysis&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;</description></item><item><title>Which NVIDIA Quarter Does Samsung's Earnings Selloff Resemble?</title><link>https://koreainvestinsights.com/post/samsung-earnings-selloff-nvidia-q4fy26-rebound-trigger-2026-07-08/</link><pubDate>Wed, 08 Jul 2026 14:30:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/samsung-earnings-selloff-nvidia-q4fy26-rebound-trigger-2026-07-08/</guid><description>
 &lt;blockquote&gt;
 &lt;p&gt;This is a follow-up to &lt;a class="link" href="https://koreainvestinsights.com/post/samsung-2q26-preview-micron-surprise-erased-core-op-hbm-2026-06-29/" &gt;Samsung 2Q26 preview&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/micron-fy3q26-ai-memory-sca-fcf-hold-2026-06-25/" &gt;Micron FY3Q26 earnings&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/nvidia-earnings-elasticity-hbm-cycle-samsung-hynix-2026-06-28/" &gt;NVIDIA elasticity and the Korea HBM cycle&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/big-tech-july-earnings-call-memory-thesis-scenarios-2026-07-07/" &gt;late-July Big Tech calls and the memory thesis&lt;/a&gt;, and &lt;a class="link" href="https://koreainvestinsights.com/post/hyperscaler-financing-race-ai-capex-memory-bottleneck-2026-07-07/" &gt;hyperscaler financing and memory bottlenecks&lt;/a&gt;. Related hubs: &lt;a class="link" href="https://koreainvestinsights.com/page/exclusive-analysis-hub/" &gt;Exclusive Analysis&lt;/a&gt; and &lt;a class="link" href="https://koreainvestinsights.com/page/korea-semiconductor-hbm-kospi-hub/" &gt;AI HBM&lt;/a&gt;.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;p&gt;Samsung Electronics&amp;rsquo; 2Q26 preliminary earnings were strong on the surface. Revenue was about KRW 171 trillion and operating profit was about KRW 89.4 trillion. Sales rose 27.74% quarter over quarter and operating profit rose 56.21%. Operating profit was up 1,810.26% year over year. These figures were disclosed by Samsung Electronics.&lt;/p&gt;
&lt;p&gt;Yet the stock fell 6.92% on July 7 and another 6.25% on July 8. The two-day decline was 12.7%.&lt;/p&gt;
&lt;p&gt;That is not the same as the famous NVIDIA August 2024 sell-on reaction. NVIDIA fell 6.4% on the first reaction day in August 2024, but rebounded 1.5% the next day. Samsung did not bounce. It fell again.&lt;/p&gt;
&lt;p&gt;Across 17 NVIDIA post-earnings reactions, the closest analogue is NVIDIA&amp;rsquo;s Q4 FY26 report on February 25, 2026. NVIDIA fell 5.5% on the first reaction day and 4.2% the next day, for a two-day decline of 9.4%. The reaction-vector distance from Samsung&amp;rsquo;s path is 2.55, the smallest in the sample.&lt;/p&gt;
&lt;p&gt;The key point is not that a strong earnings print is useless. The point is that a strong print alone was not enough once the market started asking whether AI infrastructure growth was slowing. NVIDIA&amp;rsquo;s large post-print drawdowns eventually recovered, but the recovery came only after a new catalyst directly killed the fear that had caused the drawdown. For Samsung, the equivalent catalyst is late-July Big Tech CapEx guidance.&lt;/p&gt;
&lt;p&gt;The practical conclusion is Wait. This is not a mechanical dip-buy setup yet. The next gates are late-July hyperscaler CapEx commentary, DRAM/NAND price momentum, and Samsung&amp;rsquo;s clean 3Q operating-profit path.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="1-what-was-strange-about-samsungs-earnings-reaction"&gt;1. What was strange about Samsung&amp;rsquo;s earnings reaction
&lt;/h2&gt;&lt;p&gt;The numbers were strong.&lt;/p&gt;
&lt;p&gt;Samsung Electronics announced 2Q26 preliminary earnings on July 7, 2026. The company disclosed the following consolidated figures.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Metric&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2Q26 preliminary&lt;/th&gt;
 &lt;th style="text-align: right"&gt;1Q26&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2Q25&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about KRW 171T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 133.87T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 74.57T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Operating profit&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about KRW 89.4T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 57.23T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 4.68T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Revenue growth&lt;/td&gt;
 &lt;td style="text-align: right"&gt;&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+27.74% QoQ&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+129.31% YoY&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Operating profit growth&lt;/td&gt;
 &lt;td style="text-align: right"&gt;&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+56.21% QoQ&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+1,810.26% YoY&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Samsung also disclosed estimate ranges of KRW 170T to KRW 172T for sales and KRW 89.3T to KRW 89.5T for operating profit. Korean disclosure rules required the company to publish a single median figure.&lt;/p&gt;
&lt;p&gt;The stock reaction was the problem. Samsung closed at KRW 318,000 on July 6, KRW 296,000 on July 7, and KRW 277,500 on July 8. Based on yfinance daily close data:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Date&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Close&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Daily return&lt;/th&gt;
 &lt;th&gt;Read-through&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-07-06&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 318,000&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+2.75%&lt;/td&gt;
 &lt;td&gt;pre-announcement base&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-07-07&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 296,000&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-6.92%&lt;/td&gt;
 &lt;td&gt;announcement-day selloff&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-07-08&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 277,500&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-6.25%&lt;/td&gt;
 &lt;td&gt;no bounce&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Two-day cumulative&lt;/td&gt;
 &lt;td style="text-align: right"&gt;&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-12.74%&lt;/td&gt;
 &lt;td&gt;post-print path&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Compared with the June 18 high of KRW 362,500, the July 8 close was already 23.45% lower. Samsung was not falling from a calm setup. It was already volatile before the earnings print, and the good number failed to stop the selling.&lt;/p&gt;
&lt;p&gt;The key is that the market was no longer asking only how much Samsung earned. It was asking whether the pace of earnings improvement could keep accelerating.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="2-why-compare-samsung-with-nvidia"&gt;2. Why compare Samsung with NVIDIA
&lt;/h2&gt;&lt;p&gt;Samsung is a memory company. NVIDIA is a GPU platform company. Their businesses are different. But both have been core price signals for the AI infrastructure cycle.&lt;/p&gt;
&lt;p&gt;NVIDIA moved first when the AI infrastructure boom began. As bottlenecks moved downstream into HBM, DRAM, eSSD, substrates, and power infrastructure, Samsung and SK hynix began facing the same market question: is the next leg still accelerating?&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Question&lt;/th&gt;
 &lt;th&gt;Meaning&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Did earnings improve?&lt;/td&gt;
 &lt;td&gt;confirmed result&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Did earnings beat the bar?&lt;/td&gt;
 &lt;td&gt;surprise versus expectations&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Will the next quarter be better?&lt;/td&gt;
 &lt;td&gt;guidance and pricing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Is the growth rate still rising?&lt;/td&gt;
 &lt;td&gt;the part stocks care about most&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Will hyperscalers keep spending?&lt;/td&gt;
 &lt;td&gt;the final payer for AI memory demand&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Samsung&amp;rsquo;s two-day selloff came from that last layer. The print was strong, but expectations were already high. Some investors had expected operating profit above KRW 90T. At the same time, the market was worried about possible hyperscaler CapEx moderation. Korean single-stock leverage products, futures, and ETF flows amplified the move.&lt;/p&gt;
&lt;p&gt;This analysis therefore does not compare Samsung&amp;rsquo;s business model with NVIDIA&amp;rsquo;s business model. It compares post-earnings price paths and the fear behind those paths.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="3-timing-adjustment-nvidia-reports-after-the-close"&gt;3. Timing adjustment: NVIDIA reports after the close
&lt;/h2&gt;&lt;p&gt;NVIDIA usually reports after the US market close. The earnings information is not reflected in the announcement-day close. The real reaction appears in the next regular trading session.&lt;/p&gt;
&lt;p&gt;Samsung&amp;rsquo;s guidance was reflected in the Korean session on the announcement day. So the timing has to be aligned as follows.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Company&lt;/th&gt;
 &lt;th&gt;Reporting style&lt;/th&gt;
 &lt;th&gt;First trading session with full reaction&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td&gt;reflected in the Korean session&lt;/td&gt;
 &lt;td&gt;announcement day&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NVIDIA&lt;/td&gt;
 &lt;td&gt;after-market release&lt;/td&gt;
 &lt;td&gt;next trading day&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;In this note, NVIDIA&amp;rsquo;s reaction D0 means the first trading day after the after-market release. NVIDIA&amp;rsquo;s D+1 means the following trading day. This aligns NVIDIA&amp;rsquo;s D0 with Samsung&amp;rsquo;s July 7 and NVIDIA&amp;rsquo;s D+1 with Samsung&amp;rsquo;s July 8.&lt;/p&gt;
&lt;p&gt;Without this adjustment, the comparison would be wrong.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="4-nvidias-17-quarter-post-earnings-reaction-table"&gt;4. NVIDIA&amp;rsquo;s 17-quarter post-earnings reaction table
&lt;/h2&gt;&lt;p&gt;The table below combines NVIDIA quarterly revenue, growth rates, and post-earnings stock reactions. Revenue and report timing are sourced from NVIDIA IR and SEC 8-K releases. Stock returns are calculated from yfinance daily close data.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Quarter&lt;/th&gt;
 &lt;th&gt;Report date&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Revenue&lt;/th&gt;
 &lt;th style="text-align: right"&gt;QoQ&lt;/th&gt;
 &lt;th style="text-align: right"&gt;YoY&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Reaction D0&lt;/th&gt;
 &lt;th style="text-align: right"&gt;D+1&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Two-day&lt;/th&gt;
 &lt;th style="text-align: right"&gt;T+20&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Q1 FY23&lt;/td&gt;
 &lt;td&gt;2022-05-25&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$8.3B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;n/a&lt;/td&gt;
 &lt;td style="text-align: right"&gt;n/a&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+5.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+5.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+10.8%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+0.9%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q2 FY23&lt;/td&gt;
 &lt;td&gt;2022-08-24&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$6.7B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-19.3%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;n/a&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+4.0%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-9.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-5.6%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-27.1%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q3 FY23&lt;/td&gt;
 &lt;td&gt;2022-11-16&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$5.9B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-11.9%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;n/a&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.7%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-3.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+6.5%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q4 FY23&lt;/td&gt;
 &lt;td&gt;2023-02-22&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$6.1B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+3.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;n/a&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+14.0%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.6%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+12.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+27.5%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q1 FY24&lt;/td&gt;
 &lt;td&gt;2023-05-24&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$7.2B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+18.0%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-13.3%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+24.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+2.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+27.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+38.2%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q2 FY24&lt;/td&gt;
 &lt;td&gt;2023-08-23&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$13.5B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+87.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+101.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+0.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-2.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-2.3%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-12.9%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q3 FY24&lt;/td&gt;
 &lt;td&gt;2023-11-21&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$18.1B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+34.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+206.8%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-2.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.9%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-4.3%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-3.7%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q4 FY24&lt;/td&gt;
 &lt;td&gt;2024-02-21&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$22.1B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+22.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+262.3%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+16.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+0.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+16.8%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+33.9%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q1 FY25&lt;/td&gt;
 &lt;td&gt;2024-05-22&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$26.0B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+17.6%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+261.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+9.3%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+2.6%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+12.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+33.3%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q2 FY25&lt;/td&gt;
 &lt;td&gt;2024-08-28&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$30.0B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+15.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+122.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-6.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+1.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-5.0%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.2%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q3 FY25&lt;/td&gt;
 &lt;td&gt;2024-11-20&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$35.1B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+17.0%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+93.9%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+0.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-3.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-2.7%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-10.4%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q4 FY25&lt;/td&gt;
 &lt;td&gt;2025-02-26&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$39.3B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+12.0%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+77.8%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-8.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+4.0%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-4.8%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-13.3%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q1 FY26&lt;/td&gt;
 &lt;td&gt;2025-05-28&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$44.1B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+12.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+69.6%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+3.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-2.9%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+0.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+15.0%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q2 FY26&lt;/td&gt;
 &lt;td&gt;2025-08-27&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$46.7B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+5.9%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+55.7%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-0.8%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-3.3%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-4.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-2.2%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q3 FY26&lt;/td&gt;
 &lt;td&gt;2025-11-19&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$57.0B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+22.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+62.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-3.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.0%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-4.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-6.6%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q4 FY26&lt;/td&gt;
 &lt;td&gt;2026-02-25&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$68.1B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+19.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+73.3%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-5.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-4.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-9.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-8.6%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q1 FY27&lt;/td&gt;
 &lt;td&gt;2026-05-20&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$81.6B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+19.8%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+85.0%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.8%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.9%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-3.6%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-5.7%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The key message is clear. Strong results did not always lift the stock. As the AI cycle matured, the market became more sensitive to the change in growth rate than to the level of revenue.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="5-the-closest-analogue-is-nvidia-q4-fy26"&gt;5. The closest analogue is NVIDIA Q4 FY26
&lt;/h2&gt;&lt;p&gt;Samsung&amp;rsquo;s path was:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Company/event&lt;/th&gt;
 &lt;th style="text-align: right"&gt;D0&lt;/th&gt;
 &lt;th style="text-align: right"&gt;D+1&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Two-day&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung 2Q26 preliminary earnings&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-6.92%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-6.25%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-12.74%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;We compare this two-day vector with all 17 NVIDIA quarters.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th style="text-align: right"&gt;Rank&lt;/th&gt;
 &lt;th&gt;NVIDIA quarter&lt;/th&gt;
 &lt;th&gt;Report date&lt;/th&gt;
 &lt;th style="text-align: right"&gt;D0&lt;/th&gt;
 &lt;th style="text-align: right"&gt;D+1&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Two-day&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Distance from Samsung&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;1&lt;/td&gt;
 &lt;td&gt;Q4 FY26&lt;/td&gt;
 &lt;td&gt;2026-02-25&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-5.46%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-4.16%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-9.39%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2.55&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;2&lt;/td&gt;
 &lt;td&gt;Q3 FY24&lt;/td&gt;
 &lt;td&gt;2023-11-21&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-2.46%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.93%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-4.34%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;6.21&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;3&lt;/td&gt;
 &lt;td&gt;Q3 FY26&lt;/td&gt;
 &lt;td&gt;2025-11-19&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-3.15%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-0.97%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-4.10%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;6.48&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;4&lt;/td&gt;
 &lt;td&gt;Q1 FY27&lt;/td&gt;
 &lt;td&gt;2026-05-20&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.77%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.90%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-3.64%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;6.74&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;5&lt;/td&gt;
 &lt;td&gt;Q2 FY26&lt;/td&gt;
 &lt;td&gt;2025-08-27&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-0.79%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-3.32%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-4.09%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;6.79&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Q4 FY26 is the clear match. It is not just the numbers. The drivers were similar.&lt;/p&gt;
&lt;p&gt;NVIDIA&amp;rsquo;s Q4 FY26 revenue was $68.1B, up 73% year over year and 20% quarter over quarter. The next-quarter revenue guide was $78B. The print was strong. But the stock fell 5.5% on the first reaction day and 4.2% on the next day.&lt;/p&gt;
&lt;p&gt;The market was asking whether the next step could still get better. Customer concentration, hyperscaler spending durability, AI CapEx peak risk, and competition were all in focus.&lt;/p&gt;
&lt;p&gt;Samsung faced the same structure. KRW 89.4T of operating profit was strong, but investors had already moved the bar higher. The market wanted proof that the growth rate was still accelerating, not only proof that the current profit level was high.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="6-why-the-august-2024-nvidia-sell-on-is-the-wrong-template"&gt;6. Why the August 2024 NVIDIA sell-on is the wrong template
&lt;/h2&gt;&lt;p&gt;The intuitive comparison is NVIDIA Q2 FY25 in August 2024. NVIDIA fell 6.4% on the first reaction day. That resembles Samsung&amp;rsquo;s July 7 decline.&lt;/p&gt;
&lt;p&gt;But the next day changes the conclusion.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Case&lt;/th&gt;
 &lt;th style="text-align: right"&gt;D0&lt;/th&gt;
 &lt;th style="text-align: right"&gt;D+1&lt;/th&gt;
 &lt;th&gt;Read-through&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;NVIDIA Q2 FY25, August 2024&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-6.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+1.5%&lt;/td&gt;
 &lt;td&gt;one-day selloff, then bounce&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NVIDIA Q4 FY25, February 2025&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-8.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+4.0%&lt;/td&gt;
 &lt;td&gt;one-day selloff, then strong bounce&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NVIDIA Q4 FY26, February 2026&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-5.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-4.2%&lt;/td&gt;
 &lt;td&gt;two-day selloff&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung 2Q26&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-6.92%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-6.25%&lt;/td&gt;
 &lt;td&gt;two-day selloff&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The August 2024 and February 2025 NVIDIA reactions were dip-and-bounce. Samsung was dip-and-continue. That distinction matters. A one-day sell-on can be a dip-buy setup. A two-day selloff can mean the market has not finished digesting the fear.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="7-did-nvidia-recover-immediately-after-large-post-print-selloffs"&gt;7. Did NVIDIA recover immediately after large post-print selloffs?
&lt;/h2&gt;&lt;p&gt;No.&lt;/p&gt;
&lt;p&gt;There were three NVIDIA quarters with a first-reaction-day decline worse than 5%: Q2 FY25, Q4 FY25, and Q4 FY26. All three were still below the pre-announcement price after 20 trading days.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;NVIDIA quarter&lt;/th&gt;
 &lt;th style="text-align: right"&gt;D0&lt;/th&gt;
 &lt;th style="text-align: right"&gt;T+5&lt;/th&gt;
 &lt;th style="text-align: right"&gt;T+10&lt;/th&gt;
 &lt;th style="text-align: right"&gt;T+20&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Q2 FY25, August 2024&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-6.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-14.6%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-5.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.2%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q4 FY25, February 2025&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-8.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-10.6%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-11.8%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-13.3%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q4 FY26, February 2026&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-5.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-6.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-4.9%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-8.6%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;All three fell further by T+5. All three failed to recover the pre-announcement price by T+20.&lt;/p&gt;
&lt;p&gt;The sample is small, so this is not a deterministic rule. But it is enough to reject the simple idea that strong earnings plus a large drop automatically equals an immediate buying opportunity.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="8-the-real-low-and-recovery-took-weeks"&gt;8. The real low and recovery took weeks
&lt;/h2&gt;&lt;p&gt;The second attached analysis goes further. It asks when NVIDIA actually bottomed and when it recovered.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Anchor&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Pre-print base&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Low&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Max drawdown&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Time to low&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Breakeven&lt;/th&gt;
 &lt;th style="text-align: right"&gt;New high&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Low-to-later-high&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Q2 FY25, August 2024&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$125.55&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$102.78 on 2024-09-06&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-18.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;6 sessions&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2024-10-07, 5.5 weeks&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2024-10-14, 6.5 weeks&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+45.3% to 2025-01-06&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q4 FY25, February 2025&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$131.24&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$94.29 around 2025-04-04 to 2025-04-08&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-28.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;27 sessions&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2025-05-14, 11 weeks&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2025-06-25, 16 weeks&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+119.6% to 2025-10-29&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q4 FY26, February 2026&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$195.55&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$165.17 on 2026-03-30&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-15.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;23 sessions&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2026-04-14, 6.5 weeks&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2026-04-24, 8 weeks&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+42.7% to 2026-05-14&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The pattern is important.&lt;/p&gt;
&lt;p&gt;First, the low did not arrive immediately. Even the fastest case took six trading sessions. The slower cases took more than a month.&lt;/p&gt;
&lt;p&gt;Second, breakeven took 5.5 to 11 weeks.&lt;/p&gt;
&lt;p&gt;Third, the rebound was eventually large. But it did not come from cheapness alone. It came after a catalyst removed the fear behind the selloff.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="9-the-rebound-trigger-had-to-kill-the-exact-fear"&gt;9. The rebound trigger had to kill the exact fear
&lt;/h2&gt;&lt;h3 id="9-1-august-2024-the-fed-addressed-macro-fear"&gt;9-1. August 2024: the Fed addressed macro fear
&lt;/h3&gt;&lt;p&gt;The August 2024 drawdown was about macro fear. Weak labor-market data raised recession concerns. Regulatory noise also hurt sentiment.&lt;/p&gt;
&lt;p&gt;The rebound trigger was the Fed&amp;rsquo;s 50bp cut on September 18, 2024. The fear was rates and recession. The Fed directly addressed that fear.&lt;/p&gt;
&lt;h3 id="9-2-february-2025-tariff-fear-required-a-policy-reversal"&gt;9-2. February 2025: tariff fear required a policy reversal
&lt;/h3&gt;&lt;p&gt;The February 2025 post-print decline later deepened because of tariff shock. NVIDIA moved toward a 12-month low in early April.&lt;/p&gt;
&lt;p&gt;The rebound trigger was the 90-day tariff pause. A policy shock created the fear; a policy reversal removed it.&lt;/p&gt;
&lt;h3 id="9-3-february-2026-gtc-addressed-ai-capex-peak-fear"&gt;9-3. February 2026: GTC addressed AI CapEx peak fear
&lt;/h3&gt;&lt;p&gt;This is the most relevant anchor for Samsung.&lt;/p&gt;
&lt;p&gt;NVIDIA&amp;rsquo;s Q4 FY26 print was strong, but the market worried that AI investment might peak in 2026. Hyperscaler concentration, AI return-on-investment questions, and customer spending durability all mattered.&lt;/p&gt;
&lt;p&gt;The rebound catalyst was GTC 2026. Jensen Huang said NVIDIA had visibility into at least $1T of demand or orders through 2027, compared with the prior year&amp;rsquo;s $500B figure through 2026. Data Center Dynamics and Constellation Research reported the comments.&lt;/p&gt;
&lt;p&gt;This directly addressed the fear. The market worried that AI CapEx was peaking; NVIDIA answered that demand through 2027 was much larger.&lt;/p&gt;
&lt;p&gt;Even then, the bottom came after GTC, not on GTC day. The stock bottomed on March 30, about two weeks later.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="10-applying-the-lesson-to-samsung"&gt;10. Applying the lesson to Samsung
&lt;/h2&gt;&lt;p&gt;Samsung&amp;rsquo;s selloff fear has two parts.&lt;/p&gt;
&lt;p&gt;First, memory pricing and HBM demand may have been priced too quickly.&lt;/p&gt;
&lt;p&gt;Second, if hyperscalers slow AI infrastructure spending, memory pricing and volume could both weaken.&lt;/p&gt;
&lt;p&gt;Therefore, the rebound does not need another reminder that 2Q earnings were good. The market already knows that. The rebound needs evidence that the fear is wrong.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Samsung selloff fear&lt;/th&gt;
 &lt;th&gt;Required catalyst&lt;/th&gt;
 &lt;th&gt;Timing&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Big Tech AI CapEx slowdown&lt;/td&gt;
 &lt;td&gt;Alphabet, Microsoft, Meta, and Amazon maintaining or raising 2026 to 2027 CapEx&lt;/td&gt;
 &lt;td&gt;late-July 2026 earnings calls&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Memory price-growth slowdown&lt;/td&gt;
 &lt;td&gt;DRAM/NAND spot and contract-price momentum staying firm&lt;/td&gt;
 &lt;td&gt;weekly checks and 3Q negotiations&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM catch-up doubt&lt;/td&gt;
 &lt;td&gt;HBM4/HBM4E qualification, production shipments, NVIDIA mix&lt;/td&gt;
 &lt;td&gt;2H26 IR and earnings calls&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2Q one-off cost confusion&lt;/td&gt;
 &lt;td&gt;clean 3Q operating-profit path after bonus/provision adjustments&lt;/td&gt;
 &lt;td&gt;3Q preview and 3Q earnings&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;flow-driven liquidation&lt;/td&gt;
 &lt;td&gt;foreign and program-selling slowdown, lower leverage-product volatility&lt;/td&gt;
 &lt;td&gt;daily flow checks&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The most important gate is late-July Big Tech earnings. If Alphabet, Microsoft, Meta, and Amazon keep or raise AI infrastructure spending plans, Samsung&amp;rsquo;s selloff can be reclassified as a flow correction rather than demand damage.&lt;/p&gt;
&lt;p&gt;If two or more hyperscalers talk about slowing the pace of AI infrastructure investment, the NVIDIA Q4 FY26 template becomes more relevant and the correction can last longer.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="11-why-the-fed-put-is-weaker-this-time"&gt;11. Why the Fed put is weaker this time
&lt;/h2&gt;&lt;p&gt;Two of NVIDIA&amp;rsquo;s prior recoveries had macro or policy support.&lt;/p&gt;
&lt;p&gt;In August 2024, the Fed cut rates by 50bp.&lt;/p&gt;
&lt;p&gt;In early 2025, a tariff pause reversed a policy shock.&lt;/p&gt;
&lt;p&gt;In February 2026, GTC provided a company-driven demand reset.&lt;/p&gt;
&lt;p&gt;Samsung now has less help from the first two channels. Since June 2026, strong US labor data and inflation concern have made rates a constraint again. The market is not confidently pricing a near-term Fed rescue.&lt;/p&gt;
&lt;p&gt;That makes the third channel more important: direct demand confirmation. Late-July hyperscaler CapEx commentary can play the same role for Samsung that GTC played for NVIDIA in March 2026.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="12-practical-view-wait-for-proof-then-scale-in"&gt;12. Practical view: wait for proof, then scale in
&lt;/h2&gt;&lt;p&gt;The conclusion is not a strong sell call. It is also not an automatic buy call.&lt;/p&gt;
&lt;p&gt;The stance is Wait.&lt;/p&gt;
&lt;p&gt;A strong print followed by a two-day selloff can be a signal that the market has moved from current profit level to next growth rate. That is a different regime.&lt;/p&gt;
&lt;p&gt;The entry gates are:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Gate&lt;/th&gt;
 &lt;th&gt;Meaning&lt;/th&gt;
 &lt;th&gt;If confirmed&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;late-July Big Tech CapEx maintained or raised&lt;/td&gt;
 &lt;td&gt;final demand for AI memory remains intact&lt;/td&gt;
 &lt;td&gt;selloff fear weakens&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM/NAND price momentum holds&lt;/td&gt;
 &lt;td&gt;2Q pricing logic continues into 3Q&lt;/td&gt;
 &lt;td&gt;3Q earnings support improves&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung clean 3Q QoQ remains healthy&lt;/td&gt;
 &lt;td&gt;underlying earnings path survives provision noise&lt;/td&gt;
 &lt;td&gt;quality of earnings improves&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;If two of these three gates pass, the case for staged entry improves. If all three weaken, the correction can extend.&lt;/p&gt;
&lt;p&gt;The main mistake would be to assume that record earnings automatically lead to a bounce. NVIDIA&amp;rsquo;s history says otherwise. The rebound required a catalyst that killed the specific fear behind the selloff.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="13-samsung-versus-sk-hynix"&gt;13. Samsung versus SK hynix
&lt;/h2&gt;&lt;p&gt;This note focuses on Samsung, but the read-through also matters for SK hynix.&lt;/p&gt;
&lt;p&gt;Samsung has a huge earnings base, a lower multiple, and HBM4/HBM4E catch-up optionality. But investors also have to deal with HBM purity, foundry and S.LSI losses, bonus provisions, and foreign-flow pressure.&lt;/p&gt;
&lt;p&gt;SK hynix has cleaner HBM exposure and a stronger position in NVIDIA&amp;rsquo;s supply chain. But it is already priced as the winner. If Big Tech CapEx remains strong, SK hynix is the trend-reacceleration candidate, while Samsung is the re-rating candidate where the Micron/NVIDIA read-through has been partly erased.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Lens&lt;/th&gt;
 &lt;th&gt;Samsung Electronics&lt;/th&gt;
 &lt;th&gt;SK hynix&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Strength&lt;/td&gt;
 &lt;td&gt;earnings level, low multiple, HBM catch-up option&lt;/td&gt;
 &lt;td&gt;HBM purity, customer trust, leadership position&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Burden&lt;/td&gt;
 &lt;td&gt;HBM certification, foundry/S.LSI, provision interpretation&lt;/td&gt;
 &lt;td&gt;high expectations, crowded-winner risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Rebound condition&lt;/td&gt;
 &lt;td&gt;clean 3Q earnings and HBM customer proof&lt;/td&gt;
 &lt;td&gt;Big Tech CapEx durability and HBM allocation&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Style&lt;/td&gt;
 &lt;td&gt;under-reflected alpha candidate&lt;/td&gt;
 &lt;td&gt;trend re-acceleration candidate&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Both depend on the same final question: will hyperscalers keep spending?&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="14-conclusion-samsung-needs-better-disproof-not-just-better-earnings"&gt;14. Conclusion: Samsung needs better disproof, not just better earnings
&lt;/h2&gt;&lt;p&gt;Samsung&amp;rsquo;s 2Q26 preliminary earnings were strong. But the stock fell sharply for two straight days. Compared with 17 NVIDIA post-earnings reactions, Samsung&amp;rsquo;s path looks much closer to NVIDIA Q4 FY26 in February 2026 than to the August 2024 one-day sell-on.&lt;/p&gt;
&lt;p&gt;In that NVIDIA case, the company posted strong numbers but failed to recover the pre-announcement price for about a month. The recovery began only after GTC 2026 directly challenged the AI CapEx peak fear with a $1T through-2027 demand statement.&lt;/p&gt;
&lt;p&gt;Samsung&amp;rsquo;s equivalent proof point is not another reminder that 2Q earnings were strong. It is hyperscaler confirmation that AI infrastructure spending remains firm into 2027.&lt;/p&gt;
&lt;p&gt;That is why the late-July Big Tech earnings calls matter. Until then, the right discipline is to wait for evidence rather than buying only because the stock has fallen after a record print.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="evidence-classification"&gt;Evidence classification
&lt;/h2&gt;&lt;h3 id="fact"&gt;Fact
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Evidence&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung 2Q26 preliminary earnings&lt;/td&gt;
 &lt;td&gt;Revenue about KRW 171T and operating profit about KRW 89.4T, Samsung official newsroom, 2026-07-07&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung comparison periods&lt;/td&gt;
 &lt;td&gt;1Q26 revenue KRW 133.87T and operating profit KRW 57.23T; 2Q25 revenue KRW 74.57T and operating profit KRW 4.68T, Samsung official newsroom&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung price path&lt;/td&gt;
 &lt;td&gt;July 6 KRW 318,000, July 7 KRW 296,000, July 8 KRW 277,500, yfinance daily close calculation&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung two-day reaction&lt;/td&gt;
 &lt;td&gt;July 7 -6.92%, July 8 -6.25%, two-day -12.74%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NVIDIA Q4 FY26 event&lt;/td&gt;
 &lt;td&gt;February 25, 2026 event confirmed on NVIDIA IR&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NVIDIA Q1 FY27&lt;/td&gt;
 &lt;td&gt;Revenue $81.6B, up 85% YoY and 20% QoQ, NVIDIA IR financial reports&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;GTC 2026 $1T comment&lt;/td&gt;
 &lt;td&gt;Demand/order visibility through 2027 reported by Data Center Dynamics and Constellation Research&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NVIDIA price-path calculations&lt;/td&gt;
 &lt;td&gt;yfinance daily closes, 17 quarters aligned with next-day reaction for after-market releases&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h3 id="inference"&gt;Inference
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung/NVIDIA comparison&lt;/td&gt;
 &lt;td&gt;This is a price-path and fear comparison, not a business-model comparison.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q4 FY26 similarity&lt;/td&gt;
 &lt;td&gt;Two-day path is the closest, and the fear set also overlaps: expectations, CapEx durability, and customer concentration.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Late-July Big Tech calls&lt;/td&gt;
 &lt;td&gt;If AI CapEx concern drove the Samsung selloff, hyperscaler guidance is the most direct disproof.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Wait stance&lt;/td&gt;
 &lt;td&gt;The print was strong, but the stock is asking about growth-rate durability. Confirmation should come before aggressive buying.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h3 id="speculation"&gt;Speculation
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Possible path&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Positive path&lt;/td&gt;
 &lt;td&gt;Big Tech CapEx holds or rises, DRAM/NAND price momentum holds, and Samsung&amp;rsquo;s clean 3Q earnings path improves.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Negative path&lt;/td&gt;
 &lt;td&gt;Two or more hyperscalers slow AI infrastructure spending or memory price momentum weakens.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Recovery timing&lt;/td&gt;
 &lt;td&gt;NVIDIA analogues suggest breakeven can take weeks, not days. For Samsung, mid-August to September is more realistic if catalysts turn positive.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h3 id="blocked"&gt;Blocked
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Why it is blocked&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung 2Q26 segment detail&lt;/td&gt;
 &lt;td&gt;Preliminary earnings do not yet disclose DS, foundry, or S.LSI detail.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Late-July hyperscaler CapEx guidance&lt;/td&gt;
 &lt;td&gt;Not yet reported.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung clean 3Q QoQ&lt;/td&gt;
 &lt;td&gt;Requires later earnings and IR detail after provision adjustments.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Exact consensus beat for all 17 NVIDIA quarters&lt;/td&gt;
 &lt;td&gt;Some quarters are verified, but a uniform consensus data set for every quarter was not fully obtained.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="sources-and-calculation-notes"&gt;Sources and calculation notes
&lt;/h2&gt;&lt;ol&gt;
&lt;li&gt;&lt;a class="link" href="https://news.samsung.com/global/samsung-electronics-announces-earnings-guidance-for-second-quarter-2026" target="_blank" rel="noopener"
 &gt;Samsung Electronics, Earnings Guidance for Second Quarter 2026&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://investor.nvidia.com/financial-info/financial-reports/default.aspx" target="_blank" rel="noopener"
 &gt;NVIDIA Investor Relations, Financial Reports&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://investor.nvidia.com/events-and-presentations/events-and-presentations/event-details/2026/NVIDIA-4th-Quarter-FY26-Financial-Results-2026-sO6kGS3C2P/default.aspx" target="_blank" rel="noopener"
 &gt;NVIDIA Investor Relations, 4th Quarter FY26 Financial Results&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.datacenterdynamics.com/en/news/nvidia-ceo-jensen-huang-says-company-has-one-trillion-dollars-in-orders-through-2027/" target="_blank" rel="noopener"
 &gt;Data Center Dynamics, Jensen Huang says NVIDIA has one trillion dollars in orders through 2027&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.constellationr.com/insights/news/nvidia-gtc-2026-ai-inference-fueling-demand-boom-1-trillion-order-flow" target="_blank" rel="noopener"
 &gt;Constellation Research, Nvidia GTC 2026 and $1 trillion order flow&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;Stock calculations: yfinance daily closes for &lt;code&gt;005930.KS&lt;/code&gt; and &lt;code&gt;NVDA&lt;/code&gt;, from 2022-05-20 to 2026-07-08. For NVIDIA, after-market releases are aligned so the next regular trading day is reaction D0.&lt;/li&gt;
&lt;/ol&gt;</description></item><item><title>The Big Tech Financing Relay: AI Capex Has Not Been Cut, And The Memory Bottleneck Looks Stronger</title><link>https://koreainvestinsights.com/post/hyperscaler-financing-race-ai-capex-memory-bottleneck-2026-07-07/</link><pubDate>Tue, 07 Jul 2026 21:45:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/hyperscaler-financing-race-ai-capex-memory-bottleneck-2026-07-07/</guid><description>&lt;p&gt;This is a follow-up to the late-July Big Tech earnings-call memory-thesis preview, the Micron FY3Q26 review, the 2027 semiconductor consensus-payability work, and the AI data-center capex bottleneck series.&lt;/p&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;p&gt;The Big Tech financing relay says one thing clearly: AI infrastructure capex has not been cut. If anything, it has outgrown the old assumption that elite hyperscalers could fund everything internally.&lt;/p&gt;
&lt;p&gt;Top hyperscaler 2026 capex guidance now sits around $695bn to $750bn. That compares with about $410bn in 2025 and about $245bn in 2024, implying roughly three consecutive years of 60% plus growth. The more important change is funding mix. The path has moved from internal cash to bonds, multi-currency bonds, off-balance-sheet structures, and now equity issuance.&lt;/p&gt;
&lt;p&gt;For memory, the implication is cleaner than for hyperscaler equities. Microsoft has tied roughly $25bn of its 2026 capex to component-price inflation. Meta has cited component pricing, especially memory, as one reason for higher 2026 capex. Amazon has also warned about component and memory-cost pressure. The buyers are absorbing the price burden. Memory suppliers receive it.&lt;/p&gt;
&lt;p&gt;The public-equity read-through is therefore not “own hyperscalers simply because they are spending.” It is more specific: watch the memory and HBM bottleneck nodes that can turn that spending into ASP, margin, and long-term contract visibility.&lt;/p&gt;
&lt;h2 id="the-financing-timeline"&gt;The Financing Timeline
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Timing&lt;/th&gt;
 &lt;th&gt;Issuer&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Type&lt;/th&gt;
 &lt;th&gt;Size&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;2025-10-30&lt;/td&gt;
 &lt;td&gt;Meta&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Bonds&lt;/td&gt;
 &lt;td&gt;$30bn&lt;/td&gt;
 &lt;td&gt;Long-term AI and data-center funding&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2025-11&lt;/td&gt;
 &lt;td&gt;Amazon&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Bonds&lt;/td&gt;
 &lt;td&gt;about $15bn&lt;/td&gt;
 &lt;td&gt;First major return to US dollar bonds in years&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2025-11&lt;/td&gt;
 &lt;td&gt;Alphabet&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Bonds&lt;/td&gt;
 &lt;td&gt;about $17.5bn&lt;/td&gt;
 &lt;td&gt;Pre-funding AI infrastructure&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-02&lt;/td&gt;
 &lt;td&gt;Alphabet&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Bonds&lt;/td&gt;
 &lt;td&gt;about $32bn&lt;/td&gt;
 &lt;td&gt;Multi-currency deal including a 100-year tech bond&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-02&lt;/td&gt;
 &lt;td&gt;Oracle&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Debt-led funding&lt;/td&gt;
 &lt;td&gt;$25bn&lt;/td&gt;
 &lt;td&gt;Part of a larger cloud-infrastructure funding plan&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-03-10/11&lt;/td&gt;
 &lt;td&gt;Amazon&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Bonds&lt;/td&gt;
 &lt;td&gt;about $54bn&lt;/td&gt;
 &lt;td&gt;Dollar plus euro record-scale funding&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-04-30&lt;/td&gt;
 &lt;td&gt;Meta&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Bonds&lt;/td&gt;
 &lt;td&gt;$25bn&lt;/td&gt;
 &lt;td&gt;Issued just after Q1 results and higher capex guide&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-05-12&lt;/td&gt;
 &lt;td&gt;Amazon&lt;/td&gt;
 &lt;td style="text-align: right"&gt;CHF bonds&lt;/td&gt;
 &lt;td&gt;CHF 2.82bn&lt;/td&gt;
 &lt;td&gt;Currency diversification&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-06-01/03&lt;/td&gt;
 &lt;td&gt;Alphabet&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Equity&lt;/td&gt;
 &lt;td&gt;$84.75bn&lt;/td&gt;
 &lt;td&gt;The decisive signal: dilution accepted for AI compute&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-07-07&lt;/td&gt;
 &lt;td&gt;Amazon&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Reported bond launch&lt;/td&gt;
 &lt;td&gt;$25bn plus&lt;/td&gt;
 &lt;td&gt;Reported eight-tranche US IG deal. Final terms not yet confirmed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The July 7 Amazon deal should not be mixed with the March record financing. As of publication, the final July 7 size, tranche spreads, new-issue concession and order book were not confirmed in public sources. It is therefore treated as a reported latest financing step, not a fully settled fact.&lt;/p&gt;
&lt;h2 id="capex-is-still-rising"&gt;Capex Is Still Rising
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Company&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2026E capex&lt;/th&gt;
 &lt;th style="text-align: right"&gt;YoY&lt;/th&gt;
 &lt;th&gt;Key point&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Amazon&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about $200bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about +50%&lt;/td&gt;
 &lt;td&gt;AWS, custom chips and data-center buildout&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Microsoft&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about $190bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+61%&lt;/td&gt;
 &lt;td&gt;Roughly $25bn attributed to component-price inflation&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Alphabet&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$180bn to $190bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about 2x&lt;/td&gt;
 &lt;td&gt;2027 expected to rise substantially again&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Meta&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$125bn to $145bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about 2x&lt;/td&gt;
 &lt;td&gt;Memory and component pricing cited as pressure points&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Oracle&lt;/td&gt;
 &lt;td style="text-align: right"&gt;volatile&lt;/td&gt;
 &lt;td style="text-align: right"&gt;high&lt;/td&gt;
 &lt;td&gt;Cloud buildout is highly capital-intensive&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Top 5&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$695bn to $750bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about +67%&lt;/td&gt;
 &lt;td&gt;Versus about $410bn in 2025 and $245bn in 2024&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The absolute dollar amount is still rising. The risk is the second derivative: even if 2027 capex reaches $1tn, the growth rate will likely slow from the 2025-2026 pace. That matters more for hyperscaler equity multiples than for memory suppliers. If memory keeps taking a larger share of the server bill of materials, memory revenue can still grow faster than total capex.&lt;/p&gt;
&lt;h2 id="why-funding-mix-matters"&gt;Why Funding Mix Matters
&lt;/h2&gt;&lt;p&gt;The escalation path is visible:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Internal cash.&lt;/li&gt;
&lt;li&gt;Straight bonds.&lt;/li&gt;
&lt;li&gt;Multi-currency bonds.&lt;/li&gt;
&lt;li&gt;SPVs, private credit and off-balance-sheet structures.&lt;/li&gt;
&lt;li&gt;Equity issuance.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;Alphabet&amp;rsquo;s June 2026 equity raise is the key tell. This is not a weak company. It is one of the strongest cash-flow machines in the world. If it chooses dilution to fund AI compute, the capital intensity of the AI buildout has moved beyond the old internal-cash model.&lt;/p&gt;
&lt;p&gt;Microsoft looks different. Its visible public benchmark funding has been less aggressive than Alphabet&amp;rsquo;s or Amazon&amp;rsquo;s, which may point to stronger internal funding capacity rather than lower AI capex ambition.&lt;/p&gt;
&lt;h2 id="the-memory-read-through"&gt;The Memory Read-Through
&lt;/h2&gt;&lt;p&gt;The key change is not simply that memory prices are up. It is that hyperscalers are not cutting demand in response. They are increasing budgets, signing long-term contracts and raising capital.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Signal&lt;/th&gt;
 &lt;th&gt;Memory implication&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Microsoft component-inflation comment&lt;/td&gt;
 &lt;td&gt;More dollars are needed for similar capacity&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Meta memory-price comment&lt;/td&gt;
 &lt;td&gt;Memory is moving the whole capex budget&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Amazon component-cost warning&lt;/td&gt;
 &lt;td&gt;Costs are higher, but capex continues&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron SCAs&lt;/td&gt;
 &lt;td&gt;Customers are locking in long-term supply&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM4/HBM4E transition&lt;/td&gt;
 &lt;td&gt;Higher value per accelerator and more wafer intensity&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This is why hyperscaler equities and memory suppliers should be separated. Hyperscalers bear the cost, FCF drag and payback risk. Memory suppliers receive the pricing and allocation benefit.&lt;/p&gt;
&lt;h2 id="public-equity-checklist"&gt;Public Equity Checklist
&lt;/h2&gt;&lt;p&gt;For Samsung Electronics, the questions are DS core profit, HBM4/HBM4E customer qualification, 3Q DRAM and NAND pricing, and the scale of foundry/System LSI losses.&lt;/p&gt;
&lt;p&gt;For SK Hynix, the questions are HBM4 long-term contract structure, 2027 allocation, and whether US-listed access through the SKHY ADR broadens the investor base without distorting common-share pricing.&lt;/p&gt;
&lt;p&gt;For Micron, the question is whether the SCA structure becomes the benchmark for how investors value memory-cycle durability.&lt;/p&gt;
&lt;p&gt;For hyperscalers, the clean stance is more cautious: demand is real, but FCF, debt, dilution and payback period risk are also real.&lt;/p&gt;
&lt;h2 id="scenarios"&gt;Scenarios
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Scenario&lt;/th&gt;
 &lt;th&gt;Condition&lt;/th&gt;
 &lt;th&gt;Memory read-through&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Bull&lt;/td&gt;
 &lt;td&gt;Amazon order book strong, late-July calls maintain or lift 2027 capex, memory costs remain absorbed&lt;/td&gt;
 &lt;td&gt;HBM and server DRAM pricing durability strengthens&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Base&lt;/td&gt;
 &lt;td&gt;2026 capex maintained, 2027 growth continues but slows, memory-price increases moderate&lt;/td&gt;
 &lt;td&gt;Memory thesis remains intact but stock reactions stay selective&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Bear&lt;/td&gt;
 &lt;td&gt;Capex guides cut, payback concerns rise, component inflation causes deployment delays&lt;/td&gt;
 &lt;td&gt;ASP thesis weakens until contract pricing is confirmed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Structural damage&lt;/td&gt;
 &lt;td&gt;Customers reduce memory content per server, HBM contract prices fall, bond demand weakens&lt;/td&gt;
 &lt;td&gt;Memory thesis needs full re-underwriting&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="final-view"&gt;Final View
&lt;/h2&gt;&lt;p&gt;The Big Tech financing relay confirms that AI infrastructure demand remains real. But it is not a simple hyperscaler-equity bullish signal. The more precise read-through is that memory is one of the bottlenecks taking a larger share of AI infrastructure dollars.&lt;/p&gt;
&lt;p&gt;The next tests are the final terms of Amazon&amp;rsquo;s July 7 deal, the late-July Big Tech calls, Samsung&amp;rsquo;s July 30 earnings call, SK Hynix&amp;rsquo;s 2Q call, and 3Q/4Q memory contract-price data.&lt;/p&gt;
&lt;h2 id="sources"&gt;Sources
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/1652044/000119312526251733/d160205dfwp.htm" target="_blank" rel="noopener"
 &gt;Alphabet proposed equity raise, SEC FWP&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://abc.xyz/investor/news/news-details/2026/Alphabet-Announces-Upsize-and-Pricing-of-84-75-Billion-Equity-Capital-Raise-to-Expand-AI-Infrastructure--and-Compute-2026-QzN3D9yMAj/default.aspx" target="_blank" rel="noopener"
 &gt;Alphabet upsized $84.75bn equity raise&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://blog.google/alphabet/investor-presentation-june-2026/" target="_blank" rel="noopener"
 &gt;Alphabet investor presentation&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://ir.aboutamazon.com/news-release/news-release-details/2026/Amazon-com-Announces-First-Quarter-Results/default.aspx" target="_blank" rel="noopener"
 &gt;Amazon Q1 2026 results&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/1018724/000110465926026393/tm266670d7_fwp.htm" target="_blank" rel="noopener"
 &gt;Amazon March 2026 SEC FWP&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.investing.com/news/sec-filings/amazon-closes-1447-billion-multitranche-eurodenominated-bond-sale-93CH-4564195" target="_blank" rel="noopener"
 &gt;Amazon euro notes filing report&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.tomshardware.com/tech-industry/big-tech/big-techs-ai-spending-plans-reach-725-billion" target="_blank" rel="noopener"
 &gt;Big Tech AI capex spending plans&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://investors.micron.com/quarterly-results" target="_blank" rel="noopener"
 &gt;Micron quarterly results&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;</description></item><item><title>Late-July Big Tech Earnings Calls and the Memory Thesis Scenario Map</title><link>https://koreainvestinsights.com/post/big-tech-july-earnings-call-memory-thesis-scenarios-2026-07-07/</link><pubDate>Tue, 07 Jul 2026 20:59:50 +0900</pubDate><guid>https://koreainvestinsights.com/post/big-tech-july-earnings-call-memory-thesis-scenarios-2026-07-07/</guid><description>
 &lt;blockquote&gt;
 &lt;p&gt;Context
This is a follow-up to &lt;a class="link" href="https://koreainvestinsights.com/post/micron-fy3q26-ai-memory-sca-fcf-hold-2026-06-25/" &gt;Micron FY3Q26 earnings&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/samsung-2q26-preview-micron-surprise-erased-core-op-hbm-2026-06-29/" &gt;Samsung Electronics 2Q26 preview&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/semiconductor-2027-earnings-hyperscaler-payability-memory-nvidia-2026-06-21/" &gt;Who pays for 2027 semiconductor consensus?&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/nvidia-earnings-elasticity-hbm-cycle-samsung-hynix-2026-06-28/" &gt;NVIDIA&amp;rsquo;s elasticity as a memory-cycle guide&lt;/a&gt; and &lt;a class="link" href="https://koreainvestinsights.com/post/h1-2026-ai-infra-bottleneck-korea-narrow-market-postmortem-2026-06-30/" &gt;the 1H26 AI infrastructure bottleneck review&lt;/a&gt;. Related hubs: &lt;a class="link" href="https://koreainvestinsights.com/page/exclusive-analysis-hub/" &gt;Exclusive Analysis&lt;/a&gt; and &lt;a class="link" href="https://koreainvestinsights.com/page/korea-semiconductor-hbm-kospi-hub/" &gt;AI HBM Hub&lt;/a&gt;.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;p&gt;The late-July earnings calls from Alphabet, Microsoft, Meta and Amazon are the next stress test for the memory thesis. My probability map is: strengthening 50%, unchanged 35%, weakening 15%. Within the weakening bucket, true structural damage is only around 4%.&lt;/p&gt;
&lt;p&gt;The key question is not whether Big Tech cuts capex. The four companies are still constrained by compute, power and component availability. The real questions are narrower: whether 2027 capex falls meaningfully short of the market&amp;rsquo;s roughly $1 trillion expectation, and whether memory pricing plus memory&amp;rsquo;s share of server BOM stay intact.&lt;/p&gt;
&lt;p&gt;The execution rule is simple: do not chase before the calls. Wait for the checklist. Korean memory stocks have already risen sharply over the last year, so the upside from good news can be smaller than the downside from a bad surprise.&lt;/p&gt;
&lt;h2 id="samsungs-july-7-update"&gt;Samsung&amp;rsquo;s July 7 Update
&lt;/h2&gt;&lt;p&gt;Samsung Electronics released its 2Q26 preliminary result on July 7. It guided to revenue of about KRW 171 trillion and operating profit of about KRW 89.4 trillion. The official ranges were KRW 170-172 trillion for revenue and KRW 89.3-89.5 trillion for operating profit. Source: &lt;a class="link" href="https://news.samsung.com/global/samsung-electronics-announces-earnings-guidance-for-second-quarter-2026" target="_blank" rel="noopener"
 &gt;Samsung Electronics 2Q26 earnings guidance&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;That is broadly in line with the earlier reported OP range of KRW 87-92 trillion. It does not settle the debate. The important event is the July 30 Samsung earnings call, where investors need to separate reported OP from core semiconductor profit, check 3Q DRAM and NAND pricing tone, and listen for HBM4/HBM4E customer qualification and allocation comments.&lt;/p&gt;
&lt;h2 id="what-the-event-judges"&gt;What the Event Judges
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Local timing&lt;/th&gt;
 &lt;th&gt;Event&lt;/th&gt;
 &lt;th&gt;Status&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;July 7&lt;/td&gt;
 &lt;td&gt;Samsung Electronics 2Q26 preliminary result&lt;/td&gt;
 &lt;td&gt;Officially released&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Around July 22-23 US&lt;/td&gt;
 &lt;td&gt;Alphabet 2Q26 call&lt;/td&gt;
 &lt;td&gt;Aggregator estimate, official IR date still needs confirmation&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Around July 28 US&lt;/td&gt;
 &lt;td&gt;Microsoft FY4Q26 call&lt;/td&gt;
 &lt;td&gt;Aggregator estimate, official IR page still needs confirmation&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Late July&lt;/td&gt;
 &lt;td&gt;SK Hynix 2Q26 result&lt;/td&gt;
 &lt;td&gt;Aggregators conflict, official page needs confirmation&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Around July 29 US&lt;/td&gt;
 &lt;td&gt;Meta 2Q26 call&lt;/td&gt;
 &lt;td&gt;Aggregator estimate&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Around July 30 US&lt;/td&gt;
 &lt;td&gt;Amazon 2Q26 call&lt;/td&gt;
 &lt;td&gt;Aggregator estimate&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;August&lt;/td&gt;
 &lt;td&gt;SK Hynix Nasdaq ADR listing&lt;/td&gt;
 &lt;td&gt;Pending SEC process&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The headline question is not &amp;ldquo;will Big Tech cut spending?&amp;rdquo; The real question is whether 2027 spending growth is still steep enough to support the memory earnings path.&lt;/p&gt;
&lt;h2 id="big-tech-capex-baseline"&gt;Big Tech Capex Baseline
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Company&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2025 actual&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2026 plan&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Growth&lt;/th&gt;
 &lt;th&gt;2027 commentary&lt;/th&gt;
 &lt;th&gt;Demand evidence&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Alphabet&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$91.4B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$180-190B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 2x&lt;/td&gt;
 &lt;td&gt;CFO said 2027 spending should rise materially above 2026&lt;/td&gt;
 &lt;td&gt;Cloud backlog of $462B, roughly doubled sequentially&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Microsoft&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About $118B, inferred&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About $190B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+61%&lt;/td&gt;
 &lt;td&gt;Double-digit revenue and operating income growth, but no explicit capex figure&lt;/td&gt;
 &lt;td&gt;RPO $627B, +99% YoY&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Meta&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$72B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$125-145B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+87.5% at midpoint&lt;/td&gt;
 &lt;td&gt;No number, flexible language&lt;/td&gt;
 &lt;td&gt;Contractual commitments rose $107B in 1Q&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Amazon&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$125B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About $200B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+60%&lt;/td&gt;
 &lt;td&gt;No number&lt;/td&gt;
 &lt;td&gt;AWS backlog $364B, Anthropic contract separate&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Total&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About $410B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About $710B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+73%&lt;/td&gt;
 &lt;td&gt;Market expectation above $1T, +41%&lt;/td&gt;
 &lt;td&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Alphabet is the strongest setup. It raised 2026 capex guidance, said 2027 would be meaningfully higher, and later announced an $80B capital raise to expand AI infrastructure and compute. Microsoft is the key swing factor: RPO is huge, but sequential RPO growth needs to reaccelerate. Meta is two-sided: it raised guidance because of component and memory costs, but it also kept flexibility language for 2027. Amazon has the largest absolute capex plan, but its free cash flow pressure makes its tone worth watching.&lt;/p&gt;
&lt;h2 id="memory-price-facts"&gt;Memory Price Facts
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Quarter&lt;/th&gt;
 &lt;th style="text-align: right"&gt;DRAM&lt;/th&gt;
 &lt;th style="text-align: right"&gt;NAND&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;1Q26&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+55-60%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+33-38%&lt;/td&gt;
 &lt;td&gt;Server DRAM was even stronger&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2Q26&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+58-63%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+70-75%&lt;/td&gt;
 &lt;td&gt;Peak rate of increase&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;3Q26 forecast&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+13-18%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+10-15%&lt;/td&gt;
 &lt;td&gt;Still rising, but at a slower pace&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;4Q26 reference&lt;/td&gt;
 &lt;td style="text-align: right"&gt;PC DRAM +3-8%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Not specified&lt;/td&gt;
 &lt;td&gt;A consumer-side reference point&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Source: &lt;a class="link" href="https://www.trendforce.com/presscenter/news/20260703-13134.html" target="_blank" rel="noopener"
 &gt;TrendForce 3Q26 price outlook&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;The 3Q deceleration is not a demand collapse. It reflects base effects, consumer affordability limits and a rising share of long-term contracts. The more important point is that server memory remains tight. If Big Tech capex grows more slowly while memory takes a larger share of server BOM, memory revenue can still grow faster than aggregate capex.&lt;/p&gt;
&lt;h2 id="micron-sca-matters"&gt;Micron SCA Matters
&lt;/h2&gt;&lt;p&gt;Micron disclosed 16 strategic customer agreements in its FY3Q26 materials. The contracts run from 2026 through 2030, include take-or-pay structures, and cover about 20% of DRAM volume and roughly one-third of NAND volume. Minimum contracted revenue is around $100B, and customer deposits plus commitments are around $22B. Source: &lt;a class="link" href="https://investors.micron.com/static-files/631b1a32-5537-46ae-8f40-82e42fc79dfe" target="_blank" rel="noopener"
 &gt;Micron prepared remarks&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;This is the most important underappreciated change in the cycle. Big Tech is not merely accepting higher memory prices. It is signing multi-year, non-cancelable supply assurance contracts. That lowers the historical cyclicality discount for memory earnings. It does not eliminate risk, because price ceilings may cap upside on contracted volume, but it changes the downside profile.&lt;/p&gt;
&lt;p&gt;For SK Hynix, the key question is whether its long-term contracts have similar floors without the same upside caps. That detail is not officially confirmed. It should be the first question on the SK Hynix 2Q call.&lt;/p&gt;
&lt;h2 id="scenario-map"&gt;Scenario Map
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Scenario&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Probability&lt;/th&gt;
 &lt;th&gt;Meaning for memory stocks&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Strengthening&lt;/td&gt;
 &lt;td style="text-align: right"&gt;50%&lt;/td&gt;
 &lt;td&gt;Earnings revision cycle extends. Upside depends on how explicit 2027 capex becomes.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Neutral&lt;/td&gt;
 &lt;td style="text-align: right"&gt;35%&lt;/td&gt;
 &lt;td&gt;2026 plans remain intact, but 2027 language stays cautious. Short-term headline noise likely.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Weakening&lt;/td&gt;
 &lt;td style="text-align: right"&gt;15%&lt;/td&gt;
 &lt;td&gt;Split into narrative weakening and structural damage.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Narrative weakening&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 11%&lt;/td&gt;
 &lt;td&gt;Stocks correct, but price and contract data stay healthy. Could become an entry window.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Structural damage&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 4%&lt;/td&gt;
 &lt;td&gt;Thesis reset. Examples: 2026 capex cut, memory-content redesign, digestion language from multiple companies.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="call-by-call-update-rules"&gt;Call-by-Call Update Rules
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Sequence&lt;/th&gt;
 &lt;th&gt;Result&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Strengthening probability update&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Alphabet&lt;/td&gt;
 &lt;td&gt;Plan maintained, 2027 reaffirmed, backlog up&lt;/td&gt;
 &lt;td style="text-align: right"&gt;50 → 55&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Alphabet&lt;/td&gt;
 &lt;td&gt;2027 language retreats or backlog stalls&lt;/td&gt;
 &lt;td style="text-align: right"&gt;50 → below 40&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Microsoft&lt;/td&gt;
 &lt;td&gt;$190B plan maintained, supply tightness, 2027 growth tone, RPO growth resumes&lt;/td&gt;
 &lt;td style="text-align: right"&gt;55 → 62&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Microsoft&lt;/td&gt;
 &lt;td&gt;Normalization or single-digit growth tone&lt;/td&gt;
 &lt;td style="text-align: right"&gt;55 → 45&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Meta&lt;/td&gt;
 &lt;td&gt;Plan maintained, commitments rise, cloud monetization becomes official&lt;/td&gt;
 &lt;td style="text-align: right"&gt;62 → 66&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Meta&lt;/td&gt;
 &lt;td&gt;April-style flexibility language returns with numbers&lt;/td&gt;
 &lt;td style="text-align: right"&gt;62 → 50&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Amazon&lt;/td&gt;
 &lt;td&gt;$200B maintained, memory price absorbed&lt;/td&gt;
 &lt;td style="text-align: right"&gt;66 → 68-70&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Amazon&lt;/td&gt;
 &lt;td&gt;Amazon-only pacing comments&lt;/td&gt;
 &lt;td style="text-align: right"&gt;66 → 60, not enough to kill the thesis&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;If two or more companies cut plans or use digestion language, the table no longer applies. That becomes a thesis reset exercise.&lt;/p&gt;
&lt;h2 id="korean-memory-checklist"&gt;Korean Memory Checklist
&lt;/h2&gt;&lt;p&gt;This section is a public research checklist, not a trading instruction. The late-July Big Tech calls matter here because they help test the Korean memory thesis.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Name&lt;/th&gt;
 &lt;th&gt;Role&lt;/th&gt;
 &lt;th&gt;What to verify&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td&gt;Korean memory core with DRAM and NAND volume leverage&lt;/td&gt;
 &lt;td&gt;DS core profit, 3Q memory price assumptions and HBM4 customer qualification.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td&gt;HBM purity and long-term contract structure&lt;/td&gt;
 &lt;td&gt;HBM shipment, ASP durability, contract pricing structure and HBM4 allocation.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="phrases-to-listen-for"&gt;Phrases To Listen For
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Direction&lt;/th&gt;
 &lt;th&gt;Phrases&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Positive&lt;/td&gt;
 &lt;td&gt;supply cannot keep up with demand / customer commitments rising / backlog rising / component or memory price pressure absorbed by budgets / 2027 capex expansion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Negative&lt;/td&gt;
 &lt;td&gt;digestion phase / lower utilization / investment pacing / optimization / delayed customer commitments / easing component price pressure / memory supply improving earlier than expected / memory content redesign&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="kill-conditions"&gt;Kill Conditions
&lt;/h2&gt;&lt;p&gt;One item triggers reassessment. Two or more start the thesis-reset process.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Any of the four companies lowers 2026 capex plans.&lt;/li&gt;
&lt;li&gt;Microsoft RPO or Alphabet Cloud backlog turns negative sequentially.&lt;/li&gt;
&lt;li&gt;Companies use &amp;ldquo;digestion&amp;rdquo; or &amp;ldquo;pacing purchases to power availability&amp;rdquo; language.&lt;/li&gt;
&lt;li&gt;A concrete server redesign reduces memory content per server because of price.&lt;/li&gt;
&lt;li&gt;TrendForce cuts 3Q memory price growth back to single digits while Samsung DS core profit disappoints.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;The fourth item is the real kill condition. Supply-driven forced reductions are different from price-driven voluntary memory-content cuts.&lt;/p&gt;
&lt;h2 id="evidence-classification"&gt;Evidence Classification
&lt;/h2&gt;&lt;p&gt;Facts: Samsung&amp;rsquo;s 2Q26 preliminary result, Samsung&amp;rsquo;s July 30 call schedule, Alphabet&amp;rsquo;s 2026 capex range and AI infrastructure capital raise, Microsoft&amp;rsquo;s $190B capex run-rate and $627B RPO, Meta&amp;rsquo;s $125-145B plan and $107B commitment increase, Amazon&amp;rsquo;s roughly $200B plan and AWS backlog, TrendForce&amp;rsquo;s 3Q price outlook, Micron&amp;rsquo;s FY3Q26 SCA disclosures.&lt;/p&gt;
&lt;p&gt;Inferences: the 50/35/15 scenario map, the idea that +73% to +41% capex deceleration is already embedded, the view that Meta&amp;rsquo;s cloud report is more monetization than capex-cut signal, and the view that long-term contracts reduce memory cyclicality.&lt;/p&gt;
&lt;p&gt;Speculation: SK Hynix contract upside without price ceilings, possible Microsoft financing, the exact timing of a 2027 digestion phase, and Meta&amp;rsquo;s cloud economics.&lt;/p&gt;
&lt;p&gt;Blocked: verified 2027 consensus multiples for Samsung and SK Hynix, Marvell&amp;rsquo;s exact share in Trainium 3 and 4, SK Hynix contract details, company-level 2027 capex consensus distributions, memory supplier allocation by hyperscaler and some official late-July earnings dates.&lt;/p&gt;
&lt;p&gt;The conclusion is practical: the odds favor a stronger memory thesis, but price already reflects part of that. The right way to trade the event is to update probabilities as evidence arrives, not to make a blind prediction before the calls.&lt;/p&gt;</description></item><item><title>SK Square Investment Memo: When Does SK Hynix Cash Flow Become SK Square Cash?</title><link>https://koreainvestinsights.com/post/sk-square-hynix-cash-flow-nav-discount-listed-ai-semi-allocator-2026-07-04/</link><pubDate>Sat, 04 Jul 2026 21:40:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/sk-square-hynix-cash-flow-nav-discount-listed-ai-semi-allocator-2026-07-04/</guid><description>
 &lt;blockquote&gt;
 &lt;p&gt;Context
This memo follows &lt;a class="link" href="https://koreainvestinsights.com/post/sk-hynix-f1a-f6-skhy-adr-hbm-capex-2026-07-04/" &gt;SK Hynix F-1/A and F-6&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/sk-hynix-adr-000660-price-path-leverage-etf-plumbing-2026-07-04/" &gt;SK Hynix ADR price path&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/sk-hynix-hbm-market-share-ai-memory-demand-2026/" &gt;SK Hynix HBM market share 2026&lt;/a&gt; and &lt;a class="link" href="https://koreainvestinsights.com/post/micron-fy3q26-ai-memory-sca-fcf-hold-2026-06-25/" &gt;Micron FY3Q26 earnings&lt;/a&gt;.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;p&gt;SK Square should not be framed as the investment vehicle for the entire SK Group. SK Inc. is the official group holding company and portfolio coordinator. For public-market investors, SK Square is better defined as:&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;A listed AI and semiconductor capital allocator built on SK Hynix equity value, NAV discount compression and a call option on future SK Hynix special cash dividends.&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;The key variable is not SK Hynix net income by itself. Higher SK Hynix earnings lift SK Square&amp;rsquo;s equity-method income and NAV, but they do not automatically become SK Square cash. Cash reaches SK Square through SK Hynix dividends, stake sales, collateralization or SK Square&amp;rsquo;s own portfolio exits.&lt;/p&gt;
&lt;p&gt;Current view: &lt;strong&gt;Wait / Conditional Buy&lt;/strong&gt;. The buy case strengthens when SK Hynix free cash flow converts into cash dividends, SK Square receives that cash, and the company allocates it to buyback cancellation or Hynix-linked AI and semiconductor investments.&lt;/p&gt;
&lt;h2 id="1-correct-framing"&gt;1. Correct Framing
&lt;/h2&gt;&lt;p&gt;SK Inc. is the official holding company for SK Group. It manages the broader group portfolio across energy, AI, semiconductors and other businesses. SK Square is different. It is an ICT and semiconductor investment company whose NAV is overwhelmingly driven by SK Hynix.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;SK Inc.&lt;/th&gt;
 &lt;th&gt;SK Square&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Role&lt;/td&gt;
 &lt;td&gt;Group holding company and portfolio manager&lt;/td&gt;
 &lt;td&gt;ICT and semiconductor investment company&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Main exposure&lt;/td&gt;
 &lt;td&gt;Broad SK Group portfolio&lt;/td&gt;
 &lt;td&gt;SK Hynix stake and AI/semi portfolio&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Investor thesis&lt;/td&gt;
 &lt;td&gt;Group restructuring and holding-company discount&lt;/td&gt;
 &lt;td&gt;SK Hynix NAV, cash conversion, NAV discount compression&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Main risk&lt;/td&gt;
 &lt;td&gt;Conglomerate complexity&lt;/td&gt;
 &lt;td&gt;Hynix dividends may not convert into SK Square cash&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This distinction matters. Calling SK Square the group&amp;rsquo;s treasury overstates the case. The cleaner thesis is a listed AI and semiconductor allocator tied to SK Hynix cash-flow conversion.&lt;/p&gt;
&lt;h2 id="2-nav-is-mostly-sk-hynix"&gt;2. NAV Is Mostly SK Hynix
&lt;/h2&gt;&lt;p&gt;SK Square&amp;rsquo;s official NAV page showed the following as of July 2, 2026.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Value&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Total NAV&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 324.7T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix stake value&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 319.52T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Net cash&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 0.79T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NAV per share&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 2,460,288&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Square share price&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 1,525,000&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NAV discount&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 38.0%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The discount is calculated as &lt;code&gt;1 - 1,525,000 / 2,460,288&lt;/code&gt;. SK Square holds 146,124,531 SK Hynix shares, or 20.5%.&lt;/p&gt;
&lt;h2 id="3-earnings-are-not-cash"&gt;3. Earnings Are Not Cash
&lt;/h2&gt;&lt;p&gt;SK Hynix earnings affect SK Square through accounting income and NAV. But SK Square needs actual cash for buybacks, cancellations, dividends or new investments.&lt;/p&gt;
&lt;p&gt;SK Hynix&amp;rsquo;s fixed dividend for 2025-2027 is KRW 1,500 per share. Applied to SK Square&amp;rsquo;s 146,124,531 shares, this equals KRW 219.2B, or about KRW 0.219T per year. That is about 28% of SK Square&amp;rsquo;s KRW 0.79T net cash. It is meaningful, but not enough to re-rate a holding company with KRW 319.52T of SK Hynix stake value.&lt;/p&gt;
&lt;p&gt;The real option is special or additional cash dividends.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th style="text-align: right"&gt;Total SK Hynix cash dividend&lt;/th&gt;
 &lt;th style="text-align: right"&gt;SK Square pre-tax receipt&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Estimated after-tax receipt&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;KRW 1T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 0.205T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 0.194T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;KRW 5T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 1.025T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 0.969T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;KRW 10T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 2.050T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 1.937T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;KRW 20T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 4.100T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About KRW 3.875T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The after-tax estimates use a simplified Korean dividend received deduction framework. A domestic parent owning at least 20% and less than 50% of a domestic subsidiary may exclude 80% of dividends from taxable income. Applying a 27.5% tax rate to the remaining 20% implies about 5.5% leakage.&lt;/p&gt;
&lt;h2 id="4-nav-discount-math"&gt;4. NAV Discount Math
&lt;/h2&gt;&lt;p&gt;Holding NAV per share constant at KRW 2,460,288, the implied share price changes materially as the discount narrows.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th style="text-align: right"&gt;NAV discount&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Implied price&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;50%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 1,230,144&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;45%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 1,353,158&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;40%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 1,476,173&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;38%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 1,525,379&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;35%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 1,599,187&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;30%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 1,722,202&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;25%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 1,845,216&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;SK Square has stated an objective of reducing its NAV discount to 30% or less by 2028. That is a target, not a guarantee. The market needs evidence: Hynix cash dividends, SK Square buyback cancellations, non-core exits and disciplined AI/semi investments.&lt;/p&gt;
&lt;h2 id="5-shareholder-return-and-capital-allocation"&gt;5. Shareholder Return And Capital Allocation
&lt;/h2&gt;&lt;p&gt;SK Square&amp;rsquo;s 2023-2025 shareholder-return policy uses at least 30% of recurring dividend income and part of meaningful investment gains for shareholder returns. Methods include buybacks, cancellations and cash dividends.&lt;/p&gt;
&lt;p&gt;At a high NAV discount, buyback cancellation is powerful. If a company trades at a 38% NAV discount, buying back stock is economically similar to buying KRW 1 of NAV for KRW 0.62. That means low-return investments are hard to justify unless they have clear Hynix-linked strategic value.&lt;/p&gt;
&lt;h2 id="6-us-cash-is-not-automatically-trapped-cash"&gt;6. U.S. Cash Is Not Automatically Trapped Cash
&lt;/h2&gt;&lt;p&gt;SK Hynix&amp;rsquo;s U.S. customer revenue is large. Reported 2025 U.S. customer revenue was KRW 66.885T, about 69% of total revenue. SK hynix America reportedly generated KRW 58.693T of sales and KRW 235.9B of net income, implying a net margin of about 0.40%.&lt;/p&gt;
&lt;p&gt;That low margin suggests the U.S. sales entity may be a limited-margin distributor rather than the place where most economic profit accumulates. Cash can return to the Korean parent through product payment, transfer-pricing settlement, intercompany payments, dividends, royalties or other structures.&lt;/p&gt;
&lt;p&gt;The important question for SK Square is not whether U.S. revenue is large. It is whether SK Hynix parent-level free cash flow grows and becomes cash dividends.&lt;/p&gt;
&lt;h2 id="7-skhy-adr-read-through"&gt;7. SKHY ADR Read-Through
&lt;/h2&gt;&lt;p&gt;SK Hynix is pursuing a U.S. ADR listing. Reuters reported that the company may raise up to $29.4B, with proceeds for Korean semiconductor factories and EUV equipment.&lt;/p&gt;
&lt;p&gt;For SK Square, the ADR is positive for investor access and potentially for SK Hynix&amp;rsquo;s valuation. But it is not a direct cash-in event for SK Square. The proceeds go to SK Hynix. SK Square benefits through NAV, valuation and future dividends, while also facing potential ownership dilution from new share issuance.&lt;/p&gt;
&lt;h2 id="8-investment-ideas"&gt;8. Investment Ideas
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Idea&lt;/th&gt;
 &lt;th&gt;Thesis&lt;/th&gt;
 &lt;th&gt;Required proof&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Hynix FCF-to-cash option&lt;/td&gt;
 &lt;td&gt;Hynix earnings matter only when they become SK Square cash&lt;/td&gt;
 &lt;td&gt;FCF after capex and meaningful cash dividends&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Listed AI/semi allocator&lt;/td&gt;
 &lt;td&gt;SK Square becomes a disciplined Hynix-linked investment platform&lt;/td&gt;
 &lt;td&gt;Investments tied to Hynix ecosystem, not generic AI hype&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NAV discount compression&lt;/td&gt;
 &lt;td&gt;Discount narrows from about 38% toward 30%&lt;/td&gt;
 &lt;td&gt;Buyback cancellation, non-core exits, capital discipline&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;U.S. trapped-cash misunderstanding&lt;/td&gt;
 &lt;td&gt;Large U.S. revenue does not mean large trapped U.S. profit&lt;/td&gt;
 &lt;td&gt;Parent-level FCF and tax-efficient cash flow&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="9-practical-view"&gt;9. Practical View
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Security&lt;/th&gt;
 &lt;th&gt;View&lt;/th&gt;
 &lt;th&gt;Reason&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Square&lt;/td&gt;
 &lt;td&gt;Wait / Conditional Buy&lt;/td&gt;
 &lt;td&gt;NAV discount is meaningful, but cash conversion and capital allocation need confirmation&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td&gt;Watchlist&lt;/td&gt;
 &lt;td&gt;Source of cash flow and NAV&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Inc.&lt;/td&gt;
 &lt;td&gt;Separate thesis&lt;/td&gt;
 &lt;td&gt;Official holding company, but less direct Hynix cash-flow exposure&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Entry conditions:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Condition&lt;/th&gt;
 &lt;th&gt;What to see&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Wider discount&lt;/td&gt;
 &lt;td&gt;Updated NAV discount at or above 45% while the Hynix HBM thesis remains intact&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cash conversion&lt;/td&gt;
 &lt;td&gt;Hynix special cash dividend or early additional shareholder return, followed by SK Square buyback cancellation or Hynix-synergy investment&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Allocator trust&lt;/td&gt;
 &lt;td&gt;Non-core exits, AI/semi discipline, no group-support capital allocation&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Invalidation:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Risk&lt;/th&gt;
 &lt;th&gt;Meaning&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Hynix earnings are strong but FCF is absorbed by capex and working capital&lt;/td&gt;
 &lt;td&gt;No SK Square cash conversion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Hynix returns capital mainly through buybacks rather than cash dividends&lt;/td&gt;
 &lt;td&gt;Limited SK Square cash inflow&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Square uses Hynix dividends for group support&lt;/td&gt;
 &lt;td&gt;Holding-company discount widens&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Low-return investments take priority over buyback cancellation&lt;/td&gt;
 &lt;td&gt;Capital allocation trust declines&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;ADR dilution outweighs lower cost of capital&lt;/td&gt;
 &lt;td&gt;SK Square&amp;rsquo;s Hynix stake economics weaken&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="final-view"&gt;Final View
&lt;/h2&gt;&lt;p&gt;SK Square is not cash-rich today. Official net cash is KRW 0.79T. But it owns KRW 319.52T of SK Hynix stake value. If SK Hynix AI-memory earnings convert into free cash flow and cash dividends, SK Square can receive trillions of won over time.&lt;/p&gt;
&lt;p&gt;The strongest thesis is not “SK Hynix is good, therefore SK Square is good.” It is more specific: SK Hynix FCF must become cash dividends, SK Square must allocate that cash to buyback cancellation and Hynix-linked investments, and the NAV discount must compress structurally.&lt;/p&gt;
&lt;h2 id="sources"&gt;Sources
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://www.sksquare.com/eng/ir/nav.do" target="_blank" rel="noopener"
 &gt;SK Square Shareholding Structure &amp;amp; NAV&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.sksquare.com/eng/ir/return.do" target="_blank" rel="noopener"
 &gt;SK Square Shareholder Return&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.skhynix.com/ir/UI-FR-IR04/" target="_blank" rel="noopener"
 &gt;SK hynix Shareholder Status&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.skhynix.com/ir/UI-FR-IR05/" target="_blank" rel="noopener"
 &gt;SK hynix Shareholder Return&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.sk-inc.com/en/main/mainpage.aspx" target="_blank" rel="noopener"
 &gt;SK Inc.&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://taxsummaries.pwc.com/republic-of-korea/corporate/income-determination" target="_blank" rel="noopener"
 &gt;PwC Korea corporate income determination&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.irs.gov/pub/irs-trty/korea.pdf" target="_blank" rel="noopener"
 &gt;U.S.-Korea tax treaty PDF&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.reuters.com/world/asia-pacific/south-koreas-sk-hynix-says-raise-29-bln-us-adr-listing-2026-06-24/" target="_blank" rel="noopener"
 &gt;Reuters: SK Hynix targets up to $29.4bn U.S. ADR listing&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://news.skhynix.com/sk-hynix-announces-fy25-financial-results/" target="_blank" rel="noopener"
 &gt;SK hynix FY25 financial results&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.asiae.co.kr/en/article/2026031717471236785" target="_blank" rel="noopener"
 &gt;AsiaE: SK hynix U.S. revenue coverage&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;em&gt;Disclaimer: For research and information purposes only. Not investment advice.&lt;/em&gt;&lt;/p&gt;</description></item><item><title>How SK Hynix's ADR Could Change 000660 Price Discovery And Leveraged ETF Flows</title><link>https://koreainvestinsights.com/post/sk-hynix-adr-000660-price-path-leverage-etf-plumbing-2026-07-04/</link><pubDate>Sat, 04 Jul 2026 07:18:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/sk-hynix-adr-000660-price-path-leverage-etf-plumbing-2026-07-04/</guid><description>
 &lt;blockquote&gt;
 &lt;p&gt;Context
This is a follow-up to &lt;a class="link" href="https://koreainvestinsights.com/post/sk-hynix-f1a-f6-skhy-adr-hbm-capex-2026-07-04/" &gt;SK Hynix F-1/A and F-6: how the SKHY ADR packages HBM capex&lt;/a&gt;. The previous note focused on the filing mechanics. This note focuses on price transmission, ADR-common-share arbitrage and single-stock leveraged ETF flows.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;p&gt;The SK Hynix ADR is structurally positive over the medium term, but it is not a near-term volatility dampener. It widens U.S. investor access and may allow SK Hynix to be compared more directly with U.S.-listed AI memory peers. But right after listing, the key variables are the offer discount, the first week of ADR trading quality, the ADR premium or discount to the KRX common share and the impact on Korea&amp;rsquo;s single-stock leveraged ETF plumbing.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Question&lt;/th&gt;
 &lt;th&gt;View&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Is the ADR positive for 000660?&lt;/td&gt;
 &lt;td&gt;Medium-term yes, because investor access and peer framing improve.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Is it an immediate buy signal?&lt;/td&gt;
 &lt;td&gt;No. Pricing, discount and initial premium or discount must be checked first.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;What is the key mechanism?&lt;/td&gt;
 &lt;td&gt;ADR-common-share arbitrage. If the ADR trades rich to implied parity, it can create KRX common-share buying pressure. If it trades cheap, it can pressure the common share.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Can the ADR absorb Korea&amp;rsquo;s 2x ETF overheating?&lt;/td&gt;
 &lt;td&gt;Not in the short run. The bottleneck is KRX close rebalancing, not U.S. access.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Current stance&lt;/td&gt;
 &lt;td&gt;Watch the common share and the ADR. Avoid leveraged wrappers if the goal is HBM exposure.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="1-what-is-confirmed"&gt;1. What Is Confirmed
&lt;/h2&gt;&lt;p&gt;SK Hynix has filed an F-1/A for a planned Nasdaq Global Select Market ADS listing under &lt;code&gt;SKHY&lt;/code&gt;.&lt;sup id="fnref:1"&gt;&lt;a href="#fn:1" class="footnote-ref" role="doc-noteref"&gt;1&lt;/a&gt;&lt;/sup&gt; The F-6 filed by Citibank states that each ADS represents the right to receive one-tenth of one common share.&lt;sup id="fnref:2"&gt;&lt;a href="#fn:2" class="footnote-ref" role="doc-noteref"&gt;2&lt;/a&gt;&lt;/sup&gt; Press reports indicate a maximum new-share issuance of 17.79 million common shares, around 2.5% of existing common shares.&lt;sup id="fnref:3"&gt;&lt;a href="#fn:3" class="footnote-ref" role="doc-noteref"&gt;3&lt;/a&gt;&lt;/sup&gt;&lt;/p&gt;
&lt;p&gt;The final ADS count, offer price and net proceeds remain blank in the F-1/A. That matters. The investment signal is not merely that an ADR is coming. The signal will be in the final price, the discount to the Korean common share and the first few days of trading.&lt;/p&gt;
&lt;h2 id="2-access-premium"&gt;2. Access Premium
&lt;/h2&gt;&lt;p&gt;The first effect is access. U.S. investors can buy SK Hynix exposure in dollars, during U.S. hours, without using a Korea trading account or indirect ETF exposure. That can shift the comparison set from “Korean memory stock” to “U.S.-tradable AI memory and HBM supplier.”&lt;/p&gt;
&lt;p&gt;This is positive, but not new information for the stock. Much of the obvious ADR rerating argument has already been anticipated. The remaining alpha is whether U.S. investors pay at or above implied KRX parity after the ADR starts trading.&lt;/p&gt;
&lt;h2 id="3-new-shares-and-dilution"&gt;3. New Shares And Dilution
&lt;/h2&gt;&lt;p&gt;The ADR is linked to new share issuance, not merely a secondary sale by existing holders. The headline maximum is about 2.5% of existing shares. That is dilution in ownership terms, but not necessarily large economic dilution if the offer price is close to market and proceeds are invested into HBM, eSSD, advanced packaging and EUV-related capacity.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Scenario&lt;/th&gt;
 &lt;th&gt;Read-through&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Small discount and strong order book&lt;/td&gt;
 &lt;td&gt;U.S. investors validate the common-share price. Positive for 000660.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Large discount and weak demand&lt;/td&gt;
 &lt;td&gt;U.S. market rejects the prevailing valuation. Negative for 000660.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;ADR premium after listing&lt;/td&gt;
 &lt;td&gt;U.S. demand can support KRX common shares.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Persistent ADR discount&lt;/td&gt;
 &lt;td&gt;U.S. trading becomes a new downside anchor.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="4-adr-common-share-arbitrage"&gt;4. ADR-Common-Share Arbitrage
&lt;/h2&gt;&lt;p&gt;With one ADS representing 0.1 common share, parity is simple:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Calculation&lt;/th&gt;
 &lt;th&gt;Formula&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Implied ADR price from KRX common share&lt;/td&gt;
 &lt;td&gt;&lt;code&gt;KRX common price × 0.1 ÷ USD/KRW&lt;/code&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Implied common-share price from ADR&lt;/td&gt;
 &lt;td&gt;&lt;code&gt;ADR price × USD/KRW × 10&lt;/code&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;If the ADR trades above implied parity, arbitrageurs may buy KRX common shares and sell or create ADR exposure. That creates common-share buying pressure. If the ADR trades below parity, arbitrageurs may buy ADRs and sell common-share exposure, pressuring the KRX line.&lt;/p&gt;
&lt;p&gt;Therefore the first week is not about the absolute ADR move. It is about the premium or discount versus the Korean common share, after conversion costs, FX, settlement and stock-borrow constraints.&lt;/p&gt;
&lt;h2 id="5-us-price-discovery"&gt;5. U.S. Price Discovery
&lt;/h2&gt;&lt;p&gt;The ADR extends SK Hynix price discovery into U.S. trading hours. This helps incorporate Nvidia, Micron, SOX, rates and AI capex news faster. But it also imports U.S. overnight gap risk into the Korean open. The ADR is a price-discovery channel, not automatically a volatility cushion.&lt;/p&gt;
&lt;h2 id="6-leveraged-etf-plumbing"&gt;6. Leveraged ETF Plumbing
&lt;/h2&gt;&lt;p&gt;The ADR does not directly remove Korea&amp;rsquo;s single-stock 2x ETF rebalancing pressure. Korean 2x ETFs must maintain daily target exposure through common shares, futures, swaps and liquidity provider hedging around the KRX close. The Korean Financial Services Commission described the daily 2x structure when single-stock leveraged products were introduced.&lt;sup id="fnref:4"&gt;&lt;a href="#fn:4" class="footnote-ref" role="doc-noteref"&gt;4&lt;/a&gt;&lt;/sup&gt; The Korea Capital Market Institute estimated SK Hynix leveraged ETF AUM at KRW 9.15tn as of June 19 and highlighted how daily rebalancing can amplify volatility.&lt;sup id="fnref:5"&gt;&lt;a href="#fn:5" class="footnote-ref" role="doc-noteref"&gt;5&lt;/a&gt;&lt;/sup&gt;&lt;/p&gt;
&lt;p&gt;The core issue is mechanical. When SK Hynix rises, 2x ETF exposure may need to increase. When it falls, exposure may need to decrease. That creates a short-gamma-like flow. An ADR only offsets this if U.S. long-only ADR demand is large enough to replace the domestic leveraged flow. That is possible later, but not proven now.&lt;/p&gt;
&lt;h2 id="7-what-to-watch"&gt;7. What To Watch
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Watch item&lt;/th&gt;
 &lt;th&gt;Why it matters&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Final offer price and discount&lt;/td&gt;
 &lt;td&gt;Tests U.S. demand quality.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;First 3 to 5 days of ADR premium or discount&lt;/td&gt;
 &lt;td&gt;Determines whether U.S. trading supports or pressures 000660.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;ADR trading value&lt;/td&gt;
 &lt;td&gt;Shows whether SKHY can become a real price-discovery line.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Korean leveraged ETF AUM&lt;/td&gt;
 &lt;td&gt;Shows whether the domestic short-gamma pipe is cooling.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM ASP, shipment and capex commentary&lt;/td&gt;
 &lt;td&gt;Keeps the ADR story grounded in earnings.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="final-view"&gt;Final View
&lt;/h2&gt;&lt;p&gt;SKHY is a major capital-market event for SK Hynix. It can broaden the investor base and reframe the company as a U.S.-tradable AI memory supplier. But the ADR does not immediately solve the domestic leveraged ETF problem. The near-term stance is watchlist for the common share and ADR, and avoidance of leveraged wrappers if the goal is clean HBM exposure.&lt;/p&gt;
&lt;div class="footnotes" role="doc-endnotes"&gt;
&lt;hr&gt;
&lt;ol&gt;
&lt;li id="fn:1"&gt;
&lt;p&gt;&lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/2120882/000119312526289345/d32785df1a.htm" target="_blank" rel="noopener"
 &gt;SEC, SK Hynix Form F-1/A, filed June 30, 2026&lt;/a&gt;&amp;#160;&lt;a href="#fnref:1" class="footnote-backref" role="doc-backlink"&gt;&amp;#x21a9;&amp;#xfe0e;&lt;/a&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li id="fn:2"&gt;
&lt;p&gt;&lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/1472033/000119380526000898/e665622_f6-skhynix.htm" target="_blank" rel="noopener"
 &gt;SEC, SK Hynix Form F-6, filed July 1, 2026&lt;/a&gt;&amp;#160;&lt;a href="#fnref:2" class="footnote-backref" role="doc-backlink"&gt;&amp;#x21a9;&amp;#xfe0e;&lt;/a&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li id="fn:3"&gt;
&lt;p&gt;&lt;a class="link" href="https://www.thebell.co.kr/front/newsview.asp?key=202606241720222280104920" target="_blank" rel="noopener"
 &gt;The Bell, SK Hynix Nasdaq ADR listing report&lt;/a&gt;&amp;#160;&lt;a href="#fnref:3" class="footnote-backref" role="doc-backlink"&gt;&amp;#x21a9;&amp;#xfe0e;&lt;/a&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li id="fn:4"&gt;
&lt;p&gt;&lt;a class="link" href="https://www.fsc.go.kr/no010107/86981" target="_blank" rel="noopener"
 &gt;Korea Financial Services Commission, single-stock leveraged product notice&lt;/a&gt;&amp;#160;&lt;a href="#fnref:4" class="footnote-backref" role="doc-backlink"&gt;&amp;#x21a9;&amp;#xfe0e;&lt;/a&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li id="fn:5"&gt;
&lt;p&gt;&lt;a class="link" href="https://www.kcmi.re.kr/publications/pub_detail_view?cno=6801&amp;amp;syear=2026&amp;amp;zcd=002001016&amp;amp;zno=1922" target="_blank" rel="noopener"
 &gt;Korea Capital Market Institute, single-stock leveraged and inverse ETF flows&lt;/a&gt;&amp;#160;&lt;a href="#fnref:5" class="footnote-backref" role="doc-backlink"&gt;&amp;#x21a9;&amp;#xfe0e;&lt;/a&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;
&lt;/div&gt;</description></item><item><title>SK Hynix F-1/A and F-6: What The SKHY ADR Says About HBM Capex</title><link>https://koreainvestinsights.com/post/sk-hynix-f1a-f6-skhy-adr-hbm-capex-2026-07-04/</link><pubDate>Sat, 04 Jul 2026 05:20:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/sk-hynix-f1a-f6-skhy-adr-hbm-capex-2026-07-04/</guid><description>
 &lt;blockquote&gt;
 &lt;p&gt;Context
This is a follow-up to &lt;a class="link" href="https://koreainvestinsights.com/post/kr-deep-dive-sk-hynix-2026-04-16/" &gt;SK Hynix: the HBM leader in the AI revolution&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/sk-hynix-hbm-market-share-ai-memory-demand-2026/" &gt;SK Hynix HBM market share 2026&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/sk-hynix-vs-micron-hbm-premium-ai-memory-platform-2026-05-31/" &gt;SK Hynix vs Micron&lt;/a&gt; and &lt;a class="link" href="https://koreainvestinsights.com/post/micron-fy3q26-ai-memory-sca-fcf-hold-2026-06-25/" &gt;Micron FY3Q26 earnings&lt;/a&gt;.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;p&gt;SK Hynix&amp;rsquo;s SEC filings are not just an ADR wrapper. The sequence from the June 24 F-1, the June 30 F-1/A and the July 1 F-6 shows the company packaging its HBM leadership, eSSD, Custom HBM and Korea capex roadmap for U.S. AI-memory investors.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Confirmed&lt;/th&gt;
 &lt;th&gt;Investor interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;F-1/A&lt;/td&gt;
 &lt;td&gt;Filed June 30, 2026. Nasdaq Global Select Market application under &lt;code&gt;SKHY&lt;/code&gt;. ADS count, price and net proceeds are still blank.&lt;/td&gt;
 &lt;td&gt;Not final terms yet. A further amendment or final prospectus is required.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;F-6&lt;/td&gt;
 &lt;td&gt;Filed July 1, 2026. Citibank is depositary. Each ADS represents one-tenth of one common share.&lt;/td&gt;
 &lt;td&gt;This fills the key ADR ratio that remained blank in the F-1/A cover.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;New-share ceiling&lt;/td&gt;
 &lt;td&gt;Board authorization for up to 17,790,000 new common shares, 2.50% of 712,702,365 existing shares.&lt;/td&gt;
 &lt;td&gt;Maximum dilution is limited, but actual issuance remains unknown.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The most important caveat: the F-6 registration of 1,780,000,000 ADSs is not the offering size. Form F-6 registers the ADR facility and the related depositary-fee calculation. The actual equity offering size depends on the final ADS count and offering price in the F-1/A or final prospectus.&lt;/p&gt;
&lt;p&gt;The stock read-through is Watchlist. This is an access and ownership-base event, not a fresh discovery of the HBM thesis. The alpha question is whether U.S. long-only and AI-memory investors use &lt;code&gt;SKHY&lt;/code&gt; as a cleaner vehicle for SK Hynix exposure.&lt;/p&gt;
&lt;h2 id="1-why-the-june-24-f-1-is-not-enough"&gt;1. Why The June 24 F-1 Is Not Enough
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Filing date&lt;/th&gt;
 &lt;th&gt;Form&lt;/th&gt;
 &lt;th&gt;Document&lt;/th&gt;
 &lt;th&gt;What changed&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-06-24&lt;/td&gt;
 &lt;td&gt;F-1&lt;/td&gt;
 &lt;td&gt;Registration No. 333-296987&lt;/td&gt;
 &lt;td&gt;Initial ADS offering and Nasdaq listing preparation; key offering terms blank&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-06-30&lt;/td&gt;
 &lt;td&gt;F-1/A&lt;/td&gt;
 &lt;td&gt;Amendment No. 1&lt;/td&gt;
 &lt;td&gt;Offering structure, new-share ceiling, use of proceeds and capex tables updated; ADS count and price still blank&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-07-01&lt;/td&gt;
 &lt;td&gt;F-6&lt;/td&gt;
 &lt;td&gt;Registration No. 333-297185&lt;/td&gt;
 &lt;td&gt;Citibank depositary and one ADS = one-tenth common share&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The company is SK hynix Inc. Its common shares trade on KOSPI under &lt;code&gt;000660&lt;/code&gt;. The ADSs are expected to trade on the Nasdaq Global Select Market under &lt;code&gt;SKHY&lt;/code&gt;, subject to effectiveness and listing approval.&lt;/p&gt;
&lt;h2 id="2-offering-mechanics"&gt;2. Offering Mechanics
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Latest status&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Issuer&lt;/td&gt;
 &lt;td&gt;SK hynix Inc.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;ADS ticker&lt;/td&gt;
 &lt;td&gt;&lt;code&gt;SKHY&lt;/code&gt;, Nasdaq Global Select Market application&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Depositary&lt;/td&gt;
 &lt;td&gt;Citibank, N.A.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Global coordinators&lt;/td&gt;
 &lt;td&gt;BofA Securities, Citigroup, Goldman Sachs, J.P. Morgan&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;ADS ratio&lt;/td&gt;
 &lt;td&gt;F-6: 1 ADS = 0.1 common share&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;ADS count&lt;/td&gt;
 &lt;td&gt;Blank in F-1/A&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Offering price&lt;/td&gt;
 &lt;td&gt;Blank in F-1/A&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Net proceeds&lt;/td&gt;
 &lt;td&gt;Blank in F-1/A&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Lock-up&lt;/td&gt;
 &lt;td&gt;90-day lock-up may apply to the company and certain affiliates, subject to exceptions&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The board authorized up to 17,790,000 new common shares. Against 712,702,365 existing shares, that is 2.50%. If fully issued, new shares would be about 2.44% of the post-issue share count. This is a ceiling, not the actual offering size.&lt;/p&gt;
&lt;h2 id="3-f-6-ads-registration-is-not-equity-proceeds"&gt;3. F-6 ADS Registration Is Not Equity Proceeds
&lt;/h2&gt;&lt;p&gt;The F-6 registers 1,780,000,000 ADSs, each representing one-tenth of one common share. It also uses a fee calculation of $5.00 per 100 ADSs, giving a proposed maximum aggregate fee-basis amount of $89,000,000 and a registration fee of $12,290.90.&lt;/p&gt;
&lt;p&gt;That fee basis is not the gross equity offering. Investors should not read 1.78 billion ADSs as the number of ADSs that will be sold in this offering. The real offering size remains blocked until the final ADS count and offer price are filed.&lt;/p&gt;
&lt;h2 id="4-use-of-proceeds-yongin-pt7-and-euv"&gt;4. Use Of Proceeds: Yongin, P&amp;amp;T7 And EUV
&lt;/h2&gt;&lt;p&gt;The F-1/A frames proceeds around general corporate purposes, including capital expenditures. The filing specifically discusses KRW 45.5 trillion of Korea production-facility capex and approximately KRW 11.9 trillion of EUV scanner cost, with EUV delivery expected by December 2027.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Project&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Target completion of planned investments&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Total expected cost&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Invested by May 31, 2026&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Additional planned investment&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Yongin Fab 1&lt;/td&gt;
 &lt;td style="text-align: right"&gt;End of 2030&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 31.0T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 4.4T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 26.6T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cheongju P&amp;amp;T7&lt;/td&gt;
 &lt;td style="text-align: right"&gt;End of 2030&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 19.0T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 0.1T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 18.9T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Total&lt;/td&gt;
 &lt;td style="text-align: right"&gt;End of 2030&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 50.0T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 4.5T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 45.5T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Project&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2026&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2027&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2028&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2029&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2030&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Total&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Yongin Fab 1&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 7.4T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 10.1T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 6.6T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 2.5T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 0.0T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 26.6T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cheongju P&amp;amp;T7&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 0.5T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 2.1T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 2.7T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 5.8T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 7.8T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 18.9T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Total&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 7.9T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 12.2T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 9.3T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 8.3T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 7.8T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 45.5T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The business section and capex table use different milestones. The business section points to the first Yongin cleanroom opening in 1Q27 and P&amp;amp;T7 construction completion by end-2027. The capex table targets completion of planned investments by end-2030. That is not necessarily a contradiction; it separates phase-opening from full investment completion and equipment installation.&lt;/p&gt;
&lt;h2 id="5-the-narrative-hbm-custom-hbm-and-essd"&gt;5. The Narrative: HBM, Custom HBM And eSSD
&lt;/h2&gt;&lt;p&gt;The F-1/A does not present SK Hynix as a plain DRAM-cycle company. It emphasizes HBM, server DRAM and eSSD as structural AI-infrastructure products.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Market&lt;/th&gt;
 &lt;th&gt;IDC-based position in 1Q26&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM including HBM&lt;/td&gt;
 &lt;td&gt;No. 2 by revenue, 29.1% share&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM&lt;/td&gt;
 &lt;td&gt;No. 1 by revenue, 56.4% share&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NAND flash&lt;/td&gt;
 &lt;td&gt;No. 2 by revenue, 18.5% share&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Custom HBM is especially important. The filing defines it as HBM that integrates certain GPU or ASIC functions into the HBM base die according to customer configuration. That shifts the story from memory pricing alone toward memory becoming part of the accelerator design stack.&lt;/p&gt;
&lt;h2 id="6-recent-financials"&gt;6. Recent Financials
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Metric&lt;/th&gt;
 &lt;th style="text-align: right"&gt;1Q25&lt;/th&gt;
 &lt;th style="text-align: right"&gt;1Q26&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Change&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 17.639T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 52.576T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+198.1%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Profit for the period&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 8.108T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 40.346T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~5.0x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Total assets&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 176.108T at 2025-end&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 222.829T at Mar. 31, 2026&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+26.5%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Total equity&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 120.667T at 2025-end&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 164.380T at Mar. 31, 2026&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+36.2%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Product&lt;/th&gt;
 &lt;th style="text-align: right"&gt;1Q26 revenue&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Share&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 40.659T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;77.3%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NAND&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 11.574T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;22.0%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Other&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 0.343T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0.7%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The U.S. accounted for KRW 33.999T, or 64.7% of 1Q26 revenue. That makes the ADR event strategically coherent: the investor access point is moving closer to the customer-revenue geography.&lt;/p&gt;
&lt;h2 id="7-shareholders-dividends-and-governance"&gt;7. Shareholders, Dividends And Governance
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Shareholder&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Shares&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Ownership&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Square&lt;/td&gt;
 &lt;td style="text-align: right"&gt;146,100,000&lt;/td&gt;
 &lt;td style="text-align: right"&gt;20.50%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;National Pension Service&lt;/td&gt;
 &lt;td style="text-align: right"&gt;57,439,774&lt;/td&gt;
 &lt;td style="text-align: right"&gt;8.06%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;BlackRock&lt;/td&gt;
 &lt;td style="text-align: right"&gt;36,407,157&lt;/td&gt;
 &lt;td style="text-align: right"&gt;5.11%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Capital Research and Management&lt;/td&gt;
 &lt;td style="text-align: right"&gt;25,149,374&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3.53%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Others&lt;/td&gt;
 &lt;td style="text-align: right"&gt;445,979,195&lt;/td&gt;
 &lt;td style="text-align: right"&gt;62.58%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Treasury shares&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1,626,865&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0.23%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The filing says the maximum offering size reflects the Korean Monopoly Regulation and Fair Trade Act requirement that SK Square maintain at least 20% ownership.&lt;/p&gt;
&lt;p&gt;The 2025-2027 base dividend policy is KRW 1,500 per share per year, paid quarterly at KRW 375. The 2025 total dividend was KRW 3,000 per share after additional annual distribution. This ADR story is not a dividend story; it is a capex and AI-memory access story.&lt;/p&gt;
&lt;h2 id="8-risks"&gt;8. Risks
&lt;/h2&gt;&lt;p&gt;The main risk factors remain memory cyclicality, HBM manufacturing complexity, AI infrastructure slowdown, customer concentration, U.S.-China restrictions, raw materials, water, power, equipment, foreign-exchange volatility and ADS-specific rights. ADS holders exercise voting and dividend rights through the depositary and can face conversion, cancellation or redeposit limitations.&lt;/p&gt;
&lt;h2 id="9-investment-view"&gt;9. Investment View
&lt;/h2&gt;&lt;p&gt;SKHY can make SK Hynix easier to own for U.S. AI-memory investors. It can also make the comparison with Micron&amp;rsquo;s U.S.-listed AI-memory premium more direct.&lt;/p&gt;
&lt;p&gt;But this is not yet an immediate buy signal. The final ADS count, offering price, net proceeds, effective date and first trading date remain unavailable. The event belongs on the watchlist until those terms are filed.&lt;/p&gt;
&lt;h2 id="watch-list"&gt;Watch List
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Watch item&lt;/th&gt;
 &lt;th&gt;Why it matters&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Next F-1/A or final prospectus&lt;/td&gt;
 &lt;td&gt;ADS count, offering price and net proceeds&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SEC effectiveness&lt;/td&gt;
 &lt;td&gt;Start of real offering timeline&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;First &lt;code&gt;SKHY&lt;/code&gt; trading date&lt;/td&gt;
 &lt;td&gt;U.S. investor access becomes observable&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KRX common vs ADS parity&lt;/td&gt;
 &lt;td&gt;Premium, discount or arbitrage pressure&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Offer price relative to KRX&lt;/td&gt;
 &lt;td&gt;Demand strength and near-term flow risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;U.S. AI-memory ETF inclusion&lt;/td&gt;
 &lt;td&gt;Potential structural demand&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2Q/3Q HBM ASP and shipments&lt;/td&gt;
 &lt;td&gt;Whether the filing narrative converts into earnings&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Custom HBM and eSSD commentary&lt;/td&gt;
 &lt;td&gt;Whether SK Hynix becomes a memory platform, not only an HBM supplier&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Yongin, P&amp;amp;T7 and Indiana milestones&lt;/td&gt;
 &lt;td&gt;Whether capex turns into supply capacity on schedule&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="sources"&gt;Sources
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/2120882/000119312526289345/d32785df1a.htm" target="_blank" rel="noopener"
 &gt;SK hynix Form F-1/A, June 30, 2026&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/2120882/000119312526289345/0001193125-26-289345-index.htm" target="_blank" rel="noopener"
 &gt;SK hynix Form F-1/A filing index&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/2120882/000119380526000898/e665622_f6-skhynix.htm" target="_blank" rel="noopener"
 &gt;SK hynix Form F-6, July 1, 2026&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/2120882/000119380526000898/0001193805-26-000898-index.htm" target="_blank" rel="noopener"
 &gt;SK hynix Form F-6 filing index&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/2120882/000119312526280172/d32785df1.htm" target="_blank" rel="noopener"
 &gt;SK hynix initial Form F-1, June 24, 2026&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;</description></item><item><title>FADU 2Q26 Earnings Preview: Likely Beat, Not a Mega Surprise</title><link>https://koreainvestinsights.com/post/fadu-2q26-earnings-preview-essd-controller-moderate-beat-2026-07-04/</link><pubDate>Sat, 04 Jul 2026 05:00:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/fadu-2q26-earnings-preview-essd-controller-moderate-beat-2026-07-04/</guid><description>
 &lt;blockquote&gt;
 &lt;p&gt;Context
This is a follow-up to &lt;a class="link" href="https://koreainvestinsights.com/post/fadu-sandisk-ai-storage-korea-beta-jeju-semiconductor-2026-05-31/" &gt;FADU as Korea&amp;rsquo;s Sandisk beta&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/fadu-ai-infra-storage-bottleneck-p-q-new-segment-2026-06-02/" &gt;FADU&amp;rsquo;s P, Q and new-segment test&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/korea-semiconductor-top50-two-month-alpha-candidates-samsung-hynix-2026-06-13/" &gt;Korean semiconductor Top 50 two-month alpha candidates&lt;/a&gt; and the &lt;a class="link" href="https://koreainvestinsights.com/post/hbf-hbc-commercialization-stock-trigger-catalyst-calendar-2026-06-28/" &gt;HBF/HBC commercialization calendar&lt;/a&gt;. FADU should be analyzed as an eSSD controller and AI-storage bottleneck name, not as a confirmed HBF beneficiary.&lt;/p&gt;

 &lt;/blockquote&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Result update, July 22, 2026
FADU&amp;rsquo;s preliminary Q2 result was KRW 73.16bn of revenue, KRW 16.27bn of operating profit and a 22.2% operating margin. Revenue fell just below this preview&amp;rsquo;s KRW 75bn bull threshold, while operating profit exceeded the KRW 15.4bn bull-case upper bound. The post-mortem and Gen6/QLC moat assessment continue in &lt;a class="link" href="https://koreainvestinsights.com/post/fadu-2q26-earnings-surprise-gen6-essd-controller-narrow-moat-2026-07-22/" &gt;FADU&amp;rsquo;s Q2 Surprise and Its Real Moat&lt;/a&gt;.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;p&gt;My FADU 2Q26 estimate is KRW 69-72B revenue and KRW 11-12.5B operating profit. The midpoint is KRW 70.5B revenue, KRW 11.8B operating profit and 16.7% operating margin.&lt;/p&gt;
&lt;p&gt;The local consensus snapshot as of July 3, 2026 shows 2Q26 revenue of KRW 60.5B and operating profit of KRW 9.3B. The midpoint estimate is 16.5% above revenue consensus and 26.9% above operating-profit consensus. A beat is likely, but this is not a mega-surprise bar.&lt;/p&gt;
&lt;p&gt;The key is not the gross amount of 2Q contract disclosures. Major supply contracts disclosed during the quarter total about KRW 150.8B. A mechanical pro-rata allocation would imply roughly KRW 35.4B of 2Q exposure. But amended disclosures in June pushed some contract end dates and payment terms later. Treating the full pro-rata amount as incremental 2Q revenue would be too aggressive. A more reasonable 2Q incremental recognition range is KRW 9.5-12.5B.&lt;/p&gt;
&lt;p&gt;The stock call is Wait. FADU closed at KRW 79,600 on July 3, 2026, implying roughly KRW 3.98T market cap. Against FY2026E revenue of KRW 301.5B and operating profit of KRW 44.8B, that is about 13.2x sales and 89x operating profit. At this valuation, a good 2Q print is not enough. Investors need evidence of second-half delivery normalization, backlog conversion and controller-mix durability.&lt;/p&gt;
&lt;h2 id="0-setup"&gt;0. Setup
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Detail&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Company&lt;/td&gt;
 &lt;td&gt;FADU&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Ticker&lt;/td&gt;
 &lt;td&gt;440110 / KOSDAQ&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Analysis date&lt;/td&gt;
 &lt;td&gt;2026-07-04 KST&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Price basis&lt;/td&gt;
 &lt;td&gt;July 3, 2026 close, KRW 79,600, local KR database&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Consensus basis&lt;/td&gt;
 &lt;td&gt;July 3, 2026 local consensus snapshot, 2Q26 revenue KRW 60.5B, OP KRW 9.3B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Main question&lt;/td&gt;
 &lt;td&gt;How much of the 2Q contract disclosures can actually be recognized as 2Q revenue?&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Stance&lt;/td&gt;
 &lt;td&gt;Wait; conditional entry after proof&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This preview uses public filings, public media reports and local market data. It does not use private management color.&lt;/p&gt;
&lt;h2 id="1-baseline-from-1q26"&gt;1. Baseline From 1Q26
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Metric&lt;/th&gt;
 &lt;th style="text-align: right"&gt;1Q26&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 59.54B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Operating profit&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 7.69B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Net income&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 10.21B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SSD controller revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 47.9B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SSD controller mix&lt;/td&gt;
 &lt;td style="text-align: right"&gt;80.4%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SSD finished-product revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 9.9B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SSD finished-product mix&lt;/td&gt;
 &lt;td style="text-align: right"&gt;16.6%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The important point is mix. FADU earns a higher-quality multiple only if it is seen as an eSSD controller and firmware bottleneck company, not as a finished-SSD assembler. The 1Q controller mix of 80.4% is therefore more important than the headline profit turn.&lt;/p&gt;
&lt;p&gt;The reported 1Q gross profit was KRW 34.65B. Against revenue of KRW 59.54B, this implies a 58.2% gross margin. Subtracting KRW 7.69B operating profit gives implied SG&amp;amp;A of about KRW 26.96B.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Formula&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Value&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;1Q gross margin&lt;/td&gt;
 &lt;td&gt;34.65 / 59.54&lt;/td&gt;
 &lt;td style="text-align: right"&gt;58.2%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Implied 1Q SG&amp;amp;A&lt;/td&gt;
 &lt;td&gt;34.65 - 7.69&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 26.96B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;1Q operating margin&lt;/td&gt;
 &lt;td&gt;7.69 / 59.54&lt;/td&gt;
 &lt;td style="text-align: right"&gt;12.9%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="2-backlog-and-2026-delivery-base"&gt;2. Backlog And 2026 Delivery Base
&lt;/h2&gt;&lt;p&gt;FADU&amp;rsquo;s 1Q-end backlog was reported at $116.208M, roughly KRW 175.9B. The 2026 delivery portion was $95.131M, roughly KRW 144.0B. Adding 1Q revenue gives a visible annual revenue base above KRW 200B, before the rest of the 2Q disclosures.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Amount&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;1Q-end backlog&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$116.208M&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KRW equivalent&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~KRW 175.9B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2026 delivery portion&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$95.131M&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KRW equivalent&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~KRW 144.0B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;1Q26 revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 59.54B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The caveat is timing. A 2026 delivery commitment does not mean 2Q recognition.&lt;/p&gt;
&lt;h2 id="3-2q-contract-disclosures"&gt;3. 2Q Contract Disclosures
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Disclosure timing&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Product&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Contract amount&lt;/th&gt;
 &lt;th&gt;Contract period&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Pro-rata 2Q exposure&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Apr 13&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Enterprise SSD controller&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 15.196B&lt;/td&gt;
 &lt;td&gt;Apr 10-Oct 9&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~KRW 6.8B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;May 6&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Enterprise SSD controller&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 50.012B&lt;/td&gt;
 &lt;td&gt;May 5-Dec 8 after amendment&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~KRW 13.1B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;May 22&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Enterprise SSD controller&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 28.684B&lt;/td&gt;
 &lt;td&gt;May 21-Dec 11&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~KRW 5.7B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;May 28&lt;/td&gt;
 &lt;td style="text-align: right"&gt;eSSD controller&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~KRW 46.5B&lt;/td&gt;
 &lt;td&gt;May 28-Jan 1, 2027&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~KRW 7.2B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;May 29&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Finished SSD&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 10.373B&lt;/td&gt;
 &lt;td&gt;May 28-Oct 11&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~KRW 2.6B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Total&lt;/td&gt;
 &lt;td style="text-align: right"&gt;&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~KRW 150.8B&lt;/td&gt;
 &lt;td&gt;&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~KRW 35.4B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The June amendments matter. The KRW 50B contract end date moved from Nov. 13 to Dec. 8, and payment terms moved from the next Friday after delivery to within 30 days after delivery. A separate KRW 20.3B controller contract was also extended from June 12 to July 17. These do not prove demand weakness, but they do weaken the 2Q cutoff.&lt;/p&gt;
&lt;h2 id="4-revenue-model"&gt;4. Revenue Model
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Component&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Estimate&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;1Q revenue baseline&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 59.5B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Incremental 2Q recognition from new contracts&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 9.5-12.5B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Finished products and other movement&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+/- KRW 1.0B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2Q26 revenue estimate&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 69-72B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Midpoint&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 70.5B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This is more conservative than the earlier June framework that looked for KRW 90-105B of 2Q revenue. The reason is new information: amended disclosures increased the risk that part of the revenue recognition slips into 3Q.&lt;/p&gt;
&lt;h2 id="5-operating-profit-model"&gt;5. Operating Profit Model
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Segment&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Revenue assumption&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Gross margin assumption&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Gross profit&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Controllers&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 57.5B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;61%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 35.1B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Finished SSD / modules&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 11.0B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;15%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 1.7B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Other&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 2.0B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;50%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 1.0B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Total&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 70.5B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~53.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~KRW 37.7B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Using KRW 25.9B of SG&amp;amp;A, the midpoint operating profit is about KRW 11.8B, or a 16.7% margin. The implied 2Q gross margin is below 1Q&amp;rsquo;s 58.2% to reflect finished-product mix and cutoff risk.&lt;/p&gt;
&lt;h2 id="6-scenario-table"&gt;6. Scenario Table
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Scenario&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Revenue&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Operating profit&lt;/th&gt;
 &lt;th style="text-align: right"&gt;OPM&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Bear&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 62-65B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 9-10.5B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;14.5-16.2%&lt;/td&gt;
 &lt;td&gt;In line to slight beat&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Base&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 69-72B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 11-12.5B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;15.9-17.4%&lt;/td&gt;
 &lt;td&gt;Clear beat&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Bull&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 78-83B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 14.5-17B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;18.6-20.5%&lt;/td&gt;
 &lt;td&gt;Major surprise&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Base is the central case. Bull requires faster shipment, limited impact from amendments and controller mix near 80%.&lt;/p&gt;
&lt;h2 id="7-consensus-bar"&gt;7. Consensus Bar
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Metric&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Consensus&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Midpoint estimate&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Delta&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Delta %&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 60.5B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 70.5B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 10.0B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+16.5%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Operating profit&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 9.3B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 11.8B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 2.5B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+26.9%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;OPM&lt;/td&gt;
 &lt;td style="text-align: right"&gt;15.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;16.7%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+1.3ppt&lt;/td&gt;
 &lt;td style="text-align: right"&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;A meaningful beat needs revenue above KRW 70B and OP above KRW 11.5B. A true estimate-reset print likely needs revenue above KRW 75B and OP above KRW 14B.&lt;/p&gt;
&lt;h2 id="8-price-flow-and-valuation"&gt;8. Price, Flow And Valuation
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Value&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;July 3 close&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 79,600&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;5D return&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-4.1%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;20D return&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-27.5%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;60D return&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+57.9%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Estimated market cap&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~KRW 3.98T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;20D foreign net buying&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~KRW 227.3B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;20D institutional net selling&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~KRW -117.6B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Foreign ownership&lt;/td&gt;
 &lt;td style="text-align: right"&gt;24.11% on Jun 8 to 28.25% on Jul 3&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Foreign flow is the positive signal. Institutional confirmation is not there yet. On valuation, FY2026E revenue of KRW 301.5B and OP of KRW 44.8B imply about 13.2x sales and 89x operating profit. This is not cheap. It requires continued proof.&lt;/p&gt;
&lt;h2 id="9-entry-catalysts-and-falsifiers"&gt;9. Entry, Catalysts And Falsifiers
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Action rule&lt;/th&gt;
 &lt;th&gt;Requirement&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Earnings-confirmation entry&lt;/td&gt;
 &lt;td&gt;2Q revenue above KRW 70B, OP above KRW 11.5B, controller mix near 80%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Strong-beat entry&lt;/td&gt;
 &lt;td&gt;Revenue above KRW 75B, OP above KRW 14B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Valuation entry&lt;/td&gt;
 &lt;td&gt;If the print is only moderate and the stock derates toward below KRW 3.0T market cap or below 10x FY2026E sales&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Catalysts are the 2Q result, 2Q-end backlog, 2026 delivery update, controller mix, normalization of amended contracts and 3Q revenue acceleration.&lt;/p&gt;
&lt;p&gt;Falsifiers are revenue below KRW 65B, OP below KRW 10B, controller mix below 75%, gross margin below 50%, further delays in the KRW 50B or KRW 20.3B contracts, no 3Q backlog-conversion guide, or continued dependence on two customers near 80%.&lt;/p&gt;
&lt;h2 id="final-view"&gt;Final View
&lt;/h2&gt;&lt;p&gt;FADU probably beats consensus in 2Q26. My midpoint is KRW 70.5B revenue and KRW 11.8B operating profit. But this is a moderate beat, not a clean buy signal at any price.&lt;/p&gt;
&lt;p&gt;FADU remains a valid AI-storage bottleneck candidate. The stock needs stricter evidence now: 2Q revenue above KRW 70B, controller mix near 80%, and proof that delayed contracts convert into 2H revenue.&lt;/p&gt;
&lt;p&gt;Sources: &lt;a class="link" href="https://v.daum.net/v/20260703070207071" target="_blank" rel="noopener"
 &gt;Daum News&lt;/a&gt;, &lt;a class="link" href="https://www.thelec.kr/news/articleView.html?idxno=58121" target="_blank" rel="noopener"
 &gt;The Elec&lt;/a&gt;, &lt;a class="link" href="https://www.dailyinvest.kr/news/articleView.html?idxno=70694" target="_blank" rel="noopener"
 &gt;DailyInvest&lt;/a&gt;, &lt;a class="link" href="https://kind.krx.co.kr/common/disclsviewer.do?acptno=20260512000812&amp;amp;docno=&amp;amp;method=search&amp;amp;viewerhost=" target="_blank" rel="noopener"
 &gt;KIND filing&lt;/a&gt;, &lt;a class="link" href="https://www.thebell.co.kr/front/newsview.asp?key=202605130923581040107666" target="_blank" rel="noopener"
 &gt;The Bell&lt;/a&gt;, &lt;a class="link" href="https://www.datatooza.com/article/20260413101307147852ef3fffef_80" target="_blank" rel="noopener"
 &gt;Datatooza&lt;/a&gt;, local KR price, flow and consensus databases as of July 3, 2026.&lt;/p&gt;</description></item><item><title>H1 2026 Review: AI Infrastructure Bottlenecks Broadened, but the Market Did Not</title><link>https://koreainvestinsights.com/post/h1-2026-ai-infra-bottleneck-korea-narrow-market-postmortem-2026-06-30/</link><pubDate>Tue, 30 Jun 2026 22:35:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/h1-2026-ai-infra-bottleneck-korea-narrow-market-postmortem-2026-06-30/</guid><description>
 &lt;blockquote&gt;
 &lt;p&gt;Related context
This is a half-year follow-up to &lt;a class="link" href="https://koreainvestinsights.com/post/samsung-2q26-preview-micron-surprise-erased-core-op-hbm-2026-06-29/" &gt;Samsung 2Q26 preview&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/micron-fy3q26-ai-memory-sca-fcf-hold-2026-06-25/" &gt;Micron FY3Q26 earnings&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/nvidia-earnings-elasticity-hbm-cycle-samsung-hynix-2026-06-28/" &gt;the NVIDIA elasticity framework for Samsung and SK Hynix&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/kospi-benchmark-hard-to-beat-narrow-market-monte-carlo-2026-06-20/" &gt;how hard it was to beat the KOSPI benchmark&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/kospi-disparity60-overheat-framework-partial-trim-2026-06-21/" &gt;the KOSPI 60-day disparity framework&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/real-money-flow-framework-korea-institution-quality-2026-06-03/" &gt;the Real Money Flow framework&lt;/a&gt;, and &lt;a class="link" href="https://koreainvestinsights.com/post/ai-supercycle-midgame-rate-risk-yellow-not-red-2026-06-06/" &gt;the AI supercycle midgame note&lt;/a&gt;.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;p&gt;H1 2026 looked like a broad risk-on period from the index level. Under the surface, it was much narrower. Korea was a KOSPI megacap crowding market. The U.S. was an AI infrastructure market where leadership moved from megacap AI platforms into memory, storage, servers, semiconductor equipment and power. Neither market was a broad “everything went up” market.&lt;/p&gt;
&lt;p&gt;The headline Korean index was extraordinary. KOSPI rose from 4,214.17 on December 30, 2025 to 8,476.48 on June 30, 2026, up 101.1%. KOSDAQ fell from 925.47 to 916.18, down 1.0%. Across roughly 2,700 Korean stocks, 856 rose and 1,788 fell. The median stock return was -12.6%. The index doubled, but the average stock was in a weak market.&lt;/p&gt;
&lt;p&gt;The U.S. story was also more specific than the index suggests. The S&amp;amp;P 500 gained 9.2% and Nasdaq gained 11.1%, but the real story was AI infrastructure bottlenecks. SanDisk, Micron, Western Digital, Dell, Marvell, ARM, Applied Materials and Lam Research were much stronger than NVIDIA. The market was buying the places where the AI factory actually bottlenecks.&lt;/p&gt;
&lt;p&gt;The H2 strategy is not simply “find cheap stocks.” The better rule is to buy proven bottlenecks with real revenue and contracts on pullbacks. Option-like equipment, gaming and biotech event names should be sized smaller. Continuous-revenue bottlenecks in memory, eSSD, MLCC, substrates, power, shipbuilding, defense and large biotech platforms deserve more attention.&lt;/p&gt;
&lt;div class="thesis-callout"&gt;
 &lt;div class="thesis-callout__label"&gt;One-Line Conclusion&lt;/div&gt;
 &lt;div class="thesis-callout__body"&gt;
 H1 was risk-on, but it was not broad. Korea was dominated by Samsung Electronics and SK Hynix crowding, while the U.S. rewarded the physical bottlenecks beneath AI compute. H2 is about separating recurring-revenue bottlenecks from option-like stories and leveraged flow risk.
 &lt;/div&gt;
&lt;/div&gt;
&lt;h2 id="0-data-basis-and-reconciliation"&gt;0. Data Basis and Reconciliation
&lt;/h2&gt;&lt;p&gt;The data basis is:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Area&lt;/th&gt;
 &lt;th&gt;Basis&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Korea indices and stocks&lt;/td&gt;
 &lt;td&gt;June 30, 2026 close&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;U.S. indices and stocks&lt;/td&gt;
 &lt;td&gt;June 29, 2026 close&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Korea flows&lt;/td&gt;
 &lt;td&gt;KR local DB, Kiwoom amount basis&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;U.S. stock returns&lt;/td&gt;
 &lt;td&gt;US local DB and yfinance adjusted close&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Secondary inputs&lt;/td&gt;
 &lt;td&gt;Naver Finance, local screeners and session flow analysis&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Some figures differ slightly by dataset and reference date. For example, the final KOSPI return is +101.1% using the June 30 close, while some local basket snapshots show +96.7%, and the June 23 interim review showed +90.4%. U.S. AI stock returns also differ depending on adjusted-close treatment and vendor handling.&lt;/p&gt;
&lt;p&gt;This post reconciles those differences as follows.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Data type&lt;/th&gt;
 &lt;th&gt;How it is used&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;June 30 Korea close&lt;/td&gt;
 &lt;td&gt;Main H1 performance benchmark&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;June 29 U.S. close&lt;/td&gt;
 &lt;td&gt;Main U.S. H1 performance benchmark&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;June 23 interim snapshot&lt;/td&gt;
 &lt;td&gt;Supporting evidence for breadth and buyer structure&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Adjusted-close versus local DB differences&lt;/td&gt;
 &lt;td&gt;Used directionally when conclusions are unchanged&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Less-verified single figures&lt;/td&gt;
 &lt;td&gt;Treated as H2 watch items, not as core proof&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The point is not false precision. The conclusion is stable: indices were strong, breadth was weak, and AI infrastructure returns moved from GPUs into memory, storage, servers, equipment and power.&lt;/p&gt;
&lt;h2 id="1-index-performance-kospi-doubled-kosdaq-fell"&gt;1. Index Performance: KOSPI Doubled, KOSDAQ Fell
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Market&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Basis&lt;/th&gt;
 &lt;th style="text-align: right"&gt;H1 return&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;KOSPI&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4,214.17 on 2025-12-30 to 8,476.48 on 2026-06-30&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+101.1%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KOSDAQ&lt;/td&gt;
 &lt;td style="text-align: right"&gt;925.47 to 916.18&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.0%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;S&amp;amp;P 500&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2025-12-31 to 2026-06-29&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+9.2%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Nasdaq&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2025-12-31 to 2026-06-29&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+11.1%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Russell 2000&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2025-12-31 to 2026-06-29&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+21.3%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SMH semiconductor ETF&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2025-12-31 to 2026-06-29&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+75.5%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SOXX semiconductor ETF&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2025-12-31 to 2026-06-29&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+104.2%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;At the index level, Korea was spectacular. KOSPI doubled in six months. But KOSDAQ fell. Within the same Korean market, large-cap AI infrastructure and growth stocks lived in different worlds.&lt;/p&gt;
&lt;p&gt;The U.S. index story also understates the internal rotation. The S&amp;amp;P 500 and Nasdaq were up around low double digits, while semiconductor ETFs were up far more. SOXX gained 104.2%, close to the KOSPI’s gain, showing how concentrated the global AI hardware trade became.&lt;/p&gt;
&lt;h2 id="2-korea-the-index-doubled-the-median-stock-fell"&gt;2. Korea: The Index Doubled, the Median Stock Fell
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Value&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Observed Korean stocks&lt;/td&gt;
 &lt;td style="text-align: right"&gt;roughly 2,700&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;H1 advancers&lt;/td&gt;
 &lt;td style="text-align: right"&gt;856&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;H1 decliners&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1,788&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Median stock return&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-12.6%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KOSPI H1 return&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+101.1%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KOSDAQ H1 return&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-1.0%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This is an unusual combination. The index was in a historic bull market, while the median stock was weak. That is why the lived investor experience was so different from the index headline.&lt;/p&gt;
&lt;p&gt;The main H1 leaders were:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Stock&lt;/th&gt;
 &lt;th style="text-align: right"&gt;H1 return&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electro-Mechanics&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+756%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Jusung Engineering&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+626%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Square&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+361%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+307%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Jeju Semiconductor&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+294%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Daeduck Electronics&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+226%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Simmtech&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+201%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+179%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The Korean market’s structure was:&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"&gt;&lt;code class="language-text" data-lang="text"&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;Samsung Electronics and SK Hynix
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Samsung Electro-Mechanics, MLCC and FC-BGA
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ substrates, PCB and semi equipment
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ power equipment, industrials and some financials
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;KOSDAQ, many biotech names, platforms and parts of secondary batteries lagged. Korea was strong, but most Korean stocks were not.&lt;/p&gt;
&lt;h2 id="3-korea-was-not-a-foreign-buying-market"&gt;3. Korea Was Not a Foreign-Buying Market
&lt;/h2&gt;&lt;p&gt;YTD flow data on the local Kiwoom amount basis:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Investor group&lt;/th&gt;
 &lt;th style="text-align: right"&gt;YTD cumulative&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Foreigners&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 142T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Institutions&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 45T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Retail&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 77.7T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Financial investment&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 75.4T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Investment trusts&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 1.7T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Private funds&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 3.8T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Pension funds and others&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 7.0T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The key point is that Korea was not a simple foreign-buying market. Foreigners sold Samsung Electronics and SK Hynix heavily. Domestic retail, financial investment accounts and institutions absorbed the flow.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Stock&lt;/th&gt;
 &lt;th style="text-align: right"&gt;YTD return&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Foreign cumulative flow&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td style="text-align: right"&gt;around +160%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 72.6T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td style="text-align: right"&gt;around +291%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 57.1T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Square&lt;/td&gt;
 &lt;td style="text-align: right"&gt;around +333%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 7.6T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics preferred&lt;/td&gt;
 &lt;td style="text-align: right"&gt;around +125%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 3.2T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Foreigners sold the core megacaps while those stocks rallied. That is not a normal foreign-led rally. It was a domestic absorption market.&lt;/p&gt;
&lt;p&gt;Foreigners did not sell everything. They rotated into select bottlenecks and global-fund-friendly assets such as Samsung Electro-Mechanics, Doosan Enerbility, Samsung SDI, FADU, Hanwha Ocean and Sanil Electric. The right description is not “foreigners bought Korea.” It is “foreigners sold the largest memory megacaps, domestic liquidity absorbed them, and foreigners reallocated into select second-layer bottlenecks.”&lt;/p&gt;
&lt;h2 id="4-the-hidden-buyer-was-financial-investment-not-long-only-real-money"&gt;4. The Hidden Buyer Was Financial Investment, Not Long-Only Real Money
&lt;/h2&gt;&lt;p&gt;Institutional buying was not the same as long-only accumulation. The strongest institutional subcategory was financial investment, not investment trusts, private funds or pension funds.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Flow category&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Financial investment&lt;/td&gt;
 &lt;td&gt;ETF, derivatives, program and rebalance-linked flow&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Investment trusts&lt;/td&gt;
 &lt;td&gt;More likely to represent long-only fund demand, but weak in H1&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Private funds&lt;/td&gt;
 &lt;td&gt;Mixed thematic, long-short and event-driven flow, but weak cumulatively&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Pension funds and others&lt;/td&gt;
 &lt;td&gt;Long-term real money, but net selling in H1&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Retail&lt;/td&gt;
 &lt;td&gt;Key absorber of megacap AI infrastructure flow and leverage&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This matters. Real money can sustain a trend. Financial-investment and program flow can be powerful, but it is more vulnerable to futures expiry, ETF caps, derivative positioning and leverage mechanics.&lt;/p&gt;
&lt;p&gt;H1 Korea was both a fundamental rerating market and a derivatives/ETF market. Confusing the two leads to repeated mistakes.&lt;/p&gt;
&lt;h2 id="5-kosdaq-lacked-buyers-not-stories"&gt;5. KOSDAQ Lacked Buyers, Not Stories
&lt;/h2&gt;&lt;p&gt;KOSDAQ had plenty of stories: biotech, robotics, CPO, SOCAMM, semiconductor equipment, gaming and AI software. The index still lagged because the buyer base was weak.&lt;/p&gt;
&lt;p&gt;The June 23 interim snapshot was already clear:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Index&lt;/th&gt;
 &lt;th style="text-align: right"&gt;YTD&lt;/th&gt;
 &lt;th style="text-align: right"&gt;From high&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;KOSPI&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+90.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-10.0%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KOSDAQ&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-5.7%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-27.3%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The final June 30 structure was the same: KOSPI +101.1%, KOSDAQ -1.0%.&lt;/p&gt;
&lt;p&gt;KOSDAQ struggled because:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Retail capital was tied up absorbing Samsung Electronics, SK Hynix and large AI infrastructure names.&lt;/li&gt;
&lt;li&gt;Institutional real money did not buy KOSDAQ broadly.&lt;/li&gt;
&lt;li&gt;Foreign buying was narrow, focused on names such as FADU, Jeju Semiconductor and Simmtech.&lt;/li&gt;
&lt;li&gt;Rates and FX pressure weighed on biotech and unprofitable growth multiples.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;The condition for KOSDAQ recovery is:&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"&gt;&lt;code class="language-text" data-lang="text"&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;KOSDAQ recovery =
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;retail capital returning
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;+ institutional real-money buying
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;+ Samsung Electronics and SK Hynix moving sideways rather than falling
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;+ KOSDAQ earnings revisions stabilizing
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;If Samsung and SK Hynix crash, money does not automatically rotate into KOSDAQ. Risk-off comes first. The best environment for KOSDAQ is not megacap collapse. It is megacap consolidation.&lt;/p&gt;
&lt;h2 id="6-the-us-market-was-not-just-nvidia"&gt;6. The U.S. Market Was Not Just NVIDIA
&lt;/h2&gt;&lt;p&gt;The real U.S. signal was not “AI went up.” It was:&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;The physical bottlenecks after the GPU outperformed the GPU leader.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;p&gt;Representative H1 return ranges:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Stock&lt;/th&gt;
 &lt;th style="text-align: right"&gt;H1 return range&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;SanDisk&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+645-764%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+263-301%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Western Digital&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+247-279%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Intel&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+235-257%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Dell&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+224-232%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Marvell&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+211-227%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;ARM&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+200-214%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Applied Materials&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+171%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AMD&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+152%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Lam Research&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+141%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Big Tech was mixed:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Stock&lt;/th&gt;
 &lt;th style="text-align: right"&gt;H1 return&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Google&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+13%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Amazon&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+4%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Apple&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+4%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NVIDIA&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+3-5%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Broadcom&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+7%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Meta&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-15%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Microsoft&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-22% to -23%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Oracle&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-25%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Palantir&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-31%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The market bought the points where the AI factory bottlenecks:&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"&gt;&lt;code class="language-text" data-lang="text"&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;GPU
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ HBM and DRAM
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ eSSD and storage
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ servers and racks
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ optical and interconnect
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ semiconductor equipment
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ power and cooling
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;This is directly relevant to Korea. HBM maps to Samsung and SK Hynix. eSSD and NAND map to Samsung, FADU and Jeju Semiconductor. MLCC and FC-BGA map to Samsung Electro-Mechanics and substrate names. Power maps to Doosan Enerbility and electrical equipment. Equipment maps to HPSP, TES, VM and Jusung.&lt;/p&gt;
&lt;h2 id="7-macro-was-not-a-pure-tailwind"&gt;7. Macro Was Not a Pure Tailwind
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Asset or indicator&lt;/th&gt;
 &lt;th style="text-align: right"&gt;H1 move&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Dollar index&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+3.1%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;USD/KRW&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+7.9%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;U.S. 10-year yield&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4.16% to 4.39%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;WTI oil&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+24.1%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Gold&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-6.8%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Bitcoin&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-33.2%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This was not a clean multiple-expansion macro environment. The dollar rose, the won weakened, U.S. yields rose and oil rose. Gold and Bitcoin fell.&lt;/p&gt;
&lt;p&gt;Stocks rose anyway because the AI infrastructure earnings cycle overwhelmed macro pressure. That is important for H2. If rates, the dollar and oil all rise together again, the most crowded bottleneck stocks can suffer multiple compression even if demand remains strong.&lt;/p&gt;
&lt;p&gt;The realistic risk is:&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"&gt;&lt;code class="language-text" data-lang="text"&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;AI demand stays strong
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;+ earnings remain good
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;+ rates, dollar and oil remain firm
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;+ valuation and leveraged flows crack first
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;That is the same stock-elasticity problem discussed in the NVIDIA/Samsung/SK Hynix framework.&lt;/p&gt;
&lt;h2 id="8-non-consensus-truth-1-leadership-moved-from-compute-to-memory"&gt;8. Non-Consensus Truth 1: Leadership Moved From Compute to Memory
&lt;/h2&gt;&lt;p&gt;The first non-consensus point is U.S. AI leadership. NVIDIA is still the center of AI. But incremental stock performance moved into Micron, SanDisk, Western Digital, Dell, Marvell and ARM.&lt;/p&gt;
&lt;p&gt;That does not mean compute no longer matters. It means the marginal return moved down the stack.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Question&lt;/th&gt;
 &lt;th&gt;Meaning&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Is this company a cost item in AI capex?&lt;/td&gt;
 &lt;td&gt;It can be pressured when customers optimize costs.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Is this company a bottleneck item in AI capex?&lt;/td&gt;
 &lt;td&gt;It can reprice if customers need supply.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Is this company only a narrative?&lt;/td&gt;
 &lt;td&gt;It is vulnerable during catalyst gaps.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Cost items get pressured. Bottlenecks rerate. Narratives require timing discipline.&lt;/p&gt;
&lt;h2 id="9-non-consensus-truth-2-the-real-top-came-before-the-index-high"&gt;9. Non-Consensus Truth 2: The Real Top Came Before the Index High
&lt;/h2&gt;&lt;p&gt;Breadth weakened before the index did. The June index high was driven by Samsung Electronics, SK Hynix and a small set of AI infrastructure names. KOSDAQ and the median stock had already weakened earlier.&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"&gt;&lt;code class="language-text" data-lang="text"&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;Weak median stock
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;+ weak KOSDAQ
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;+ strong KOSPI memory megacaps
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;= index high and weak lived market at the same time
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;For H2, the key indicators are breadth, ADR, KOSDAQ/KOSPI relative strength, megacap memory trading-value share, and KOSDAQ retail/institutional joint buying.&lt;/p&gt;
&lt;h2 id="10-non-consensus-truth-3-multiples-moved-before-fundamentals-fully-arrived"&gt;10. Non-Consensus Truth 3: Multiples Moved Before Fundamentals Fully Arrived
&lt;/h2&gt;&lt;p&gt;Samsung, SK Hynix, Micron and Samsung Electro-Mechanics all had strong fundamentals. But stock returns moved faster than near-term earnings. The market changed the multiple.&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"&gt;&lt;code class="language-text" data-lang="text"&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;Old frame:
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;memory = commodity cycle = peak-earnings discount
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;H1 frame:
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;memory = AI infrastructure bottleneck = supply shortage plus contract premium
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;The H2 risk is not just bad earnings. It is good earnings with lower multiples if the market starts treating memory as commodity again. Investors should watch EPS revision speed, beat magnitude and FY2027 visibility, not just the next quarterly beat.&lt;/p&gt;
&lt;h2 id="11-non-consensus-truth-4-leverage-plumbing-amplified-tail-risk"&gt;11. Non-Consensus Truth 4: Leverage Plumbing Amplified Tail Risk
&lt;/h2&gt;&lt;p&gt;Korean memory megacaps were affected not only by fundamentals, but also by leverage plumbing: 2x and 3x Samsung/SK Hynix ETFs listed offshore, domestic leverage and inverse products, futures and program flow.&lt;/p&gt;
&lt;p&gt;Simplified upside loop:&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"&gt;&lt;code class="language-text" data-lang="text"&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;stock up
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ leveraged ETF rebalancing buy
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ futures and program support
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ more megacap upside
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ more momentum capital
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;Downside loop:&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"&gt;&lt;code class="language-text" data-lang="text"&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;stock down
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ leveraged ETF rebalancing sell
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ futures and program pressure
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ retail margin and forced-selling risk
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ larger drawdown
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;Korea can therefore move more violently than Micron even when the fundamental direction is similar.&lt;/p&gt;
&lt;h2 id="12-non-consensus-truth-5-supply-moved-from-strong-hands-to-weaker-hands"&gt;12. Non-Consensus Truth 5: Supply Moved From Strong Hands to Weaker Hands
&lt;/h2&gt;&lt;p&gt;The important question is not only who sold. It is who bought.&lt;/p&gt;
&lt;p&gt;H1 saw foreigners and some institutional accounts reduce exposure in memory megacaps while retail and financial-investment accounts absorbed the flow.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Positive read&lt;/th&gt;
 &lt;th&gt;Negative read&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Domestic liquidity is strong&lt;/td&gt;
 &lt;td&gt;Shares may have moved into weaker leveraged hands&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KOSPI held despite foreign selling&lt;/td&gt;
 &lt;td&gt;Forced-selling risk rises if margin unwinds&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Large-cap buyer base widened&lt;/td&gt;
 &lt;td&gt;It may also represent distribution from better-informed accounts&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;H2 monitoring should combine foreign cash flow, retail margin, deposits, ETF balances, and program flow.&lt;/p&gt;
&lt;h2 id="13-non-consensus-truth-6-foreigners-bought-globally-explainable-korean-assets"&gt;13. Non-Consensus Truth 6: Foreigners Bought Globally Explainable Korean Assets
&lt;/h2&gt;&lt;p&gt;Foreigners did not buy every Korean theme. The stocks they bought were globally explainable.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Stock&lt;/th&gt;
 &lt;th style="text-align: right"&gt;YTD foreign net buying&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electro-Mechanics&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 2.28T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Doosan Enerbility&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 1.71T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Celltrion&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 1.52T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung SDI&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 1.33T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Doosan&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 1.11T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;APR&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 1.08T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Hyundai Rotem&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 0.85T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;LG Energy Solution&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 0.85T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FADU&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 0.80T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Hanwha Ocean&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 0.76T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Common features:&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"&gt;&lt;code class="language-text" data-lang="text"&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;power, nuclear, shipbuilding and defense
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;global quality industrials
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;large battery names
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;large biotech platforms
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;direct AI infrastructure bottlenecks
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;For H2, foreign long-only capital is more likely to buy assets it can explain in an investment committee than obscure small-cap stories.&lt;/p&gt;
&lt;h2 id="14-non-consensus-truth-7-revenue-pattern-beat-company-quality"&gt;14. Non-Consensus Truth 7: Revenue Pattern Beat Company Quality
&lt;/h2&gt;&lt;p&gt;In H1, the quality of the revenue pattern mattered more than abstract company quality.&lt;/p&gt;
&lt;p&gt;Weak revenue-pattern names:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Type&lt;/th&gt;
 &lt;th&gt;Weakness&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Equipment&lt;/td&gt;
 &lt;td&gt;Order gaps can create earnings gaps.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Gaming&lt;/td&gt;
 &lt;td&gt;Revenue visibility breaks around launch cycles.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Biotech event stocks&lt;/td&gt;
 &lt;td&gt;Trial and licensing gaps create narrative gaps.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Some robotics and AI software&lt;/td&gt;
 &lt;td&gt;Expectations move before revenue.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Strong revenue-pattern names:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Type&lt;/th&gt;
 &lt;th&gt;Strength&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Memory&lt;/td&gt;
 &lt;td&gt;Pricing and volume improved together.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;eSSD and storage&lt;/td&gt;
 &lt;td&gt;AI server storage bottleneck rerating.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;MLCC and substrates&lt;/td&gt;
 &lt;td&gt;Server and package bottlenecks.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Power equipment&lt;/td&gt;
 &lt;td&gt;Data-center power demand plus policy capital.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Shipbuilding&lt;/td&gt;
 &lt;td&gt;Long backlog visibility.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Defense&lt;/td&gt;
 &lt;td&gt;Export backlog and government budgets.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Large biotech platforms&lt;/td&gt;
 &lt;td&gt;Contracts, production and platform value are explainable.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The H2 rule is simple:&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Size option-like names smaller. Size recurring-revenue bottlenecks larger.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="15-bonus-hypothesis-korea-became-a-leveraged-us-ai-proxy"&gt;15. Bonus Hypothesis: Korea Became a Leveraged U.S. AI Proxy
&lt;/h2&gt;&lt;p&gt;Korea’s market identity changed in H1. Historically, Samsung and SK Hynix were proxies for EM, China and global IT inventory cycles. In H1 2026, the linkage to U.S. AI capex became much stronger.&lt;/p&gt;
&lt;p&gt;Session analysis suggested a stronger relationship between U.S. prior-day AI/semi moves and Korea next-day memory megacaps, including an interim correlation estimate of 0.42. That figure needs formal revalidation, but the price action is consistent with the direction.&lt;/p&gt;
&lt;p&gt;The issue is diversification. If Korea behaves like a leveraged U.S. AI capex proxy, domestic buffers weaken when the global AI factor cools.&lt;/p&gt;
&lt;p&gt;Another watch item is capex. Session analysis classified large SK Hynix capex commitment as a peak-confidence checkpoint. The exact amount and scope should be rechecked against filings before it becomes a core argument. But historically, peak capex announcements often arrive when demand looks strongest.&lt;/p&gt;
&lt;p&gt;Key H2 capex questions:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Question&lt;/th&gt;
 &lt;th&gt;Meaning&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Are customer contracts supporting capex?&lt;/td&gt;
 &lt;td&gt;Lowers supply-glut risk.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Does capex impair FCF?&lt;/td&gt;
 &lt;td&gt;Earnings can look good while cash flow weakens.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Does supply arrive near demand peak?&lt;/td&gt;
 &lt;td&gt;Raises 2027 cycle risk.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Is capex HBM and AI-memory focused?&lt;/td&gt;
 &lt;td&gt;Must be separated from commodity DRAM oversupply risk.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="16-h2-checklist"&gt;16. H2 Checklist
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Checkpoint&lt;/th&gt;
 &lt;th&gt;Why it matters&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung 2Q26 Core OP&lt;/td&gt;
 &lt;td&gt;Core profit power matters more than reported headline OP.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix 2Q and 3Q HBM allocation&lt;/td&gt;
 &lt;td&gt;Determines whether the crowded winner can reaccelerate.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron FY4Q26 guide delivery&lt;/td&gt;
 &lt;td&gt;Tests whether U.S. AI memory premium can persist.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM and NAND pricing&lt;/td&gt;
 &lt;td&gt;Confirms margin defense beyond HBM.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KOSDAQ retail and institutional buying&lt;/td&gt;
 &lt;td&gt;Needed for KOSDAQ recovery.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Financial-investment and program flow&lt;/td&gt;
 &lt;td&gt;Real plumbing behind megacap volatility.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;USD/KRW, U.S. 10-year yields and oil&lt;/td&gt;
 &lt;td&gt;Put a ceiling on AI bottleneck multiples.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AI capex contracts and FCF&lt;/td&gt;
 &lt;td&gt;Shows whether demand is paid for by cash flow, not just narrative.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="17-h2-strategy"&gt;17. H2 Strategy
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Area&lt;/th&gt;
 &lt;th&gt;Strategy&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;AI infrastructure core&lt;/td&gt;
 &lt;td&gt;Keep exposure, but buy pullbacks rather than chasing.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Memory megacaps&lt;/td&gt;
 &lt;td&gt;Samsung is the unreflected-alpha candidate. SK Hynix is the reacceleration candidate after pullbacks.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM lower-stack bottlenecks&lt;/td&gt;
 &lt;td&gt;Buy only where orders, margins and customer qualification are visible.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KOSDAQ&lt;/td&gt;
 &lt;td&gt;Do not buy the whole index first. Start with semiconductor equipment, test and profitable growth names.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Option-like equipment, gaming and biotech&lt;/td&gt;
 &lt;td&gt;Smaller sizing due to catalyst-gap risk.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Global quality cyclicals&lt;/td&gt;
 &lt;td&gt;Continue to monitor shipbuilding, defense, power, nuclear and large biotech platforms.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The H2 strategy in one sentence:&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Keep AI infrastructure core exposure, reduce option-like positions, and shift toward recurring-revenue bottlenecks and Korean quality cyclicals that global capital can explain.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="18-final-view"&gt;18. Final View
&lt;/h2&gt;&lt;p&gt;The crowd may remember H1 2026 as the best Korea/semiconductor supercycle year. That is not wrong. KOSPI doubled and memory, substrates, equipment and power leaders exploded.&lt;/p&gt;
&lt;p&gt;But the more important story is:&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"&gt;&lt;code class="language-text" data-lang="text"&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;leadership became extremely narrow
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;multiples moved before fundamentals fully arrived
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;foreigners sold the core megacaps while domestic liquidity absorbed them
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;financial-investment and ETF/program flows amplified the market
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;KOSDAQ lacked buyers, not stories
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;U.S. AI leadership moved from GPU into memory, storage, servers and power
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;The H2 question is not only “how much more upside is left?” The better questions are:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Does stock elasticity remain positive even after strong earnings?&lt;/li&gt;
&lt;li&gt;Can domestic liquidity keep absorbing foreign selling?&lt;/li&gt;
&lt;li&gt;Do KOSDAQ buyers return?&lt;/li&gt;
&lt;li&gt;Do AI bottlenecks convert from narrative into revenue and contracts?&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;H1 was risk-on, but it was not broad. It was a compressed leadership market. H2 requires more precision: real bottlenecks over cheap stocks, recurring revenue over one-off catalysts, and Core OP plus FCF over headline earnings.&lt;/p&gt;
&lt;h2 id="data-note"&gt;Data Note
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;Korea data: KR local DB, Kiwoom investor flow data and Naver Finance support as of the June 30, 2026 close.&lt;/li&gt;
&lt;li&gt;U.S. data: US local DB and yfinance adjusted close as of the June 29, 2026 close.&lt;/li&gt;
&lt;li&gt;Interim snapshots: June 23 KOSPI/KOSDAQ relative performance and session flow analysis.&lt;/li&gt;
&lt;li&gt;U.S. June 30 regular-session results are not included.&lt;/li&gt;
&lt;li&gt;Less-verified capex single figures are treated as H2 monitoring items rather than core proof.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;em&gt;Disclaimer: For research and information purposes only. Not investment advice. Names cited are for analytical illustration; readers should perform their own due diligence and consult licensed advisors before any investment decision.&lt;/em&gt;&lt;/p&gt;</description></item><item><title>Samsung 2Q26 Preview: Micron’s Surprise Has Been Erased, Core OP Is the Real Test</title><link>https://koreainvestinsights.com/post/samsung-2q26-preview-micron-surprise-erased-core-op-hbm-2026-06-29/</link><pubDate>Mon, 29 Jun 2026 14:35:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/samsung-2q26-preview-micron-surprise-erased-core-op-hbm-2026-06-29/</guid><description>
 &lt;blockquote&gt;
 &lt;p&gt;Related context
This is a follow-up to &lt;a class="link" href="https://koreainvestinsights.com/post/micron-fy3q26-ai-memory-sca-fcf-hold-2026-06-25/" &gt;Micron FY3Q26 earnings&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/nvidia-earnings-elasticity-hbm-cycle-samsung-hynix-2026-06-28/" &gt;the NVIDIA elasticity framework for Samsung and SK Hynix&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/samsung-hynix-micron-forward-per-parity-memory-catch-up-2026-06-03/" &gt;Samsung-Hynix-Micron parity&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/samsung-electronics-shareholder-return-ds-bonus-buyback-flow-2026-06-22/" &gt;Samsung shareholder return and DS bonus buyback flow&lt;/a&gt;, and &lt;a class="link" href="https://koreainvestinsights.com/post/samsung-sk-hynix-hbm4e-12h-sample-allocation-battle-2026-06-18/" &gt;Samsung versus SK Hynix in HBM4E 12-high&lt;/a&gt;.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;p&gt;The number that matters in Samsung Electronics’ 2Q26 preview is not simply reported operating profit. The market needs to separate reported OP from &lt;strong&gt;Core OP, defined as reported OP plus the catch-up accrual for performance-based compensation&lt;/strong&gt;.&lt;/p&gt;
&lt;p&gt;At current prices, Micron’s FY3Q26 surprise has almost disappeared from Korean memory equities. Micron retained an 8.0% gain versus its pre-earnings price as of June 26. Samsung Electronics is down 5.9% versus the June 24 close based on June 29 intraday data, while SK Hynix is down 0.5%. The June 25 reaction was real, but most of it was erased by June 26 and June 29.&lt;/p&gt;
&lt;p&gt;Samsung’s 2Q26 DRAM pricing is not the downgrade factor. Conventional DRAM contract prices are estimated to be up 58-63% QoQ, while mobile LPDDR is up 70-83% QoQ. A blended Samsung DRAM ASP assumption in the high-50% QoQ range is therefore not aggressive.&lt;/p&gt;
&lt;p&gt;The reasonable range for 2Q26 reported OP is KRW 87-92 trillion, with a midpoint of KRW 89.5 trillion. Adding back KRW 15-17 trillion of catch-up bonus accrual implies Core OP of roughly KRW 104-106 trillion. In other words, the correct reading is not “earnings fell from the KRW 100 trillion expectation to KRW 89 trillion.” It is closer to “the memory core business reached more than KRW 100 trillion of profit power, but Samsung is recognizing excess-profit sharing costs in 2Q.”&lt;/p&gt;
&lt;p&gt;The rating is &lt;strong&gt;Watchlist / Conditional Buy&lt;/strong&gt;. Conditions are clear: reported OP should be at least KRW 87 trillion, Core OP should be at least KRW 102 trillion, the downgrade should be explained by accruals rather than DRAM/NAND price or shipment weakness, and 3Q26 OP durability around KRW 100 trillion should remain credible.&lt;/p&gt;
&lt;div class="thesis-callout"&gt;
 &lt;div class="thesis-callout__label"&gt;Main Point&lt;/div&gt;
 &lt;div class="thesis-callout__body"&gt;
 Micron’s surprise was reflected in Samsung and SK Hynix for one day on June 25, but it has mostly been erased. Samsung is back to a price that barely embeds the Micron read-through. The real test is not headline 2Q OP, but Core OP above KRW 100 trillion and the durability of 3Q memory earnings.
 &lt;/div&gt;
&lt;/div&gt;
&lt;h2 id="1-price-reaction-is-microns-surprise-still-in-korea-memory"&gt;1. Price reaction: is Micron’s surprise still in Korea memory?
&lt;/h2&gt;&lt;p&gt;Micron reported FY3Q26 results after the U.S. close on June 24, 2026. For Korea, the clean baseline is the June 24 close.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Stock&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Jun 24 close&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Jun 25 close&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Jun 26 close&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Jun 29 intraday&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Current vs Jun 24&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 340,500&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 358,500&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 339,500&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 320,500&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-5.9%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 2,580,000&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 2,917,000&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 2,673,000&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 2,568,000&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-0.5%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron&lt;/td&gt;
 &lt;td style="text-align: right"&gt;USD 1,048.51&lt;/td&gt;
 &lt;td style="text-align: right"&gt;USD 1,213.56&lt;/td&gt;
 &lt;td style="text-align: right"&gt;USD 1,132.33&lt;/td&gt;
 &lt;td style="text-align: right"&gt;latest DB value Jun 26&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+8.0%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Source: KII local daily price database for Korea and U.S. stocks; June 29 intraday Korea prices from Naver Finance API at 13:19 KST.&lt;/p&gt;
&lt;p&gt;SK Hynix reacted most directly on June 25, rising 13.1%. But it has since nearly returned to its pre-event level. Samsung reacted 5.3% on June 25, then fell below the pre-event price. Micron still held an 8.0% gain as of June 26.&lt;/p&gt;
&lt;p&gt;The conclusion is simple: at current prices, Korean memory has not retained Micron’s earnings read-through. It was reflected for one day, then erased by macro pressure, KOSPI overheat adjustment and company-specific uncertainty. The unreflected alpha is more visible in Samsung. SK Hynix is more of a pullback-and-reacceleration candidate.&lt;/p&gt;
&lt;h2 id="2-what-micron-confirmed"&gt;2. What Micron confirmed
&lt;/h2&gt;&lt;p&gt;Micron FY3Q26 was a strong benchmark for the memory cycle. Revenue was USD 41.456 billion, non-GAAP EPS was USD 25.11, non-GAAP gross margin was 84.9%, and FY4Q26 guidance called for USD 50.0 billion revenue and USD 31.00 non-GAAP EPS. These are official company numbers. &lt;a class="link" href="https://www.globenewswire.com/news-release/2026/06/24/3317151/14450/en/micron-technology-inc-reports-record-results-for-the-third-quarter-of-fiscal-2026.html" target="_blank" rel="noopener"
 &gt;Micron FY3Q26 earnings release&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;The key read-throughs are:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Micron signal&lt;/th&gt;
 &lt;th&gt;Meaning for Samsung and SK Hynix&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;USD 41.456B revenue and 84.9% non-GAAP GM&lt;/td&gt;
 &lt;td&gt;Memory pricing, mix and shortage-driven operating leverage are very strong.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY4Q guidance of USD 50B revenue and USD 31 EPS&lt;/td&gt;
 &lt;td&gt;Pricing and shipments are unlikely to collapse immediately after 2Q.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Data center revenue above USD 25B&lt;/td&gt;
 &lt;td&gt;AI memory demand is broader than HBM alone.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM4 and HBM4E roadmap&lt;/td&gt;
 &lt;td&gt;2027 allocation and customer qualification remain the next battleground.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Strategic Customer Agreements&lt;/td&gt;
 &lt;td&gt;Long-term contracts may reduce the old memory cyclicality discount.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This matters because Samsung now has to prove the same point with its own numbers: whether core memory profitability is already in the KRW 100 trillion range.&lt;/p&gt;
&lt;h2 id="3-reported-op-versus-core-op"&gt;3. Reported OP versus Core OP
&lt;/h2&gt;&lt;p&gt;The working formula is:&lt;/p&gt;
&lt;div class="highlight"&gt;&lt;pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"&gt;&lt;code class="language-text" data-lang="text"&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;Reported OP = accounting operating profit
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;Core OP = Reported OP + catch-up accrual for performance-based compensation
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;Base case:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Base case&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Reported OP&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 89.5T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Catch-up bonus accrual&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 15-17T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Core OP&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 104-106T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The important point is that the expected OP cut is not driven by weaker DRAM or NAND assumptions. In the research input, Kiwoom’s 2Q26 OP estimate fell from KRW 100T in May to KRW 89T on June 29. But over the same period, DRAM ASP assumptions moved from +55% QoQ to +58% QoQ, and NAND from +72% QoQ to +75% QoQ. The downgrade is therefore better explained by bonus accruals, non-memory weakness, DX cost pressure, and continued Foundry/S.LSI losses.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;May estimate&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Late-June estimate&lt;/th&gt;
 &lt;th&gt;Meaning&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;2Q26 OP&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 100T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 89T&lt;/td&gt;
 &lt;td&gt;Headline lower&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM ASP&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+55% QoQ&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+58% QoQ&lt;/td&gt;
 &lt;td&gt;Price assumption stronger&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NAND ASP&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+72% QoQ&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+75% QoQ&lt;/td&gt;
 &lt;td&gt;Price assumption stronger&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Downgrade driver&lt;/td&gt;
 &lt;td style="text-align: right"&gt;not applicable&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Bonus, non-memory, DX, Foundry/S.LSI&lt;/td&gt;
 &lt;td&gt;Not a memory pricing problem&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="4-official-1q26-base"&gt;4. Official 1Q26 base
&lt;/h2&gt;&lt;p&gt;Samsung’s 1Q26 official results provide the base:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;1Q26&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Company revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 133.9T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Company operating profit&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 57.2T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DS revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 81.7T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DS operating profit&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 53.7T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Source: &lt;a class="link" href="https://news.samsungsemiconductor.com/global/samsung-electronics-announces-first-quarter-2026-results/" target="_blank" rel="noopener"
 &gt;Samsung Electronics 1Q26 results&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;DS operating profit of KRW 53.7T is the key base for the 2Q bonus accrual discussion. If 1Q did not fully accrue the new profit-sharing structure, 2Q can carry a catch-up cost.&lt;/p&gt;
&lt;h2 id="5-dram-pricing-is-supportive"&gt;5. DRAM pricing is supportive
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Product group&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Estimated 2Q26 price trend&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Conventional DRAM&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+58-63% QoQ&lt;/td&gt;
 &lt;td&gt;Kiwoom’s +58% assumption is near the lower end of the market range.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;PC DDR5&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+43-48% QoQ&lt;/td&gt;
 &lt;td&gt;PC demand is not strong, but pricing still rose.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;PC DDR4&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+35-40% QoQ&lt;/td&gt;
 &lt;td&gt;Lower than DDR5, but still strong.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Mobile LPDDR4X&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+70-75% QoQ&lt;/td&gt;
 &lt;td&gt;Mobile DRAM helps pull blended ASP higher.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Mobile LPDDR5X&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+78-83% QoQ&lt;/td&gt;
 &lt;td&gt;Positive for Samsung’s blended DRAM ASP.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Consumer DRAM&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+45-50% QoQ&lt;/td&gt;
 &lt;td&gt;Customer price resistance exists, but pricing is still up.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The core conclusion is that a high-50% QoQ blended DRAM ASP assumption is reasonable. If reported OP comes in near KRW 89T, it should not be interpreted as DRAM price weakness unless Samsung’s own disclosures say otherwise.&lt;/p&gt;
&lt;h2 id="6-bonus-accrual-treatment"&gt;6. Bonus accrual treatment
&lt;/h2&gt;&lt;p&gt;The existence of DS performance-based compensation is not new. The unsettled question is scale, timing and disclosure quality.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Analytical purpose&lt;/th&gt;
 &lt;th&gt;How to treat bonus accrual&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;2Q earnings quality&lt;/td&gt;
 &lt;td&gt;Add it back to estimate Core OP&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;3Q run-rate&lt;/td&gt;
 &lt;td&gt;Exclude 1Q catch-up, include normal ongoing bonus expense&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY26 EPS&lt;/td&gt;
 &lt;td&gt;Include it&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY27 valuation&lt;/td&gt;
 &lt;td&gt;Include it if recurring&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Shareholder return and FCF&lt;/td&gt;
 &lt;td&gt;Include it&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Short-term event trade&lt;/td&gt;
 &lt;td&gt;Watch both reported OP and Core OP&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The two mistakes to avoid are: treating all bonus expense as a one-off, or treating all of it as structural deterioration. The catch-up portion matters for 2Q interpretation. The recurring profit-sharing structure matters for FY26/FY27 EPS and valuation.&lt;/p&gt;
&lt;h2 id="7-sell-side-op-distribution"&gt;7. Sell-side OP distribution
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Source&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2Q26 OP estimate&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;BNK&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 77.7T&lt;/td&gt;
 &lt;td&gt;Most conservative bonus retroactive accrual assumption&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Securities&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 81.3T&lt;/td&gt;
 &lt;td&gt;DRAM ASP +46%, NAND ASP +65%, DS bonus reflected&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Mirae Asset&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 84.1T&lt;/td&gt;
 &lt;td&gt;Strong pre-accrual cost scenario near KRW 35T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FnGuide consensus&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 86.2T&lt;/td&gt;
 &lt;td&gt;Early-June average, may not fully reflect latest prices and accruals&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Kiwoom Jun 29&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 89.0T&lt;/td&gt;
 &lt;td&gt;Higher DRAM/NAND price assumptions, but bonus/non-memory costs reflected&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KB&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 90.0-90.2T&lt;/td&gt;
 &lt;td&gt;Memory price up more than 50%, DS OP in the high-KRW 80T range&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Daishin&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 91.0T&lt;/td&gt;
 &lt;td&gt;Memory price strength with partial cost reflection&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Hana&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 92.0T&lt;/td&gt;
 &lt;td&gt;Stronger pricing and shipment assumptions&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Investment framing:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Reported OP range&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Below KRW 84T&lt;/td&gt;
 &lt;td&gt;Bear case. Costs and non-memory drag are very large.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KRW 87-92T&lt;/td&gt;
 &lt;td&gt;Base cluster. This is the realistic range.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Above KRW 94T&lt;/td&gt;
 &lt;td&gt;Bull case. Costs are limited or price/shipments are even stronger.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="8-scenario-model"&gt;8. Scenario model
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Bear&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Base&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Bull&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM blended ASP&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+50%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+56-58%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+62-63%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM bit growth&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+6%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FX effect&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+1.0%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+1.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+2.0%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 86-88T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 91-93T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 95-97T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM OP margin&lt;/td&gt;
 &lt;td style="text-align: right"&gt;71-73%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;73-75%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;76-78%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM OP&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 62-64T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 67-70T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 72-75T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NAND OP&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 18-20T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 20-22T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 22-24T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Foundry/S.LSI OP&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 1.5T to -KRW 1.0T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 1.0T to 0&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0 to KRW 0.5T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Display/DX/Harman combined&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 0.5T to KRW 1.0T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0 to KRW 1.0T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 1-2T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Reported OP&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 82-86T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 87-92T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 94-98T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Base midpoint:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Midpoint&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about KRW 181T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Reported OP&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about KRW 89.5T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;OPM&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about 49.4%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Core OP&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 104-106T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="9-sensitivities"&gt;9. Sensitivities
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Variable&lt;/th&gt;
 &lt;th style="text-align: right"&gt;OP sensitivity&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM ASP +1ppt&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 0.4-0.5T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM bit growth +1ppt&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 0.4-0.5T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NAND ASP +1ppt&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+KRW 0.1-0.2T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Bonus/accrual +KRW 1T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 1.0T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DX cost +KRW 1T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 1.0T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Foundry/S.LSI loss +KRW 1T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-KRW 1.0T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The OP downgrade from KRW 100T to KRW 89T despite stronger DRAM/NAND price assumptions implies roughly KRW 12-13T of cost, accrual and non-memory drag.&lt;/p&gt;
&lt;h2 id="10-reported-op-and-core-op-matrix"&gt;10. Reported OP and Core OP matrix
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th style="text-align: right"&gt;Reported OP&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Bonus accrual&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Core OP&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;th&gt;Action&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;KRW 92T or higher&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about KRW 15T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 107T or higher&lt;/td&gt;
 &lt;td&gt;Clear beat&lt;/td&gt;
 &lt;td&gt;Buy possible&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;KRW 89-92T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about KRW 15T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 104-107T&lt;/td&gt;
 &lt;td&gt;In-line positive&lt;/td&gt;
 &lt;td&gt;Buy weakness&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;KRW 87-89T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about KRW 15T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 102-104T&lt;/td&gt;
 &lt;td&gt;Consensus area&lt;/td&gt;
 &lt;td&gt;Staged entry possible&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;KRW 84-87T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about KRW 15T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 99-102T&lt;/td&gt;
 &lt;td&gt;Weak miss&lt;/td&gt;
 &lt;td&gt;Needs explanation&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;KRW 84-87T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 25-35T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 109-122T&lt;/td&gt;
 &lt;td&gt;Headline miss, but neutral to positive if full-year pre-accrual&lt;/td&gt;
 &lt;td&gt;Verify cost nature&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;KRW 80-84T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;about KRW 15T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 95-99T&lt;/td&gt;
 &lt;td&gt;Cannot be explained by bonus alone&lt;/td&gt;
 &lt;td&gt;Wait or reduce&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;Below KRW 80T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;any&lt;/td&gt;
 &lt;td style="text-align: right"&gt;not meaningful&lt;/td&gt;
 &lt;td&gt;Thesis damage likely&lt;/td&gt;
 &lt;td&gt;Avoid&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="11-catalysts"&gt;11. Catalysts
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Event&lt;/th&gt;
 &lt;th&gt;Timing&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Importance&lt;/th&gt;
 &lt;th&gt;Watch item&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung 2Q26 preliminary results&lt;/td&gt;
 &lt;td&gt;likely early July 2026&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Very high&lt;/td&gt;
 &lt;td&gt;Reported OP at least KRW 87T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung 2Q26 final results and call&lt;/td&gt;
 &lt;td&gt;likely late July 2026&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Very high&lt;/td&gt;
 &lt;td&gt;Bonus accrual, DS core profit, 3Q outlook&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;3Q26 DRAM/NAND contract prices&lt;/td&gt;
 &lt;td&gt;July-August 2026&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Very high&lt;/td&gt;
 &lt;td&gt;Flat to up QoQ&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM4E, SOCAMM2, eSSD customer comments&lt;/td&gt;
 &lt;td&gt;2H26&lt;/td&gt;
 &lt;td style="text-align: right"&gt;High&lt;/td&gt;
 &lt;td&gt;Evidence of share gain&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Bonus-share buyback disclosure&lt;/td&gt;
 &lt;td&gt;TBD&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Medium-high&lt;/td&gt;
 &lt;td&gt;Distinguish compensation buyback from cancellation buyback&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron FY4Q26 results&lt;/td&gt;
 &lt;td&gt;likely late September 2026&lt;/td&gt;
 &lt;td style="text-align: right"&gt;High&lt;/td&gt;
 &lt;td&gt;Whether USD 50B revenue and USD 31 EPS are delivered&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="12-valuation"&gt;12. Valuation
&lt;/h2&gt;&lt;p&gt;The Kiwoom June 29 estimates summarized in the research input are:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2026F&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2027F&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 723.6T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 820.1T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Operating profit&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 372.9T&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 421.3T&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;EPS&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 43,429&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 48,177&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;PER&lt;/td&gt;
 &lt;td style="text-align: right"&gt;7.8x&lt;/td&gt;
 &lt;td style="text-align: right"&gt;7.0x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;PBR&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3.5x&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2.3x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;ROE&lt;/td&gt;
 &lt;td style="text-align: right"&gt;53.9%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;39.6%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;A KRW 430,000 target price implies 9.9x 2026F EPS and 8.9x 2027F EPS. This is valid only if 3Q26 OP remains above KRW 100T, 2027 EPS stays near KRW 48,000, bonus and stock compensation are already embedded in EPS, HBM4/eSSD contributions are visible, and DRAM/NAND prices do not fall sharply.&lt;/p&gt;
&lt;h2 id="13-red-team-and-invalidation"&gt;13. Red Team and invalidation
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Risk&lt;/th&gt;
 &lt;th&gt;What would invalidate the thesis&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Macro failure&lt;/td&gt;
 &lt;td&gt;Rates and credit spreads slow AI capex, hyperscaler capex guidance falls, AI server orders are delayed.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Micro failure&lt;/td&gt;
 &lt;td&gt;Samsung does not gain HBM4/eSSD share, HBM4 qualification is delayed, Foundry/S.LSI losses do not shrink.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Accounting failure&lt;/td&gt;
 &lt;td&gt;Reported OP falls below KRW 84T without enough bonus/accrual explanation, or Core OP falls below KRW 100T.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Capital allocation failure&lt;/td&gt;
 &lt;td&gt;Compensation buybacks are mistaken for shareholder return, while recurring profit-sharing erodes FY27 EPS and FCF.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Hard invalidation conditions:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Condition&lt;/th&gt;
 &lt;th&gt;Meaning&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Reported OP below KRW 84T and not explained by at least KRW 25T of pre-accrual&lt;/td&gt;
 &lt;td&gt;More than timing noise&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Core OP below KRW 100T&lt;/td&gt;
 &lt;td&gt;Core memory profit power weaker than expected&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;3Q26 OP outlook below KRW 100T&lt;/td&gt;
 &lt;td&gt;Peak-out risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;3Q DRAM/NAND prices fall QoQ&lt;/td&gt;
 &lt;td&gt;Pricing thesis damaged&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;No HBM4/eSSD share gains&lt;/td&gt;
 &lt;td&gt;2027 EPS and multiple expansion limited&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Foundry/S.LSI loss persists or widens&lt;/td&gt;
 &lt;td&gt;Samsung discount persists&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="final-view"&gt;Final view
&lt;/h2&gt;&lt;p&gt;At current prices, Micron’s surprise is almost gone from Samsung and SK Hynix. Samsung has returned to a price that barely embeds the Micron read-through. SK Hynix has almost returned to pre-event levels after a strong June 25 reaction.&lt;/p&gt;
&lt;p&gt;Samsung’s opportunity is now specific: if 2Q preliminary results show that the headline OP decline is mostly bonus-accrual timing, while Core OP is above KRW 100T and 3Q durability remains intact, the stock can reprice.&lt;/p&gt;
&lt;p&gt;The relative conclusion is straightforward:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Stock&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td&gt;Unreflected alpha if Core OP and 3Q durability are confirmed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td&gt;Pullback and reacceleration alpha, not a “cheap” stock&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron&lt;/td&gt;
 &lt;td&gt;Cleanest proof of the AI memory cycle, but already repriced&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;One sentence: Micron’s earnings were reflected in Korean memory names for one day, but most of that effect has been erased. For Samsung, the next battle is not headline 2Q OP but Core OP above KRW 100 trillion and 3Q durability.&lt;/p&gt;</description></item><item><title>HBF And HBC Commercialization Calendar: Milestones, Stock Triggers, And Misclassified Baskets</title><link>https://koreainvestinsights.com/post/hbf-hbc-commercialization-stock-trigger-catalyst-calendar-2026-06-28/</link><pubDate>Sun, 28 Jun 2026 23:20:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/hbf-hbc-commercialization-stock-trigger-catalyst-calendar-2026-06-28/</guid><description>&lt;h1 id="hbf-and-hbc-commercialization-calendar-milestones-stock-triggers-and-misclassified-baskets"&gt;HBF And HBC Commercialization Calendar: Milestones, Stock Triggers, And Misclassified Baskets
&lt;/h1&gt;&lt;p&gt;This note follows the earlier HBM, HBF and HBC technology comparison. The goal here is not to restate the technology. The goal is to separate timing, evidence, equity exposure and misclassified stock baskets.&lt;/p&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;HBF and HBC are real directions, but both are early.&lt;/li&gt;
&lt;li&gt;HBF should be read as second-half 2026 sample, early 2027 first inference-device sample, and 2028 or later revenue potential.&lt;/li&gt;
&lt;li&gt;HBC starts with Qualcomm AI250. Qualcomm AI200 is the pre-HBC product. AI250 with HBC Gen 1 is targeted for commercial sampling in mid-2027. Revenue relevance is more likely 2028 or later.&lt;/li&gt;
&lt;li&gt;The closest 2026 trigger is not revenue. It is physical evidence: HBF sample delivery and Qualcomm AI200 shipment.&lt;/li&gt;
&lt;li&gt;HBF equity exposure is narrow. SanDisk is the most direct listed option, but the stock is still primarily a NAND and data-center SSD cycle stock. SK Hynix has HBF exposure, but its stock is still driven by HBM. FADU is not a confirmed HBF beneficiary.&lt;/li&gt;
&lt;li&gt;HBC exposure is also narrow. Qualcomm is the home asset. In Korea, Samsung Electro-Mechanics is the most direct confirmed component exposure through FC-BGA substrates for Qualcomm AI200, based on media reports. Hanmi Semiconductor is an HBM and HBF equipment name, not an HBC name.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="why-the-distinction-matters"&gt;Why The Distinction Matters
&lt;/h2&gt;&lt;p&gt;HBF and HBC sound similar, but they are not the same layer.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;What it is&lt;/th&gt;
 &lt;th&gt;Current state&lt;/th&gt;
 &lt;th&gt;Key question&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM&lt;/td&gt;
 &lt;td&gt;High-bandwidth DRAM memory&lt;/td&gt;
 &lt;td&gt;Mass production&lt;/td&gt;
 &lt;td&gt;Can 2027 pricing and allocation hold?&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBF&lt;/td&gt;
 &lt;td&gt;NAND-based high-bandwidth flash layer&lt;/td&gt;
 &lt;td&gt;Standardization and pre-sample&lt;/td&gt;
 &lt;td&gt;Can samples and customer adoption appear?&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBC&lt;/td&gt;
 &lt;td&gt;Qualcomm accelerator architecture&lt;/td&gt;
 &lt;td&gt;Roadmap and design targets&lt;/td&gt;
 &lt;td&gt;Can Qualcomm win data-center customers?&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;SanDisk and SK Hynix have announced HBF standardization work under OCP. SanDisk has targeted first HBF samples in the second half of 2026 and first AI inference devices with HBF in early 2027. Qualcomm, meanwhile, describes HBC as a near-memory compute architecture inside its Dragonfly data-center roadmap, with HBC Gen 1 commercial sampling targeted with AI250 in mid-2027.&lt;/p&gt;
&lt;h2 id="commercialization-clock"&gt;Commercialization Clock
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Milestone&lt;/th&gt;
 &lt;th&gt;HBF&lt;/th&gt;
 &lt;th&gt;HBC&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Current state&lt;/td&gt;
 &lt;td&gt;Pre-sample. Most numbers are simulations or targets&lt;/td&gt;
 &lt;td&gt;No shipped HBC product. Public numbers are design targets&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;First physical evidence&lt;/td&gt;
 &lt;td&gt;HBF sample delivery in 2H26&lt;/td&gt;
 &lt;td&gt;AI200 shipment in 2H26, but AI200 is not HBC&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Defining point&lt;/td&gt;
 &lt;td&gt;First HBF inference-device sample in early 2027&lt;/td&gt;
 &lt;td&gt;AI250 with HBC Gen 1 sample in mid-2027&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Revenue timing&lt;/td&gt;
 &lt;td&gt;2028 or later, likely 2028-2030&lt;/td&gt;
 &lt;td&gt;2028 or later&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Largest swing factor&lt;/td&gt;
 &lt;td&gt;NVIDIA or major AI-platform adoption&lt;/td&gt;
 &lt;td&gt;Qualcomm data-center customer orders and revenue&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Direct listed exposure&lt;/td&gt;
 &lt;td&gt;SanDisk, with NAND caveat&lt;/td&gt;
 &lt;td&gt;Qualcomm; Samsung Electro-Mechanics as Korea component exposure&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The practical conclusion is simple: 2026 is a sample and shipment watch, 2027 is a validation watch, and 2028 or later is the revenue watch.&lt;/p&gt;
&lt;h2 id="hbf-stock-exposure"&gt;HBF Stock Exposure
&lt;/h2&gt;&lt;p&gt;SanDisk is the cleanest HBF equity option because it is directly involved in HBF standardization and sample timing. But it is not a pure HBF stock. The core drivers remain NAND pricing, data-center SSD demand and AI storage backlog. HBF is an asymmetric option layered on top of that.&lt;/p&gt;
&lt;p&gt;SK Hynix is involved in standardization, but HBF is not the main stock driver. HBM4 allocation, NVIDIA demand, DRAM pricing and memory operating margin remain far more important.&lt;/p&gt;
&lt;p&gt;FADU should not be put into the HBF basket without evidence of HBF products, contracts or logic-die supply. Its real discussion belongs under eSSD controllers, AI storage, ICMS and KV-cache offload. The company also carries legal uncertainty from the IPO-related case reported in late 2025 and 2026.&lt;/p&gt;
&lt;h2 id="hbc-stock-exposure"&gt;HBC Stock Exposure
&lt;/h2&gt;&lt;p&gt;Qualcomm is the home asset for HBC. But its data-center business is still small relative to its smartphone, automotive and broader QCT base. The main triggers are AI200 shipment, AI250 sampling, customer purchase orders, data-center revenue and independent benchmark validation.&lt;/p&gt;
&lt;p&gt;Samsung Electro-Mechanics is the most direct Korean component exposure. Media reports say it has begun producing FC-BGA substrates for Qualcomm AI200 at its Busan plant. This is exposure to Qualcomm&amp;rsquo;s data-center accelerator roadmap. It is not yet direct HBC Gen 1 exposure because AI200 is not HBC.&lt;/p&gt;
&lt;p&gt;Samsung Electronics, SK Hynix and Micron may have LPDDR relevance, but HBC-linked LPDDR is not as economically powerful as HBM. For Korean memory majors, HBM remains the main thesis.&lt;/p&gt;
&lt;p&gt;Hanmi Semiconductor is misclassified as an HBC beneficiary. Its TC bonder exposure belongs to HBM and potentially HBF, not Qualcomm HBC.&lt;/p&gt;
&lt;h2 id="priority-trigger-calendar"&gt;Priority Trigger Calendar
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th style="text-align: right"&gt;Priority&lt;/th&gt;
 &lt;th&gt;Theme&lt;/th&gt;
 &lt;th&gt;Trigger&lt;/th&gt;
 &lt;th&gt;Timing&lt;/th&gt;
 &lt;th&gt;Why it matters&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;1&lt;/td&gt;
 &lt;td&gt;HBF&lt;/td&gt;
 &lt;td&gt;First HBF sample delivery&lt;/td&gt;
 &lt;td&gt;2H26&lt;/td&gt;
 &lt;td&gt;First physical proof&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;2&lt;/td&gt;
 &lt;td&gt;HBC&lt;/td&gt;
 &lt;td&gt;AI200 shipment&lt;/td&gt;
 &lt;td&gt;2H26&lt;/td&gt;
 &lt;td&gt;First proof of Qualcomm data-center accelerator execution&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;3&lt;/td&gt;
 &lt;td&gt;HBC&lt;/td&gt;
 &lt;td&gt;Data-center revenue visibility&lt;/td&gt;
 &lt;td&gt;FY27&lt;/td&gt;
 &lt;td&gt;Most objective proof&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;4&lt;/td&gt;
 &lt;td&gt;HBF&lt;/td&gt;
 &lt;td&gt;NVIDIA stance&lt;/td&gt;
 &lt;td&gt;Unknown&lt;/td&gt;
 &lt;td&gt;Biggest two-way wildcard&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;5&lt;/td&gt;
 &lt;td&gt;HBC&lt;/td&gt;
 &lt;td&gt;AI250/HBC Gen 1 sample&lt;/td&gt;
 &lt;td&gt;Mid-2027&lt;/td&gt;
 &lt;td&gt;The defining HBC milestone&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;6&lt;/td&gt;
 &lt;td&gt;HBF&lt;/td&gt;
 &lt;td&gt;Anchor customer qualification&lt;/td&gt;
 &lt;td&gt;2027&lt;/td&gt;
 &lt;td&gt;From technology claim to customer adoption&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;7&lt;/td&gt;
 &lt;td&gt;Both&lt;/td&gt;
 &lt;td&gt;Independent benchmarks&lt;/td&gt;
 &lt;td&gt;2027+&lt;/td&gt;
 &lt;td&gt;Validation of design targets&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td style="text-align: right"&gt;8&lt;/td&gt;
 &lt;td&gt;HBC&lt;/td&gt;
 &lt;td&gt;Customer PO and deployment&lt;/td&gt;
 &lt;td&gt;2027-2028&lt;/td&gt;
 &lt;td&gt;Narrative turns into revenue&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="final-take"&gt;Final Take
&lt;/h2&gt;&lt;p&gt;HBF and HBC both address a real AI memory bottleneck. But the equity exposure is narrower than the theme suggests. HBF equals SanDisk option, with SK Hynix as a side branch. HBC equals Qualcomm, with Samsung Electro-Mechanics as the most direct Korean component exposure. FADU in HBF and Hanmi Semiconductor in HBC are weak classifications.&lt;/p&gt;
&lt;p&gt;The right framework is to buy confirmation, not vocabulary. In 2026, watch samples and shipments. In 2027, watch qualification and AI250. In 2028 and beyond, watch revenue.&lt;/p&gt;
&lt;p&gt;Sources include Qualcomm&amp;rsquo;s June 2026 data-center roadmap, SanDisk and SK Hynix HBF announcements, Samsung Electro-Mechanics product materials, and media reports on Samsung Electro-Mechanics and FADU. This is catalyst analysis, not investment advice.&lt;/p&gt;</description></item><item><title>What NVIDIA's Earnings Elasticity Says About Samsung And SK Hynix</title><link>https://koreainvestinsights.com/post/nvidia-earnings-elasticity-hbm-cycle-samsung-hynix-2026-06-28/</link><pubDate>Sun, 28 Jun 2026 22:00:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/nvidia-earnings-elasticity-hbm-cycle-samsung-hynix-2026-06-28/</guid><description>&lt;h1 id="what-nvidias-earnings-elasticity-says-about-samsung-and-sk-hynix"&gt;What NVIDIA&amp;rsquo;s Earnings Elasticity Says About Samsung And SK Hynix
&lt;/h1&gt;&lt;p&gt;This is a follow-up to the Micron FY3Q26 review, the Samsung-Hynix-Micron parity series, the HBM4E allocation note, and the Samsung catch-up work. The question is simple: if Samsung Electronics and SK Hynix still have strong earnings, can their stocks keep rising with the same elasticity?&lt;/p&gt;
&lt;p&gt;The NVIDIA case says no. Stocks do not react only to the level of earnings. They react to the rate of change, the second derivative of growth, and whether estimates are still being revised up faster than the stock has already moved.&lt;/p&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;NVIDIA&amp;rsquo;s first share-price elasticity slowdown started around calendar 3Q24. The more structural plateau began around calendar 4Q25 to 1Q26.&lt;/li&gt;
&lt;li&gt;Revenue kept exploding, but the stock started responding less to absolute growth and more to the durability of surprise.&lt;/li&gt;
&lt;li&gt;The Samsung and SK Hynix earnings cycle has not broken yet. Recent SK Hynix, Samsung, and Micron results still point to strong AI-memory demand.&lt;/li&gt;
&lt;li&gt;The stock-cycle risk is different. Samsung and SK Hynix entered a first overheated zone around June 2026.&lt;/li&gt;
&lt;li&gt;The main decision window is 4Q26 to 1Q27. Watch 2027 HBM contract pricing and volumes, HBM4 share, NVIDIA and hyperscaler AI capex, and memory operating-margin peak-out.&lt;/li&gt;
&lt;li&gt;My equity call is conditional buy for Samsung Electronics and wait for SK Hynix. SK Hynix is the higher-quality HBM leader, but it is now a crowded winner. Samsung still has HBM4 and Vera Rubin catch-up optionality.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="the-nvidia-lesson"&gt;The NVIDIA Lesson
&lt;/h2&gt;&lt;p&gt;NVIDIA was the cleanest listed-equity case of the AI semiconductor cycle. The stock moved first, then the earnings caught up. But once the market fully believed the earnings path, excellent results no longer produced the same stock response.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;NVIDIA reporting quarter&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Revenue growth&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Data-center growth&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Calendar-quarter stock move&lt;/th&gt;
 &lt;th&gt;Read-through&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;FY23 Q4, reported Feb 2023&lt;/td&gt;
 &lt;td style="text-align: right"&gt;YoY -21%, QoQ +2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;n/a&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2023 Q1 +90.1%&lt;/td&gt;
 &lt;td&gt;AI expectations moved before earnings&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY24 Q1, reported May 2023&lt;/td&gt;
 &lt;td style="text-align: right"&gt;YoY -13%, QoQ +19%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;DC revenue $4.28bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2023 Q2 +52.3%&lt;/td&gt;
 &lt;td&gt;Guidance shock&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY24 Q2, reported Aug 2023&lt;/td&gt;
 &lt;td style="text-align: right"&gt;YoY +101%, QoQ +88%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;YoY +171%, QoQ +141%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2023 Q3 +2.8%&lt;/td&gt;
 &lt;td&gt;First digestion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY24 Q4, reported Feb 2024&lt;/td&gt;
 &lt;td style="text-align: right"&gt;YoY +265%, QoQ +22%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;YoY +409%, QoQ +27%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2024 Q1 +82.5%&lt;/td&gt;
 &lt;td&gt;Blackwell confidence&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY25 Q2, reported Aug 2024&lt;/td&gt;
 &lt;td style="text-align: right"&gt;YoY +122%, QoQ +15%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;YoY +154%, QoQ +16%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2024 Q3 -1.7%&lt;/td&gt;
 &lt;td&gt;First inflection&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY26 Q3, reported Nov 2025&lt;/td&gt;
 &lt;td style="text-align: right"&gt;YoY +62%, QoQ +22%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;YoY +66%, QoQ +25%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2025 Q4 about 0.0%&lt;/td&gt;
 &lt;td&gt;Structural plateau begins&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY26 Q4, reported Feb 2026&lt;/td&gt;
 &lt;td style="text-align: right"&gt;YoY +73%, QoQ +20%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;YoY +75%, QoQ +22%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2026 Q1 -6.5%&lt;/td&gt;
 &lt;td&gt;Strong earnings, weak elasticity&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY27 Q1, reported May 2026&lt;/td&gt;
 &lt;td style="text-align: right"&gt;YoY +85%, QoQ +20%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;YoY +92%, QoQ +21%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2026 Q2 to Jun 26 +10.4%&lt;/td&gt;
 &lt;td&gt;Earnings follow-through, limited multiple expansion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;NVIDIA FY27 Q1 revenue was $81.6bn and data-center revenue was $75.2bn. The FY27 Q2 guide was $91.0bn, plus or minus 2%. Those are still exceptional numbers. Yet from September 2025 to June 2026, the stock barely moved compared with its earlier rise.&lt;/p&gt;
&lt;p&gt;The lesson for Korean memory equities is clear: strong earnings are not enough. The stock has to be cheaper than the next estimate revision path.&lt;/p&gt;
&lt;h2 id="why-samsung-and-sk-hynix-have-not-broken-yet"&gt;Why Samsung And SK Hynix Have Not Broken Yet
&lt;/h2&gt;&lt;p&gt;The earnings cycle is still intact.&lt;/p&gt;
&lt;p&gt;SK Hynix reported 1Q26 revenue of KRW 52.5763tn, operating profit of KRW 37.6103tn, and net income of KRW 40.3459tn. Samsung reported 1Q26 consolidated revenue of KRW 133.9tn and operating profit of KRW 57.2tn, with DS operating profit at KRW 53.7tn. Micron&amp;rsquo;s FY3Q26 was also very strong, with revenue of $41.456bn, non-GAAP EPS of $25.11, and non-GAAP gross margin of 84.9%.&lt;/p&gt;
&lt;p&gt;That does not look like a demand collapse. It looks like a profitable shortage.&lt;/p&gt;
&lt;p&gt;The problem is not the earnings cycle. The problem is the stock cycle. SK Hynix became South Korea&amp;rsquo;s most valuable company in June 2026, and Reuters reported that the stock was up more than 340% year-to-date. At that point, the market no longer needs proof that SK Hynix is an HBM winner. It needs proof that 2027 EPS estimates still have room to move up.&lt;/p&gt;
&lt;h2 id="the-timing-framework"&gt;The Timing Framework
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Stage&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Expected timing&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;th&gt;Key signal&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;First stock-elasticity slowdown&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Already around June 2026&lt;/td&gt;
 &lt;td&gt;Overheated zone&lt;/td&gt;
 &lt;td&gt;SK Hynix market-cap leadership, retail FOMO, target-price chase&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;EPS-revision peak&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4Q26 to 1Q27&lt;/td&gt;
 &lt;td&gt;Most important window&lt;/td&gt;
 &lt;td&gt;2027 HBM contract price and volume disappoint or confirm&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Earnings peak-out&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1H27 to 2H27&lt;/td&gt;
 &lt;td&gt;Base-case fundamental risk window&lt;/td&gt;
 &lt;td&gt;HBM ASP growth slows and memory margin rolls over&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Full memory downcycle&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2H27 to 2028&lt;/td&gt;
 &lt;td&gt;Not yet confirmed&lt;/td&gt;
 &lt;td&gt;Supply catches demand and inventories build&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The stock can peak before earnings. That is the central point.&lt;/p&gt;
&lt;h2 id="five-break-conditions"&gt;Five Break Conditions
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Break condition&lt;/th&gt;
 &lt;th&gt;What it means&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM ASP flattens or declines&lt;/td&gt;
 &lt;td&gt;The first sign is likely price momentum slowing, not demand disappearing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix margin falls for two straight quarters&lt;/td&gt;
 &lt;td&gt;Peak earnings risk becomes observable&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung and Micron reduce SK Hynix&amp;rsquo;s HBM premium&lt;/td&gt;
 &lt;td&gt;The monopoly premium becomes a leadership premium&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NVIDIA data-center QoQ growth falls into single digits for two quarters&lt;/td&gt;
 &lt;td&gt;AI capex is no longer accelerating fast enough&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2027 supply catches demand&lt;/td&gt;
 &lt;td&gt;HBM4 and HBM4E ramp creates a new pricing regime&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="stock-implications"&gt;Stock Implications
&lt;/h2&gt;&lt;p&gt;Samsung Electronics is the cleaner conditional buy. The point is not that Samsung&amp;rsquo;s memory business is good. The point is that its HBM laggard discount can narrow if HBM4 and SOCAMM2 revenue show up in DS margin.&lt;/p&gt;
&lt;p&gt;SK Hynix remains the quality leader, but new buying is less attractive. The stock now needs more than a strong 2026. It needs 2027 HBM contracts to prove that EPS can still be revised up.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Stock&lt;/th&gt;
 &lt;th&gt;View&lt;/th&gt;
 &lt;th&gt;Why&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td&gt;Conditional buy&lt;/td&gt;
 &lt;td&gt;HBM4 and Vera Rubin catch-up can reduce the laggard discount&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td&gt;Wait&lt;/td&gt;
 &lt;td&gt;Best HBM asset, but crowded and closer to peak-expectation risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NVIDIA&lt;/td&gt;
 &lt;td&gt;Watchlist&lt;/td&gt;
 &lt;td&gt;AI leader, but earnings growth and share-price elasticity have separated&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM equipment and substrates&lt;/td&gt;
 &lt;td&gt;Selective watch&lt;/td&gt;
 &lt;td&gt;Alpha can move down the bottleneck stack if megacap memory stalls&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="final-view"&gt;Final View
&lt;/h2&gt;&lt;p&gt;Samsung and SK Hynix are not in an earnings breakdown. They are in a stock-elasticity test. The right question is no longer whether HBM is good. The right question is whether 2027 EPS revisions can outrun the stock price.&lt;/p&gt;
&lt;p&gt;That makes Samsung Electronics the more asymmetric Korean memory expression today. SK Hynix is higher quality, but more fully believed.&lt;/p&gt;
&lt;p&gt;Sources: &lt;a class="link" href="https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-first-quarter-fiscal-2027" target="_blank" rel="noopener"
 &gt;NVIDIA FY27 Q1&lt;/a&gt;, &lt;a class="link" href="https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-fourth-quarter-and-fiscal-2026" target="_blank" rel="noopener"
 &gt;NVIDIA FY26 Q4&lt;/a&gt;, &lt;a class="link" href="https://news.samsungsemiconductor.com/global/samsung-electronics-announces-first-quarter-2026-results/" target="_blank" rel="noopener"
 &gt;Samsung 1Q26&lt;/a&gt;, &lt;a class="link" href="https://news.skhynix.com/q1-2026-business-results/" target="_blank" rel="noopener"
 &gt;SK Hynix 1Q26&lt;/a&gt;, &lt;a class="link" href="https://investors.micron.com/news-releases/news-release-details/micron-technology-inc-reports-record-results-third-quarter" target="_blank" rel="noopener"
 &gt;Micron FY3Q26&lt;/a&gt;, &lt;a class="link" href="https://www.reuters.com/world/asia-pacific/sk-hynix-overtakes-samsung-become-koreas-most-valuable-company-2026-06-22/" target="_blank" rel="noopener"
 &gt;Reuters on SK Hynix&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;&lt;small&gt;Disclaimer: This is research and information only, not investment advice.&lt;/small&gt;&lt;/p&gt;</description></item><item><title>Micron FY3Q26 Earnings: AI Memory Bottlenecks, 84.9% Gross Margin, And A $50B Guide</title><link>https://koreainvestinsights.com/post/micron-fy3q26-ai-memory-sca-fcf-hold-2026-06-25/</link><pubDate>Thu, 25 Jun 2026 15:20:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/micron-fy3q26-ai-memory-sca-fcf-hold-2026-06-25/</guid><description>&lt;h1 id="micron-fy3q26-earnings-ai-memory-bottlenecks-849-gross-margin-and-a-50b-guide"&gt;Micron FY3Q26 Earnings: AI Memory Bottlenecks, 84.9% Gross Margin, And A $50B Guide
&lt;/h1&gt;&lt;p&gt;This note follows the Samsung-Hynix-Micron parity series, the AI chip and memory P/E map, the SK Hynix versus Micron comparison, and the recent memory-price-risk work around CXMT. The central question is no longer whether Micron is an AI-memory proxy. FY3Q26 showed that the proxy now has earnings, gross margin, FCF, and customer commitments behind it.&lt;/p&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;p&gt;Micron&amp;rsquo;s FY3Q26 was not just an earnings beat. Revenue reached $41.456bn, non-GAAP EPS was $25.11, non-GAAP gross margin was 84.9%, and FY4Q26 guidance called for $50.0bn of revenue and $31.00 of non-GAAP EPS. These figures are from Micron&amp;rsquo;s official release.&lt;/p&gt;
&lt;p&gt;The key change is threefold.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Micron can no longer be treated as a plain commodity-memory equity. Management framed memory bandwidth and capacity as structural bottlenecks for AI systems, and data-center revenue exceeded $25bn in the quarter.&lt;/li&gt;
&lt;li&gt;The company has signed 16 strategic customer agreements, or SCAs. For 14 of those contracts, Micron pointed to roughly $100bn of minimum revenue at floor prices and about $22bn of customer deposits and financing commitments. Take-or-pay terms and price floors reduce, but do not eliminate, the earnings-collapse risk that normally defines memory cycles.&lt;/li&gt;
&lt;li&gt;The stock already reacted. After the release, after-hours quotes showed a double-digit rally, roughly in the mid-teens depending on the data vendor. The numbers were excellent, but the margin of safety narrowed quickly.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;My conclusion is &lt;strong&gt;HOLD, with new buying better left for pullbacks&lt;/strong&gt;. The revised 12-month target is $1,350. Versus the roughly $1,214 after-hours reference price used in this note, that implies about 11% upside. The upside looks larger versus the regular close of $1,048.51, but that price did not fully include the earnings release.&lt;/p&gt;
&lt;h2 id="event-check"&gt;Event Check
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Status&lt;/th&gt;
 &lt;th style="text-align: right"&gt;KST Timing&lt;/th&gt;
 &lt;th&gt;Evidence&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Recent earnings release&lt;/td&gt;
 &lt;td&gt;Confirmed&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2026-06-25 05:01&lt;/td&gt;
 &lt;td&gt;Micron released FY3Q26 results at 2026-06-24 16:01 ET. FY3Q26 ended 2026-05-28.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Earnings call&lt;/td&gt;
 &lt;td&gt;Confirmed&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2026-06-25 05:30&lt;/td&gt;
 &lt;td&gt;Micron scheduled its Third Quarter 2026 Financial Call at 4:30 PM ET.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Prepared remarks and deck&lt;/td&gt;
 &lt;td&gt;Confirmed&lt;/td&gt;
 &lt;td style="text-align: right"&gt;After release&lt;/td&gt;
 &lt;td&gt;Micron posted the FY3Q26 presentation and prepared remarks on its IR site.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Post-earnings analyst call&lt;/td&gt;
 &lt;td&gt;Confirmed&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2026-06-25 07:00&lt;/td&gt;
 &lt;td&gt;Reflected in the event schedule used for this review.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Next earnings date&lt;/td&gt;
 &lt;td&gt;Not officially confirmed&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Not fixed&lt;/td&gt;
 &lt;td&gt;Vendor estimates differ around late September 2026.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The FY3Q26 event is confirmed. FY4Q26 is not. Until Micron publishes the next date, late September after the US close is only an estimate.&lt;/p&gt;
&lt;h2 id="fy3q26-numbers"&gt;FY3Q26 Numbers
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Metric&lt;/th&gt;
 &lt;th style="text-align: right"&gt;FY3Q26&lt;/th&gt;
 &lt;th style="text-align: right"&gt;FY2Q26&lt;/th&gt;
 &lt;th style="text-align: right"&gt;FY3Q25&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$41.456bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$23.860bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$9.301bn&lt;/td&gt;
 &lt;td&gt;+73.7% QoQ, +345.7% YoY&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;GAAP net income / EPS&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$28.243bn / $24.67&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$13.785bn / $12.07&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$1.885bn / $1.68&lt;/td&gt;
 &lt;td&gt;Extreme operating leverage&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Non-GAAP net income / EPS&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$28.857bn / $25.11&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$14.021bn / $12.20&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$2.181bn / $1.91&lt;/td&gt;
 &lt;td&gt;Non-GAAP EPS +105.8% QoQ and +1,214.7% YoY&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;GAAP gross margin&lt;/td&gt;
 &lt;td style="text-align: right"&gt;84.6%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;74.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;37.7%&lt;/td&gt;
 &lt;td&gt;Pricing, mix, and shortage effect&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Non-GAAP gross margin&lt;/td&gt;
 &lt;td style="text-align: right"&gt;84.9%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;74.9%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;39.0%&lt;/td&gt;
 &lt;td&gt;+10.0pp QoQ and +45.9pp YoY&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Non-GAAP operating income / margin&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$33.681bn / 81.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$16.455bn / 69.0%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$2.490bn / 26.8%&lt;/td&gt;
 &lt;td&gt;More than 13 times the year-ago level&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Operating cash flow&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$25.39bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$11.90bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$4.61bn&lt;/td&gt;
 &lt;td&gt;Cash generation followed earnings&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Capex, net&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$7.084bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$5.004bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$2.660bn&lt;/td&gt;
 &lt;td&gt;Investment rising with demand&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Adjusted free cash flow&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$18.304bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$6.899bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$1.949bn&lt;/td&gt;
 &lt;td&gt;Record free cash flow&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cash, marketable investments, restricted cash&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$30.2bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;not provided&lt;/td&gt;
 &lt;td style="text-align: right"&gt;not provided&lt;/td&gt;
 &lt;td&gt;Buffer for capex and capital return&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The quality of the beat matters. GAAP EPS and non-GAAP EPS moved together. Operating income, operating cash flow, and adjusted FCF also moved together. This was not a one-line adjustment story.&lt;/p&gt;
&lt;h2 id="segment-data"&gt;Segment Data
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Segment&lt;/th&gt;
 &lt;th style="text-align: right"&gt;FY3Q26 Revenue&lt;/th&gt;
 &lt;th style="text-align: right"&gt;FY2Q26 Revenue&lt;/th&gt;
 &lt;th style="text-align: right"&gt;FY3Q25 Revenue&lt;/th&gt;
 &lt;th style="text-align: right"&gt;FY3Q26 Gross Margin&lt;/th&gt;
 &lt;th style="text-align: right"&gt;FY3Q26 Operating Margin&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Cloud Memory&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$13.769bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$7.749bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$3.386bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;83%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;78%&lt;/td&gt;
 &lt;td&gt;Core HBM and high-end DRAM growth engine&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Core Data Center&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$11.524bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$5.687bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$1.530bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;87%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;83%&lt;/td&gt;
 &lt;td&gt;Data-center DRAM and SSD demand was very strong&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Mobile and Client&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$11.521bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$7.711bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$3.255bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;87%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;86%&lt;/td&gt;
 &lt;td&gt;PC and mobile also benefited from pricing and mix&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Automotive and Embedded&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$4.634bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$2.708bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$1.127bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;79%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;75%&lt;/td&gt;
 &lt;td&gt;Auto and embedded also participated&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Cloud Memory plus Core Data Center reached about $25.3bn. That matches management&amp;rsquo;s comment that data-center revenue exceeded $25bn, or more than a $100bn annualized run-rate. This point is crucial: the quarter was not only about HBM. DRAM, NAND, SSD, client, mobile, auto, and embedded all showed pricing power.&lt;/p&gt;
&lt;h2 id="fy4q26-guidance"&gt;FY4Q26 Guidance
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Company Guidance&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$50.0bn ± $1.0bn&lt;/td&gt;
 &lt;td&gt;$49bn to $51bn range&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;GAAP gross margin&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 86%&lt;/td&gt;
 &lt;td&gt;Record-level margin sustained&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Non-GAAP gross margin&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 86%&lt;/td&gt;
 &lt;td&gt;Up again from 84.9% in FY3Q&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;GAAP EPS&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$30.73 ± $1.00&lt;/td&gt;
 &lt;td&gt;Close to non-GAAP EPS&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Non-GAAP EPS&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$31.00 ± $1.00&lt;/td&gt;
 &lt;td&gt;$30 to $32 range&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Non-GAAP operating expenses&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About $1.65bn&lt;/td&gt;
 &lt;td&gt;Strong operating leverage&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Diluted shares&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About 1.15bn&lt;/td&gt;
 &lt;td&gt;EPS denominator&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Reuters reported that the $50bn revenue guide was well above the LSEG consensus near $43.58bn. The guide says that FY3Q was not necessarily the peak. Still, management also noted that the rate of price increases can moderate, even if pricing remains favorable.&lt;/p&gt;
&lt;h2 id="consensus-beat"&gt;Consensus Beat
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Actual Or Guide&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Consensus / Estimate&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Difference&lt;/th&gt;
 &lt;th&gt;Reading&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;FY3Q revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$41.46bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$35.69bn, Investing.com&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+16.2%&lt;/td&gt;
 &lt;td&gt;Beat&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY3Q adjusted EPS&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$25.11&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$20.49, Investing.com&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+22.6%&lt;/td&gt;
 &lt;td&gt;Beat&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY3Q revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$41.46bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$35.91bn, MarketBeat&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+$5.54bn&lt;/td&gt;
 &lt;td&gt;Beat&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY3Q EPS&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$25.11&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$20.98, MarketBeat&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+$4.13&lt;/td&gt;
 &lt;td&gt;Beat&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Non-GAAP gross margin&lt;/td&gt;
 &lt;td style="text-align: right"&gt;84.9%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;81.6%, LSEG preview&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+3.3pp&lt;/td&gt;
 &lt;td&gt;Beat&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY4Q revenue guide&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$50.0bn ± $1.0bn&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$43.58bn, LSEG via Reuters&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Large beat&lt;/td&gt;
 &lt;td&gt;Beat and raise&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY4Q EPS guide&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$31.00 ± $1.00&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$25.84, Reuters preview&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Above&lt;/td&gt;
 &lt;td&gt;Estimates need to move up&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Consensus sources differ, but the conclusion is stable: Micron beat on revenue, EPS, gross margin, and the next-quarter guide.&lt;/p&gt;
&lt;h2 id="the-strategic-customer-agreement-point"&gt;The Strategic Customer Agreement Point
&lt;/h2&gt;&lt;p&gt;The most important non-earnings item was the SCA structure.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;SCA Item&lt;/th&gt;
 &lt;th&gt;Management Detail&lt;/th&gt;
 &lt;th&gt;Investment Meaning&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Number of agreements&lt;/td&gt;
 &lt;td&gt;16&lt;/td&gt;
 &lt;td&gt;Broad, not one-customer dependent&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Typical duration&lt;/td&gt;
 &lt;td&gt;Mostly 2026 to 2030, with shorter automotive structures&lt;/td&gt;
 &lt;td&gt;Longer visibility than a normal memory cycle&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Volume coverage&lt;/td&gt;
 &lt;td&gt;About 20% of DRAM volume and one-third of NAND volume&lt;/td&gt;
 &lt;td&gt;Already material&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Potential revenue coverage&lt;/td&gt;
 &lt;td&gt;Could approach more than half of revenue once fully signed&lt;/td&gt;
 &lt;td&gt;Commodity discount can shrink&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Pricing&lt;/td&gt;
 &lt;td&gt;Take-or-pay, floors, ceilings, some fixed or market-based pricing&lt;/td&gt;
 &lt;td&gt;Downside protection and some upside limits&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;RPO&lt;/td&gt;
 &lt;td&gt;More than $5bn at FY3Q end, with about $100bn for 14 signed SCAs expected in filings&lt;/td&gt;
 &lt;td&gt;Minimum revenue and floor-price framing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Deposits and commitments&lt;/td&gt;
 &lt;td&gt;Around $22bn, including about $18bn of cash deposits&lt;/td&gt;
 &lt;td&gt;Strong customer commitment&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The SCA changes the debate. It gives Micron a floor under part of the cycle. It also proves that customers care enough about supply assurance to put money and obligations behind the contracts. But it is not a free option. Price ceilings can cap part of the upside when spot markets rise sharply, and customer deposits sit in financing cash flow rather than free cash flow.&lt;/p&gt;
&lt;h2 id="what-management-and-analysts-focused-on"&gt;What Management And Analysts Focused On
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Topic&lt;/th&gt;
 &lt;th&gt;Core Message&lt;/th&gt;
 &lt;th&gt;Why It Matters&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Supply tightness&lt;/td&gt;
 &lt;td&gt;DRAM and NAND demand is above industry supply, with tightness expected beyond 2027&lt;/td&gt;
 &lt;td&gt;The bull case is supply discipline and capacity bottlenecks&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM4&lt;/td&gt;
 &lt;td&gt;High-volume shipments for a lead customer, more qualification samples, and more than $1bn of HBM4 revenue&lt;/td&gt;
 &lt;td&gt;HBM4 is already commercial, not just a future narrative&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM4E&lt;/td&gt;
 &lt;td&gt;Development for calendar 2027 volume production&lt;/td&gt;
 &lt;td&gt;The next allocation battle is already starting&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Deposits&lt;/td&gt;
 &lt;td&gt;Around $22bn of customer commitments, about $18bn in cash deposits&lt;/td&gt;
 &lt;td&gt;Customer behavior confirms scarcity&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Capex&lt;/td&gt;
 &lt;td&gt;FY4Q capex around $10bn, FY2026 capex around $27bn, FY2027 quarterly capex above FY4Q&lt;/td&gt;
 &lt;td&gt;Growth needs capital and can pressure FCF later&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Gross margin&lt;/td&gt;
 &lt;td&gt;86% guide, but price increases may moderate&lt;/td&gt;
 &lt;td&gt;The next issue is sustainability, not the absolute beat&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Buybacks&lt;/td&gt;
 &lt;td&gt;Capital return remains centered on repurchases, but no precise cadence was committed&lt;/td&gt;
 &lt;td&gt;FCF conversion and buyback scale are the next equity catalysts&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The call was not mainly about whether FY3Q beat. It was about how durable the new profit level is.&lt;/p&gt;
&lt;h2 id="upside-and-downside-triggers"&gt;Upside And Downside Triggers
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Upside Trigger&lt;/th&gt;
 &lt;th&gt;Timing&lt;/th&gt;
 &lt;th&gt;Mechanism&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;FY4Q revenue above $51bn or EPS above $32&lt;/td&gt;
 &lt;td&gt;FY4Q26 earnings&lt;/td&gt;
 &lt;td&gt;FY2027 run-rate estimates move up&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;More FY2027 HBM3E/HBM4 allocation fixed, faster HBM4E qualification&lt;/td&gt;
 &lt;td&gt;FY4Q call and second-half product updates&lt;/td&gt;
 &lt;td&gt;Longer shortage duration supports higher margins&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SCA revenue coverage moves toward half or more of revenue&lt;/td&gt;
 &lt;td&gt;Next one to two quarters&lt;/td&gt;
 &lt;td&gt;Cyclicality discount shrinks&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Large FY4Q FCF increase plus clearer buyback scale&lt;/td&gt;
 &lt;td&gt;FY4Q and 10-K&lt;/td&gt;
 &lt;td&gt;EPS accretion and FCF yield re-rating&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Non-HBM DRAM and NAND pricing remains strong&lt;/td&gt;
 &lt;td&gt;ASP and segment margin updates&lt;/td&gt;
 &lt;td&gt;Reduces HBM concentration risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Data-center revenue run-rate stays above $100bn&lt;/td&gt;
 &lt;td&gt;FY4Q and FY2027&lt;/td&gt;
 &lt;td&gt;Micron is reclassified as an AI infrastructure beneficiary&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Downside Trigger&lt;/th&gt;
 &lt;th&gt;Timing&lt;/th&gt;
 &lt;th&gt;Mechanism&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;FY4Q revenue or EPS misses the midpoint of guidance&lt;/td&gt;
 &lt;td&gt;FY4Q26 earnings&lt;/td&gt;
 &lt;td&gt;Beat-and-raise thesis weakens&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Gross margin rolls over from 86%&lt;/td&gt;
 &lt;td&gt;FY4Q and FY2027 guide&lt;/td&gt;
 &lt;td&gt;Memory supercycle multiple compresses&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;FY2027 capex consumes too much FCF&lt;/td&gt;
 &lt;td&gt;FY2026 10-K and FY2027 capex guide&lt;/td&gt;
 &lt;td&gt;FCF multiple and buyback hopes fall&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SCA ceilings cap upside&lt;/td&gt;
 &lt;td&gt;Contract and margin commentary&lt;/td&gt;
 &lt;td&gt;Agreements look like caps, not floors&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AI customers shift to cheaper memory structures&lt;/td&gt;
 &lt;td&gt;Accelerator product cycles&lt;/td&gt;
 &lt;td&gt;Premium DRAM and HBM pricing power weakens&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Chinese memory makers penetrate higher-end products&lt;/td&gt;
 &lt;td&gt;Late 2026 to 2027&lt;/td&gt;
 &lt;td&gt;Supply and price competition return&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="stock-reaction"&gt;Stock Reaction
&lt;/h2&gt;&lt;p&gt;FY3Q26 was released after the US close on 24 June 2026. Therefore the first regular session that fully reflects the news is 25 June 2026. At the original analysis time, that session had not opened.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Window&lt;/th&gt;
 &lt;th style="text-align: right"&gt;MU Price / Return&lt;/th&gt;
 &lt;th&gt;Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;2026-06-24 regular session&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$1,048.51, -0.31% using the supplied reference&lt;/td&gt;
 &lt;td&gt;Pre-release trading, not the event reaction&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;MarketBeat regular close&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$1,037.93, -1.32%&lt;/td&gt;
 &lt;td&gt;Vendor difference&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;After-hours supplied reference&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$1,213.97, +15.78%&lt;/td&gt;
 &lt;td&gt;Strong immediate reaction&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;MarketBeat after-hours&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$1,221.62, +17.70%&lt;/td&gt;
 &lt;td&gt;Another vendor reference&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;1D regular session&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Pending&lt;/td&gt;
 &lt;td&gt;Available after 25 June close&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2D regular session&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Pending&lt;/td&gt;
 &lt;td&gt;Available after 26 June close&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;5D regular session&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Pending&lt;/td&gt;
 &lt;td&gt;Available next week&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The reaction was driven by more than EPS. The FY4Q revenue guide, EPS guide, SCA commitments, long-duration supply tightness, and the FCF profile all mattered.&lt;/p&gt;
&lt;h2 id="valuation-and-recommendation"&gt;Valuation And Recommendation
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;View&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Rating&lt;/td&gt;
 &lt;td&gt;HOLD&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;New buying&lt;/td&gt;
 &lt;td&gt;Wait for pullback&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;12-month target&lt;/td&gt;
 &lt;td&gt;$1,350&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Reference price&lt;/td&gt;
 &lt;td&gt;About $1,213.97 after hours&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Upside&lt;/td&gt;
 &lt;td&gt;About +11.2%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Upside versus regular close&lt;/td&gt;
 &lt;td&gt;About +28.8%, but that price did not fully reflect the release&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The valuation frame is forward P/E. FY2026 non-GAAP EPS can be approximated as Q1 $4.78 plus Q2 $12.20 plus Q3 $25.11 plus the FY4Q guide midpoint of $31.00, or $73.09. For the base case, I use FY2027 EPS of about $100. That is below the FY4Q26 run-rate to reflect FY2027 capex, start-up costs, and price normalization risk. The base multiple is 13.5 times.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Scenario&lt;/th&gt;
 &lt;th style="text-align: right"&gt;EPS Assumption&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Multiple&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Target Range&lt;/th&gt;
 &lt;th&gt;Key Condition&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Bear&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$70-80&lt;/td&gt;
 &lt;td style="text-align: right"&gt;11-12x&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$800-950&lt;/td&gt;
 &lt;td&gt;Supply improves, ASP falls, capex and FCF pressure return&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Base&lt;/td&gt;
 &lt;td style="text-align: right"&gt;About $100&lt;/td&gt;
 &lt;td style="text-align: right"&gt;13.5x&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$1,350&lt;/td&gt;
 &lt;td&gt;FY4Q guide is achieved, HBM/SCA remain strong, FCF holds&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Bull&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$110-120&lt;/td&gt;
 &lt;td style="text-align: right"&gt;15x&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$1,650-1,800&lt;/td&gt;
 &lt;td&gt;2027 supply sold-out extends, buybacks accelerate, SCA coverage rises&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="what-would-change-the-view"&gt;What Would Change The View
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Change&lt;/th&gt;
 &lt;th&gt;Condition&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Upgrade to BUY&lt;/td&gt;
 &lt;td&gt;Stock pulls back below $1,100, or FY4Q revenue exceeds $51bn and EPS exceeds $32 while FY2027 FCF visibility remains intact&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Downgrade to SELL / reduce&lt;/td&gt;
 &lt;td&gt;Gross margin falls quickly below 80%, HBM4/HBM4E qualification is delayed, SCA additions disappoint, or customers shift toward cheaper memory architectures&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Raise target&lt;/td&gt;
 &lt;td&gt;SCA-linked revenue approaches more than half of total revenue and FY2027 EPS visibility moves above $110&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Lower target&lt;/td&gt;
 &lt;td&gt;FY2027 capex materially constrains cash return and the expected FY4Q FCF step-up does not materialize&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="korean-semiconductor-read-through"&gt;Korean Semiconductor Read-Through
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Korean Investor Question&lt;/th&gt;
 &lt;th&gt;Micron FY3Q26 Answer&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Did Samsung Electronics and SK Hynix EPS risk rise?&lt;/td&gt;
 &lt;td&gt;No. Micron&amp;rsquo;s DRAM, NAND, HBM, and SSD numbers point to strong memory demand and pricing.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Is the Micron premium more justified now?&lt;/td&gt;
 &lt;td&gt;Partly yes. 84.9% non-GAAP gross margin, the 86% FY4Q guide, and SCA structures reduce the commodity-memory discount.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Does this hurt SK Hynix&amp;rsquo;s exclusivity premium?&lt;/td&gt;
 &lt;td&gt;It raises competitive pressure, but the HBM total addressable market is also expanding.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;What does it mean for Samsung Electronics catch-up?&lt;/td&gt;
 &lt;td&gt;Samsung must prove HBM4E qualification, DS margin recovery, and customer allocation if it wants the discount to close.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;What does it mean for Korean suppliers?&lt;/td&gt;
 &lt;td&gt;HBM, SSD, SOCAMM, high-capacity DDR5, test, packaging, and power-integrity demand remain supported.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The implication is not one-sided. Micron proved that the US-listed AI-memory premium has real earnings behind it. At the same time, if Korean memory EPS is not impaired, the relative discount of Samsung Electronics and SK Hynix versus Micron may remain too wide.&lt;/p&gt;
&lt;h2 id="next-quarter-checklist"&gt;Next Quarter Checklist
&lt;/h2&gt;&lt;ol&gt;
&lt;li&gt;Does FY4Q26 revenue reach the $50bn guide and EPS reach the $31 guide?&lt;/li&gt;
&lt;li&gt;Do HBM3E and HBM4 remain fully booked through 2027?&lt;/li&gt;
&lt;li&gt;Does HBM4E qualification stay on schedule?&lt;/li&gt;
&lt;li&gt;How much revenue visibility do SCAs actually create?&lt;/li&gt;
&lt;li&gt;Does the price floor protect margin downside more than the ceiling limits upside?&lt;/li&gt;
&lt;li&gt;Can FY2027 capex rise while FCF and buybacks remain credible?&lt;/li&gt;
&lt;li&gt;Does non-HBM DRAM and NAND pricing remain strong?&lt;/li&gt;
&lt;li&gt;Do AI customers continue to demand more memory capacity and bandwidth, or do they shift toward cheaper memory architectures?&lt;/li&gt;
&lt;li&gt;Do Chinese memory suppliers or large customers weaken Micron pricing power?&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id="sources-and-limits"&gt;Sources And Limits
&lt;/h2&gt;&lt;p&gt;Core sources include Micron&amp;rsquo;s FY3Q26 official release, SEC Exhibit 99.1, the FY3Q26 presentation, prepared remarks, the Micron IR events page, Reuters coverage, MarketWatch live coverage, and the MarketBeat/Quartr transcript mirror. The Q&amp;amp;A transcript was treated as a web fallback and cross-checked against official materials where possible.&lt;/p&gt;
&lt;p&gt;Regular-session 1D, 2D, and 5D event returns were not available at the analysis time because the release came after the US close. FY4Q26&amp;rsquo;s official date was also not yet confirmed. Data vendors showed different late-September estimates.&lt;/p&gt;
&lt;h2 id="machine-readable-summary"&gt;Machine-Readable Summary
&lt;/h2&gt;&lt;div class="highlight"&gt;&lt;pre tabindex="0" style="color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;-webkit-text-size-adjust:none;"&gt;&lt;code class="language-json" data-lang="json"&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;{
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;as_of_kst&amp;#34;&lt;/span&gt;: &lt;span style="color:#e6db74"&gt;&amp;#34;2026-06-25 13:59&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;company&amp;#34;&lt;/span&gt;: {
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;name&amp;#34;&lt;/span&gt;: &lt;span style="color:#e6db74"&gt;&amp;#34;Micron Technology, Inc.&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;ticker&amp;#34;&lt;/span&gt;: &lt;span style="color:#e6db74"&gt;&amp;#34;MU&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;exchange&amp;#34;&lt;/span&gt;: &lt;span style="color:#e6db74"&gt;&amp;#34;NASDAQ&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;currency&amp;#34;&lt;/span&gt;: &lt;span style="color:#e6db74"&gt;&amp;#34;USD&amp;#34;&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; },
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;recent_earnings_release&amp;#34;&lt;/span&gt;: {
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;release_datetime_kst&amp;#34;&lt;/span&gt;: &lt;span style="color:#e6db74"&gt;&amp;#34;2026-06-25 05:01&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;confirmed&amp;#34;&lt;/span&gt;: &lt;span style="color:#66d9ef"&gt;true&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; },
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;recent_call&amp;#34;&lt;/span&gt;: {
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;exists&amp;#34;&lt;/span&gt;: &lt;span style="color:#66d9ef"&gt;true&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;call_datetime_kst&amp;#34;&lt;/span&gt;: &lt;span style="color:#e6db74"&gt;&amp;#34;2026-06-25 05:30&amp;#34;&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; },
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;upcoming_earnings_release&amp;#34;&lt;/span&gt;: {
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;release_datetime_kst&amp;#34;&lt;/span&gt;: &lt;span style="color:#66d9ef"&gt;null&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;confirmed&amp;#34;&lt;/span&gt;: &lt;span style="color:#66d9ef"&gt;false&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; },
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;recommendation&amp;#34;&lt;/span&gt;: {
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;rating&amp;#34;&lt;/span&gt;: &lt;span style="color:#e6db74"&gt;&amp;#34;HOLD&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;target_price&amp;#34;&lt;/span&gt;: &lt;span style="color:#ae81ff"&gt;1350&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;horizon_months&amp;#34;&lt;/span&gt;: &lt;span style="color:#ae81ff"&gt;12&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;upside_pct&amp;#34;&lt;/span&gt;: &lt;span style="color:#ae81ff"&gt;11.2&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; },
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;key_triggers_up&amp;#34;&lt;/span&gt;: [
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#e6db74"&gt;&amp;#34;FY4Q26 revenue above $51B or non-GAAP EPS above $32&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#e6db74"&gt;&amp;#34;Additional FY2027 HBM3E/HBM4 allocation locked in&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#e6db74"&gt;&amp;#34;SCA revenue coverage approaches half or more of company revenue&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#e6db74"&gt;&amp;#34;FY4Q free cash flow rises sharply and buyback scale becomes explicit&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#e6db74"&gt;&amp;#34;Non-HBM DRAM and NAND pricing remains strong&amp;#34;&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; ],
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#f92672"&gt;&amp;#34;key_triggers_down&amp;#34;&lt;/span&gt;: [
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#e6db74"&gt;&amp;#34;FY4Q26 results miss the midpoint of company guidance&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#e6db74"&gt;&amp;#34;Gross margin rolls over materially from the 86% guide&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#e6db74"&gt;&amp;#34;FY2027 capex reduces free cash flow and limits buybacks&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#e6db74"&gt;&amp;#34;SCA ceiling prices cap upside more than floor prices protect downside&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#e6db74"&gt;&amp;#34;AI customers shift toward cheaper memory architectures&amp;#34;&lt;/span&gt;,
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; &lt;span style="color:#e6db74"&gt;&amp;#34;Supply catches up earlier than expected and pricing power weakens&amp;#34;&lt;/span&gt;
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; ]
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;}
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;h2 id="portfolio-manager-comment"&gt;Portfolio-Manager Comment
&lt;/h2&gt;&lt;p&gt;This is not a normal memory rebound. It is pricing power attached to an AI infrastructure bottleneck. The numbers are excellent, but the market recognized that immediately. SCA makes Micron a better business than the old pure commodity-memory model, but memory remains memory: if supply returns too quickly, the stock can still be brutal. I prefer holding or buying on pullbacks rather than chasing the first after-hours reaction.&lt;/p&gt;
&lt;p&gt;Sources: &lt;a class="link" href="https://www.globenewswire.com/news-release/2026/06/24/3317151/14450/en/micron-technology-inc-reports-record-results-for-the-third-quarter-of-fiscal-2026.html" target="_blank" rel="noopener"
 &gt;Micron FY3Q26 official release&lt;/a&gt;, &lt;a class="link" href="https://investors.micron.com/quarterly-results" target="_blank" rel="noopener"
 &gt;Micron quarterly results&lt;/a&gt;, &lt;a class="link" href="https://investors.micron.com/static-files/2354ecda-77a0-4ddd-8462-a631eb491356" target="_blank" rel="noopener"
 &gt;Micron FY3Q26 presentation&lt;/a&gt;, &lt;a class="link" href="https://investors.micron.com/static-files/631b1a32-5537-46ae-8f40-82e42fc79dfe" target="_blank" rel="noopener"
 &gt;Micron prepared remarks&lt;/a&gt;, &lt;a class="link" href="https://www.marketbeat.com/earnings/reports/2026-6-24-micron-technology-inc-stock/" target="_blank" rel="noopener"
 &gt;MarketBeat/Quartr transcript&lt;/a&gt;, &lt;a class="link" href="https://www.thestar.com.my/tech/tech-news/2026/06/25/micron-forecasts-strong-quarterly-results-on-soaring-memory-chip-demand" target="_blank" rel="noopener"
 &gt;Reuters via The Star&lt;/a&gt;, &lt;a class="link" href="https://m.investing.com/news/stock-market-news/micron-forecasts-quarterly-revenue-above-estimates-4758884?ampMode=1" target="_blank" rel="noopener"
 &gt;Reuters via Investing.com&lt;/a&gt;, &lt;a class="link" href="https://www.marketwatch.com/livecoverage/micron-earnings-stock-results-memory-guidance/card/micron-s-spending-is-moving-higher-in-the-face-of-booming-demand-mLjzhU62y5AJfaUStOEC" target="_blank" rel="noopener"
 &gt;MarketWatch capex note&lt;/a&gt;, &lt;a class="link" href="https://www.marketwatch.com/livecoverage/micron-earnings-stock-results-memory-guidance/card/micron-has-no-clue-when-memory-supply-will-finally-catch-up-with-demand-cKcl56PcCiQzIAvLYTnx" target="_blank" rel="noopener"
 &gt;MarketWatch supply note&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;&lt;em&gt;Disclaimer: For research and information purposes only. Not investment advice.&lt;/em&gt;&lt;/p&gt;</description></item></channel></rss>