<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Ai-Hbm-2026 on Korea Invest Insights</title><link>https://koreainvestinsights.com/series/ai-hbm-2026/</link><description>Recent content in Ai-Hbm-2026 on Korea Invest Insights</description><generator>Hugo -- gohugo.io</generator><language>en</language><copyright>koreainvestinsights.com · @korea_invest_insights</copyright><lastBuildDate>Fri, 31 Jul 2026 12:23:31 +0900</lastBuildDate><atom:link href="https://koreainvestinsights.com/series/ai-hbm-2026/feed.xml" rel="self" type="application/rss+xml"/><item><title>Did One Hedge Fund Cause the July AI Rout? Fact-Checking Situational Awareness and the SK hynix ADR</title><link>https://koreainvestinsights.com/post/situational-awareness-liquidation-sk-hynix-adr-july-ai-rout-fact-check-2026-07-31/</link><pubDate>Fri, 31 Jul 2026 12:15:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/situational-awareness-liquidation-sk-hynix-adr-july-ai-rout-fact-check-2026-07-31/</guid><description>&lt;p&gt;A compelling explanation spread after AI infrastructure stocks staged an extraordinary rebound at the end of July 2026. The story was that Situational Awareness, the hedge fund led by former OpenAI researcher Leopold Aschenbrenner, had taken a $7 billion cornerstone position in the SK hynix ADR, received margin calls, and transferred a $16 billion public-equity book to Citadel. The fund&amp;rsquo;s liquidation supposedly explained both the July selloff and the July 31 rebound.&lt;/p&gt;
&lt;p&gt;That narrative mixes confirmed facts, credible reporting, a material misreading of the SEC prospectus, and causal claims that cannot yet be proved. The best-supported conclusion is narrower: a large transfer of Situational Awareness&amp;rsquo;s public-equity book appears to have occurred, and forced deleveraging likely amplified moves in crowded AI infrastructure stocks. But the fund did not publicly commit to a $7 billion SK hynix allocation by itself, and no public dataset can attribute 40% to 50% of the KOSPI decline to one fund.&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Data cutoff: U.S. returns use the July 30, 2026 close. Korean prices are intraday snapshots at 12:12 p.m. KST on July 31. Private-fund AUM, leverage, prime-broker relationships, and margin calls have limited public disclosure, so this report separates SEC filings from media reports and inference. It follows our &lt;a class="link" href="https://koreainvestinsights.com/en/post/china-duv-steelman-verdict-black-tuesday-korea-memory-2026-07-28/" &gt;July 28 Korean memory selloff analysis&lt;/a&gt; and &lt;a class="link" href="https://koreainvestinsights.com/en/post/ai-capex-eight-doubts-rebound-or-terminal-48hours-2026-07-29/" &gt;eight questions behind the AI CAPEX correction&lt;/a&gt;. More research is available in the &lt;a class="link" href="https://koreainvestinsights.com/en/page/korea-semiconductor-hbm-kospi-hub/" &gt;AI and HBM hub&lt;/a&gt; and &lt;a class="link" href="https://koreainvestinsights.com/en/page/exclusive-analysis-hub/" &gt;Exclusive Analysis 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;The Wall Street Journal and Axios reported that Situational Awareness sold most or all of its public-equity portfolio to Citadel after large losses. The direction of the transaction is supported by credible reporting, although the exact value and terms are not publicly filed.&lt;/li&gt;
&lt;li&gt;The SEC prospectus does not say Situational Awareness alone received a $7 billion SK hynix allocation. Baillie Gifford, Coatue-managed funds, and Situational Awareness collectively expressed non-binding interest in up to $7 billion. Individual allocations were not disclosed.&lt;/li&gt;
&lt;li&gt;Q1 2026 filings confirm meaningful exposure to Bloom Energy, CoreWeave, and SanDisk. A May Schedule 13G confirms a 5.6% Nebius stake. Micron and Nvidia exposure, however, included substantial options. The portfolio was not a simple long-only AI basket.&lt;/li&gt;
&lt;li&gt;On July 30, NBIS rose 27.13%, BE 26.49%, SNDK 25.99%, CRWV 21.51%, MU 18.36%, and SKHY 17.52%, versus 2.65% for NVDA. This is consistent with forced-seller removal and short covering, but it is not a controlled experiment proving a single-fund cause.&lt;/li&gt;
&lt;li&gt;A separate claim that Korea suffered KRW 37.7 trillion of forced margin liquidation is also incorrect. That figure appears to refer to outstanding margin-credit balances, not executed forced sales.&lt;/li&gt;
&lt;li&gt;The most defensible verdict is that Situational Awareness was likely an important marginal seller and amplifier inside a leveraged, crowded factor. Its book transfer can support short-term normalization, but it does not answer the fundamental question of AI and memory earnings after 2028.&lt;/li&gt;
&lt;/ul&gt;
&lt;div class="thesis-callout"&gt;
 &lt;div class="thesis-callout__label"&gt;Verdict&lt;/div&gt;
 &lt;div class="thesis-callout__body"&gt;
 The public-equity transfer is well reported. The alleged standalone $7 billion SK hynix allocation and a 40% to 50% attribution of the July decline are not established. This is best understood as a market-structure event that amplified price moves, not proof that fundamentals caused or ended the rout.
 &lt;/div&gt;
&lt;/div&gt;
&lt;h2 id="1-claim-by-claim-verdict"&gt;1. Claim-by-claim verdict
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Claim&lt;/th&gt;
 &lt;th&gt;Verdict&lt;/th&gt;
 &lt;th&gt;What the evidence says&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Situational Awareness transferred its public-equity book to Citadel&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;Largely supported&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;WSJ reported most of the book; Axios reported all of it&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AUM fell from $20 billion to $10 billion&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;Reported, not independently filed&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;Bloomberg and follow-on reports cited the decline&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;The fund used up to four times leverage&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;Reported, not filed&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;Media reports cite the figure; 13F cannot establish leverage&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Goldman Sachs, JPMorgan, and BofA issued margin calls&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;Reported, not filed&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;Prime-broker terms and margin-call records are private&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;The fund alone bought $7 billion of SKHY&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;False&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;$7 billion was the aggregate, non-binding indication from three investors&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SKHY triggered the margin call&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;Plausible but unproved&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;Timing fits, but allocation, entry price, and collateral terms are unknown&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;The fund was concentrated in SKHY, MU, CRWV, NBIS, BE, and SNDK&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;Partly supported&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;Q1 13F and May 13G support several names; SKHY was not yet listed; MU exposure was mixed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;The fund did not own NVDA&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;False&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;Q1 13F reported a small common position and large put exposure&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;The rebound was a controlled experiment proving causality&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;Overstated&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;Betas, options, shorts, earnings, and macro variables also moved&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Korea saw KRW 37.7 trillion of forced liquidation&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;False&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;The figure appears to be outstanding credit balances&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;One fund caused 40% to 50% of the July decline&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;Not measurable&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;Daily fund sales, swaps, hedges, and a counterfactual are unavailable&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="2-the-central-correction-7-billion-was-not-one-funds-allocation"&gt;2. The central correction: $7 billion was not one fund&amp;rsquo;s allocation
&lt;/h2&gt;&lt;p&gt;SK hynix&amp;rsquo;s final SEC prospectus covered 177.9 million ADSs at $149 each, for a gross offering of roughly $26.5 billion and estimated net proceeds of about $26.2 billion.&lt;/p&gt;
&lt;p&gt;The prospectus named three investors:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Baillie Gifford Overseas Limited&lt;/li&gt;
&lt;li&gt;funds managed by Coatue&lt;/li&gt;
&lt;li&gt;Situational Awareness Partners LP&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;They had, severally and not jointly, indicated interest in purchasing up to an aggregate of $7 billion of ADSs. The indications were explicitly non-binding, and each investor could buy more, less, or none. &lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/2120882/000119312526299963/d32785d424b4.htm" target="_blank" rel="noopener"
 &gt;SK hynix final SEC prospectus&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;The public record therefore establishes only:&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;Aggregate indication ceiling from three investors: $7 billion
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;Individual allocations: undisclosed
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;Situational Awareness allocation: undisclosed
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;Binding commitment: no
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;Situational Awareness may have received a large allocation. It may also have received far less than $7 billion. Treating the aggregate indication as a single-fund position inflated every downstream conclusion about leverage, the margin call, and market impact.&lt;/p&gt;
&lt;h2 id="3-what-is-actually-known-about-the-fund"&gt;3. What is actually known about the fund
&lt;/h2&gt;&lt;h3 id="31-the-public-equity-transfer"&gt;3.1 The public-equity transfer
&lt;/h3&gt;&lt;p&gt;The &lt;a class="link" href="https://www.wsj.com/finance/citadel-buys-situational-awarenesss-stock-portfolio-after-big-losses-in-ai-5117159b" target="_blank" rel="noopener"
 &gt;Wall Street Journal&lt;/a&gt; reported that Situational Awareness sold the bulk of its stock portfolio to Citadel after deep AI-related losses. &lt;a class="link" href="https://www.axios.com/2026/07/30/ai-hedge-fund-situational-awareness-citadel" target="_blank" rel="noopener"
 &gt;Axios&lt;/a&gt; cited one source saying it sold the entire public-equity portfolio.&lt;/p&gt;
&lt;p&gt;The source discrepancy matters. The safest wording is that most or all of the public-equity book was transferred. That does not establish that the fund itself was liquidated. Private holdings and separate vehicles may remain, and no public announcement says the manager has ceased operations.&lt;/p&gt;
&lt;p&gt;Some secondary reports put the transferred book at $16 billion. There is no public transaction filing showing whether this figure was market value, gross exposure, or some other measure. The trade&amp;rsquo;s existence is more reliable than the exact amount.&lt;/p&gt;
&lt;h3 id="32-aum-returns-leverage-and-margin-calls"&gt;3.2 AUM, returns, leverage, and margin calls
&lt;/h3&gt;&lt;p&gt;Media reports described assets rising above $20 billion and then falling to about $10 billion. Other reports cited a 439% net return through June, more than 1,000% since inception, up to four times leverage, and margin calls from multiple prime brokers.&lt;/p&gt;
&lt;p&gt;These are reported figures, not values that can be reconstructed from public SEC filings. Form 13F does not disclose:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;net asset value&lt;/li&gt;
&lt;li&gt;cash and borrowings&lt;/li&gt;
&lt;li&gt;gross and net leverage&lt;/li&gt;
&lt;li&gt;OTC derivatives and swaps&lt;/li&gt;
&lt;li&gt;prime-broker collateral&lt;/li&gt;
&lt;li&gt;monthly P&amp;amp;L and redemptions&lt;/li&gt;
&lt;li&gt;private-asset valuations&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The media reporting may be accurate, but it should not be presented as audited public fact.&lt;/p&gt;
&lt;h2 id="4-what-the-13f-and-13g-filings-show"&gt;4. What the 13F and 13G filings show
&lt;/h2&gt;&lt;p&gt;Situational Awareness&amp;rsquo;s Q1 2026 13F contained 42 reportable entries with aggregate reported value of about $13.68 billion. Option entries can reflect the market value of underlying securities rather than option premium or delta exposure. The total is therefore neither AUM nor a measure of net risk. &lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/2045724/000204572426000008/salp13fq1xml.xml" target="_blank" rel="noopener"
 &gt;Situational Awareness Q1 2026 13F&lt;/a&gt;&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Security&lt;/th&gt;
 &lt;th style="text-align: right"&gt;March 31 filing&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;Bloom Energy&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$878.7M common plus $55.3M call underlying value&lt;/td&gt;
 &lt;td&gt;Large directional exposure&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CoreWeave&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$556.1M common plus $140.6M calls&lt;/td&gt;
 &lt;td&gt;Material AI infrastructure long&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SanDisk&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$724.4M common plus $388.8M calls&lt;/td&gt;
 &lt;td&gt;Material storage exposure&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$5.9M common, $422.3M calls, $583.7M puts&lt;/td&gt;
 &lt;td&gt;Mixed options book, not a clean long&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Nvidia&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Small common position plus $1.568B of put underlying value&lt;/td&gt;
 &lt;td&gt;Material bearish option exposure&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;A more recent Schedule 13G reported 12,410,060 Nebius shares, or 5.6%, as of May 19. &lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/2045724/000093583626000303/primary_doc.xml" target="_blank" rel="noopener"
 &gt;Nebius Schedule 13G&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;These filings support the view that the fund had large AI infrastructure, power, and storage exposure. They also show why a simple held-versus-not-held table is insufficient. Micron included both calls and puts, and Nvidia was associated with substantial put exposure.&lt;/p&gt;
&lt;p&gt;SKHY cannot be verified from the Q1 filing because the ADR listed in July. A broad Q3 holdings check will not arrive until mid-November. A Schedule 13D or 13G could appear earlier if an investor crossed the 5% threshold, but current public filings do not disclose Situational Awareness&amp;rsquo;s allocation.&lt;/p&gt;
&lt;h2 id="5-the-rebound-strong-evidence-not-a-controlled-experiment"&gt;5. The rebound: strong evidence, not a controlled experiment
&lt;/h2&gt;&lt;h3 id="us-close-on-july-30"&gt;U.S. close on July 30
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Security&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Return&lt;/th&gt;
 &lt;th&gt;Public filing connection&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Nebius&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+27.13%&lt;/td&gt;
 &lt;td&gt;5.6% May 13G stake&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Bloom Energy&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+26.49%&lt;/td&gt;
 &lt;td&gt;Q1 common and calls&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SanDisk&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+25.99%&lt;/td&gt;
 &lt;td&gt;Q1 common and calls&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CoreWeave&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+21.51%&lt;/td&gt;
 &lt;td&gt;Q1 common and calls&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+18.36%&lt;/td&gt;
 &lt;td&gt;Q1 common, calls, and puts&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SKHY ADR&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+17.52%&lt;/td&gt;
 &lt;td&gt;Fund allocation undisclosed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SOXX&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+8.55%&lt;/td&gt;
 &lt;td&gt;Semiconductor benchmark&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Nvidia&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+2.65%&lt;/td&gt;
 &lt;td&gt;Q1 common and substantial puts&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h3 id="korea-intraday-at-1212-pm-kst-on-july-31"&gt;Korea intraday at 12:12 p.m. KST on July 31
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Asset&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Price&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Return&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 250,500&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+21.01%&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 1,645,000&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+24.43%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KOSPI&lt;/td&gt;
 &lt;td style="text-align: right"&gt;6,389.04&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+14.22%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The cross-section is consistent with the removal of a forced seller. It may also reflect short covering and options hedging in the names that had suffered the largest declines.&lt;/p&gt;
&lt;p&gt;It is not a controlled experiment for five reasons:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;NBIS, BE, and CRWV have much higher beta than Nvidia.&lt;/li&gt;
&lt;li&gt;A block transfer to Citadel is different from all shares being sold in the open market.&lt;/li&gt;
&lt;li&gt;Short covering and dealer gamma can magnify the rebound.&lt;/li&gt;
&lt;li&gt;The fund&amp;rsquo;s net exposure is unknown because strikes, maturities, deltas, and swaps are not disclosed.&lt;/li&gt;
&lt;li&gt;FOMC news, hyperscaler earnings, macro risk appetite, and Korean program flows moved at the same time.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;The strongest defensible conclusion is that the rebound supports a forced-flow amplification hypothesis. It does not identify the original trigger or measure one fund&amp;rsquo;s share of the decline.&lt;/p&gt;
&lt;h2 id="6-the-krw-377-trillion-margin-liquidation-error"&gt;6. The KRW 37.7 trillion margin-liquidation error
&lt;/h2&gt;&lt;p&gt;Outstanding margin credit, collateral shortfalls, forced-sale orders, and executed forced sales are different statistics.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Metric&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;Margin-credit balance&lt;/td&gt;
 &lt;td&gt;Borrowed purchase funding still outstanding&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Collateral shortfall&lt;/td&gt;
 &lt;td&gt;Amount below required maintenance margin&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Forced-sale order&lt;/td&gt;
 &lt;td&gt;Broker order submitted for repayment&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Executed forced sale&lt;/td&gt;
 &lt;td&gt;Portion of orders actually filled&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The publicly cited KRW 37.7 trillion figure appears to refer to outstanding margin-credit balances, not executed forced sales over July 24 to 28. Treating the entire stock of credit as liquidation volume materially exaggerates market impact.&lt;/p&gt;
&lt;p&gt;The same caution applies to ELS knock-in hedging. Futures selling alone cannot identify ELS hedges without product-level delta and gamma data.&lt;/p&gt;
&lt;h2 id="7-a-more-defensible-causal-chain"&gt;7. A more defensible causal chain
&lt;/h2&gt;&lt;p&gt;The following mechanism fits the evidence:&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;High-beta AI infrastructure stocks decline
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ collateral capacity deteriorates in leveraged portfolios
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ public-equity exposure is reduced or block-transferred
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ crowded-factor pressure intensifies
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ removal of the urgent seller becomes public
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ short covering, dealer hedging, and high-beta rebound
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;What remains unknown is crucial:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Situational Awareness&amp;rsquo;s actual SKHY allocation and purchase price&lt;/li&gt;
&lt;li&gt;the position that triggered the first margin call&lt;/li&gt;
&lt;li&gt;prime-broker liquidation timing and size&lt;/li&gt;
&lt;li&gt;security-level quantities and hedges transferred to Citadel&lt;/li&gt;
&lt;li&gt;daily July net sales by the fund&lt;/li&gt;
&lt;li&gt;actual Korean forced-sale and ELS hedge execution&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;SKHY may have contributed to the stress. But the public record does not prove that it was the trigger. Losses on multiple AI infrastructure longs, software shorts moving against the fund, and derivatives collateral could have combined.&lt;/p&gt;
&lt;h2 id="8-why-a-40-to-50-market-attribution-is-not-possible"&gt;8. Why a 40% to 50% market attribution is not possible
&lt;/h2&gt;&lt;p&gt;To assign a share of the KOSPI or AI-factor decline to one fund, an analyst would need:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;security-level sale quantities and execution prices&lt;/li&gt;
&lt;li&gt;SKHY ADR versus 000660 arbitrage flows&lt;/li&gt;
&lt;li&gt;futures, options, and swap hedges&lt;/li&gt;
&lt;li&gt;executed Korean margin-liquidation data&lt;/li&gt;
&lt;li&gt;issuer-level ELS hedge data&lt;/li&gt;
&lt;li&gt;high-frequency event windows&lt;/li&gt;
&lt;li&gt;a credible counterfactual group&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;None is publicly complete. Larger drawdowns and rebounds in fund-linked names show an amplification channel, not a market-wide contribution percentage.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Factor&lt;/th&gt;
 &lt;th&gt;Current assessment&lt;/th&gt;
 &lt;th&gt;Confidence&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Forced reduction of leveraged AI infrastructure exposure&lt;/td&gt;
 &lt;td&gt;Meaningful amplifier&lt;/td&gt;
 &lt;td&gt;Medium-high&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SKHY as the single initial trigger&lt;/td&gt;
 &lt;td&gt;Possible, unproved&lt;/td&gt;
 &lt;td&gt;Medium-low&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Korean credit and derivative amplification&lt;/td&gt;
 &lt;td&gt;Direction plausible, size unknown&lt;/td&gt;
 &lt;td&gt;Medium&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;China DUV, CXMT, and vendor-financing concerns&lt;/td&gt;
 &lt;td&gt;Supplied a narrative for repricing&lt;/td&gt;
 &lt;td&gt;Medium&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Macro, rates, and earnings&lt;/td&gt;
 &lt;td&gt;Independent concurrent drivers&lt;/td&gt;
 &lt;td&gt;Medium&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;One fund explaining half the KOSPI decline&lt;/td&gt;
 &lt;td&gt;Unsupported&lt;/td&gt;
 &lt;td&gt;Low&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="9-investment-implications"&gt;9. Investment implications
&lt;/h2&gt;&lt;h3 id="91-what-improved"&gt;9.1 What improved
&lt;/h3&gt;&lt;p&gt;If the public-equity transfer occurred as reported, an urgently liquidity-constrained marginal seller is likely less active. That supports a reversal of the portion of the July decline unrelated to fundamentals. The extraordinary rebound suggests that normalization had begun.&lt;/p&gt;
&lt;h3 id="92-what-did-not-improve"&gt;9.2 What did not improve
&lt;/h3&gt;&lt;p&gt;The following questions remain open:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Can AI token demand absorb memory supply growth after 2028?&lt;/li&gt;
&lt;li&gt;Can HBM4 and HBM4E market share and pricing hold?&lt;/li&gt;
&lt;li&gt;Will commodity DRAM and NAND move into oversupply?&lt;/li&gt;
&lt;li&gt;Can hyperscaler cash flow fund CAPEX and long-term purchase commitments?&lt;/li&gt;
&lt;li&gt;How will Chinese memory expansion and U.S. controls split global pricing?&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The rebound is a flow signal, not an earnings validation.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Group&lt;/th&gt;
 &lt;th&gt;Meaning of this event&lt;/th&gt;
 &lt;th&gt;Next evidence&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;Potential easing of ADR and leverage overhang; allocation unknown&lt;/td&gt;
 &lt;td&gt;Q3 13F/13G, ADR borrow and volume, HBM earnings&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td&gt;Technical normalization through the same memory factor&lt;/td&gt;
 &lt;td&gt;HBM qualification, commodity DRAM profit, program flows&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron&lt;/td&gt;
 &lt;td&gt;Mixed disclosed options make a simple forced-selling thesis difficult&lt;/td&gt;
 &lt;td&gt;Open interest, long-term contracts, guidance&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NBIS, BE, CRWV, SNDK&lt;/td&gt;
 &lt;td&gt;Forced-flow and short-covering effects likely material&lt;/td&gt;
 &lt;td&gt;Volume normalization, block trades, earnings and funding&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Nvidia&lt;/td&gt;
 &lt;td&gt;Q1 filing showed significant fund put exposure&lt;/td&gt;
 &lt;td&gt;AI demand, dealer positioning, receivables and customer concentration&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="10-citadel-is-not-permanent-unlevered-capital"&gt;10. Citadel is not permanent, unlevered capital
&lt;/h2&gt;&lt;p&gt;Describing Citadel as unlevered patient capital would be inaccurate. Citadel is a sophisticated hedge fund and trading ecosystem capable of rapidly pricing and hedging risk.&lt;/p&gt;
&lt;p&gt;The positive interpretation is narrower: positions moved from a seller that reportedly needed immediate liquidity to a much larger risk absorber. Citadel can still:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;hedge with single-stock, index, and options positions&lt;/li&gt;
&lt;li&gt;distribute shares as liquidity returns&lt;/li&gt;
&lt;li&gt;run relative-value trades&lt;/li&gt;
&lt;li&gt;retain some positions and sell others&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The overhang may have shifted from urgent to manageable. It did not necessarily disappear.&lt;/p&gt;
&lt;h2 id="11-what-to-monitor"&gt;11. What to monitor
&lt;/h2&gt;&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Schedules 13D and 13G&lt;/strong&gt; for any investor crossing 5% of SKHY.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Q3 2026 13F filings&lt;/strong&gt; in mid-November for broad SKHY and AI infrastructure changes.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;ADR versus local-share pricing&lt;/strong&gt;, borrow costs, volume, and conversion economics.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Options open interest and implied volatility&lt;/strong&gt; to assess whether dealer and short-covering pressure is fading.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Follow-on block trades&lt;/strong&gt; that could indicate redistribution of the Citadel book.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;2027-2028 earnings estimates&lt;/strong&gt;, the final test of whether price normalization is supported by cash flow.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id="12-red-team"&gt;12. Red Team
&lt;/h2&gt;&lt;h3 id="the-fund-was-an-effect-not-a-cause"&gt;The fund was an effect, not a cause
&lt;/h3&gt;&lt;p&gt;AI infrastructure may have been repriced because of valuation, rates, power constraints, and vendor-financing concerns. Situational Awareness could have been a casualty of that move rather than its origin.&lt;/p&gt;
&lt;h3 id="forced-selling-may-not-be-over"&gt;Forced selling may not be over
&lt;/h3&gt;&lt;p&gt;Citadel may hedge or distribute the acquired positions, while other leveraged funds could face margin calls. One transfer does not cleanse an entire crowded factor.&lt;/p&gt;
&lt;h3 id="the-rebound-may-be-mostly-short-covering"&gt;The rebound may be mostly short covering
&lt;/h3&gt;&lt;p&gt;A one-day 18% to 27% rebound can be driven by short covering and options gamma. If volume fades and prices revisit prior lows, the normalization interpretation weakens.&lt;/p&gt;
&lt;h3 id="invalidation"&gt;Invalidation
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Future SEC filings show little or no SKHY position at Situational Awareness.&lt;/li&gt;
&lt;li&gt;The public-equity sale reports are denied by the parties.&lt;/li&gt;
&lt;li&gt;Large follow-on block sales occur after the Citadel transfer.&lt;/li&gt;
&lt;li&gt;AI infrastructure and memory earnings estimates decline structurally.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="13-evidence-classification"&gt;13. Evidence classification
&lt;/h2&gt;&lt;h3 id="fact"&gt;[Fact]
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;SK hynix offered 177.9 million ADSs at $149.&lt;/li&gt;
&lt;li&gt;Three investors collectively indicated non-binding interest in up to $7 billion.&lt;/li&gt;
&lt;li&gt;Q1 filings show BE, CRWV, SNDK, MU, and NVDA exposures.&lt;/li&gt;
&lt;li&gt;A May Schedule 13G reported a 5.6% NBIS stake.&lt;/li&gt;
&lt;li&gt;AI infrastructure and Korean memory stocks rebounded sharply.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="reported"&gt;[Reported]
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Most or all of the public-equity book was transferred to Citadel.&lt;/li&gt;
&lt;li&gt;Fund assets fell to roughly $10 billion.&lt;/li&gt;
&lt;li&gt;The fund used up to four times leverage and faced margin calls.&lt;/li&gt;
&lt;li&gt;The transferred book was approximately $16 billion.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="inference"&gt;[Inference]
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;The transfer likely reduced urgent forced-selling pressure.&lt;/li&gt;
&lt;li&gt;Leveraged AI infrastructure positioning likely amplified the July decline.&lt;/li&gt;
&lt;li&gt;Citadel may have converted an urgent overhang into a more manageable one.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="blocked"&gt;[Blocked]
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Actual SKHY allocation and entry price&lt;/li&gt;
&lt;li&gt;Prime-broker collateral and margin-call records&lt;/li&gt;
&lt;li&gt;Exact transaction value, security quantities, and Citadel hedges&lt;/li&gt;
&lt;li&gt;Daily July net sales and total fund net exposure&lt;/li&gt;
&lt;li&gt;Executed Korean margin liquidations and ELS hedges&lt;/li&gt;
&lt;li&gt;The fund&amp;rsquo;s exact contribution to the KOSPI and AI-factor decline&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="conclusion"&gt;Conclusion
&lt;/h2&gt;&lt;p&gt;The Situational Awareness episode is an important piece of the July AI infrastructure selloff. The reported transfer of a large public-equity book and the explosive rebound in high-beta names support the view that forced deleveraging distorted prices.&lt;/p&gt;
&lt;p&gt;But three corrections are essential. The $7 billion was not Situational Awareness&amp;rsquo;s disclosed standalone SK hynix allocation. KRW 37.7 trillion was not verified forced-liquidation volume. And the rebound, while powerful evidence of flow normalization, was not a controlled experiment assigning 40% to 50% of the market decline to one fund.&lt;/p&gt;
&lt;p&gt;The most accurate summary is:&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Situational Awareness has not been proved to be the sole cause of the July AI rout. It was likely an important marginal seller and amplifier inside a leveraged, crowded market structure. The public-equity transfer can help short-term normalization, but the durability of AI and memory earnings after 2028 still requires separate fundamental proof.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;p&gt;This report is market-structure research based on public information, not a recommendation to buy or sell any security.&lt;/p&gt;</description></item><item><title>NVIDIA Vera CPU Production Expansion and Simmtech: SoCAMM2 is About Module Units, Not Bits</title><link>https://koreainvestinsights.com/post/nvidia-vera-cpu-socamm2-simmtech-module-unit-thesis-2026-07-19/</link><pubDate>Sun, 19 Jul 2026 18:05:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/nvidia-vera-cpu-socamm2-simmtech-module-unit-thesis-2026-07-19/</guid><description>&lt;p&gt;One NVIDIA Vera Rubin NVL72 contains 36 Vera CPUs and 54TB of LPDDR5X memory. That is 1.5TB per CPU. However, in June, TrendForce reported that NVIDIA would halve the capacity of its next-generation Vera SoCAMM modules while increasing total module shipments and Vera CPU production.&lt;/p&gt;
&lt;p&gt;On the surface, reduced memory capacity appears negative for component suppliers. When measuring PCB demand purely in terms of bits, it is. However, what Simmtech manufactures is not DRAM chips but PCBs for SoCAMM modules. Simmtech&amp;rsquo;s volume is more sensitive to the number of CPUs produced and the number of modules than to total memory capacity. Even if LPDDR shortage forces lower per-CPU capacity, if NVIDIA ships more Vera CPUs and modules, PCB demand may not decline.&lt;/p&gt;
&lt;p&gt;Looking at official materials, industry research, analyst estimates, and actual stock prices and supply-demand data shows that the business logic remains intact, but the price bottom has not yet been confirmed.&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Connected Context
This article is a follow-up to &lt;a class="link" href="https://koreainvestinsights.com/post/ai-pcb-thesis-system-bom-common-bottleneck-2026-05-05/" &gt;AI substrate and PCB investment thesis: System BOM as a common bottleneck&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/korea-ai-pcb-ecosystem-ten-companies-2026-05-05/" &gt;Korean AI PCB ecosystem: ten companies&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/vera-rubin-vr200-bom-memory-pcb-mlcc-korea-alpha-2026-05-21/" &gt;Vera Rubin VR200 bill of materials verification&lt;/a&gt;, and &lt;a class="link" href="https://koreainvestinsights.com/post/memory-supply-gap-spillover-substrate-essd-legacy-hana-three-reports-2026-07-14/" &gt;Memory supply gap spilling beyond HBM&lt;/a&gt;. Related articles can be found in the &lt;a class="link" href="https://koreainvestinsights.com/page/korea-ai-pcb-substrate-hub/" &gt;AI substrate and PCB hub&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/page/korea-semiconductor-hbm-kospi-hub/" &gt;AI HBM hub&lt;/a&gt;, and &lt;a class="link" href="https://koreainvestinsights.com/page/exclusive-analysis-hub/" &gt;exclusive analysis 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;NVIDIA&amp;rsquo;s official specifications show Vera Rubin NVL72 uses 36 Vera CPUs and 54TB of LPDDR5X. Standalone Vera CPU racks support up to 4 nodes per 1U tray, expanding to 256 CPUs per rack maximum. Vera&amp;rsquo;s evolution from a supporting CPU in GPU racks to an independent server platform is confirmed.&lt;/li&gt;
&lt;li&gt;TrendForce&amp;rsquo;s reported SoCAMM capacity reduction is a response to LPDDR supply shortage, not slowing demand. NVIDIA intends to lower per-module capacity while increasing total module shipments and Vera CPU production.&lt;/li&gt;
&lt;li&gt;Simmtech lists SoCAMM PCB on its official product page. However, simultaneous supply from three memory vendors and high market share are analyst estimates, not company disclosures. Actual customer-specific revenue and market share remain undisclosed.&lt;/li&gt;
&lt;li&gt;On July 16th, Simmtech closed at 107,400 KRW, down 11.68% in a single day. Over 20 trading days, institutions bought 236.2 billion KRW in net, and real money (pension, mutual funds, insurance) bought 186.4 billion KRW, while foreign investors sold 161.1 billion KRW and programs sold 185.9 billion KRW. Mid-term institutional buying pressure collides with short-term selling pressure.&lt;/li&gt;
&lt;li&gt;The 2026E PER of 27.1x is not cheap. Applying 2027E expected EPS of 6,655 KRW yields 16.1x, but only if SoCAMM revenue and profit margins expand as planned.&lt;/li&gt;
&lt;li&gt;The verdict is &lt;strong&gt;Conditional Buy&lt;/strong&gt;. A safer approach confirms support near 104,000 KRW with easing foreign and program selling, or recovery through 113,800 KRW on increased volume.&lt;/li&gt;
&lt;/ul&gt;
&lt;hr&gt;
&lt;div class="thesis-callout"&gt;
 &lt;div class="thesis-callout__label"&gt;This Article's Verdict&lt;/div&gt;
 &lt;div class="thesis-callout__body"&gt;
 Directly connecting SoCAMM2 capacity reduction to Simmtech PCB demand decline uses the wrong unit. Simmtech's revenue formula approaches Vera CPU units × modules per CPU × PCBs per module × Simmtech market share more than DRAM bit counts. However, the last two items in this formula have not yet been confirmed by company disclosures.
 &lt;/div&gt;
&lt;/div&gt;
&lt;hr&gt;
&lt;h2 id="1-numbers-to-discard-and-numbers-to-keep"&gt;1. Numbers to Discard and Numbers to Keep
&lt;/h2&gt;&lt;p&gt;The 2030 server CPU market size of 170 billion dollars, 1:1 CPU-to-GPU ratio, and 2027 DRAM shortage appearing in attached analysis are already reflected in major statements from multiple brokerages and corporate presentations. They are useful for understanding industry direction but are not sources of excess returns specific to Simmtech.&lt;/p&gt;
&lt;p&gt;Four numbers are more important when evaluating Simmtech:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;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;Vera CPUs shipped&lt;/td&gt;
 &lt;td&gt;Determines the installed base for the SoCAMM platform&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SoCAMM modules per CPU&lt;/td&gt;
 &lt;td&gt;Direct volume variable for Simmtech module PCBs&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Simmtech&amp;rsquo;s customer-specific supply share&lt;/td&gt;
 &lt;td&gt;Determines Simmtech&amp;rsquo;s portion of industry growth&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Module PCB unit price and cost&lt;/td&gt;
 &lt;td&gt;Determines the rate at which revenue growth converts to operating profit&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Using assumptions of 5 to 6 million Vera CPUs and 8 modules per CPU yields 40 to 48 million modules annually. At 5.5 million units, that is 44 million modules. However, this calculation is not an official NVIDIA shipment plan. It is a scenario derived from analyst and industry CPU shipment estimates and publicly disclosed system configurations.&lt;/p&gt;
&lt;p&gt;Therefore, the 44 million unit figure should not be used directly as a revenue projection. It is only an example for gauging order magnitude.&lt;/p&gt;
&lt;h2 id="2-official-nvidia-materials-confirm-veras-expansion-path"&gt;2. Official NVIDIA Materials Confirm Vera&amp;rsquo;s Expansion Path
&lt;/h2&gt;&lt;h3 id="21-nvl72-contains-36-vera-cpus"&gt;2.1 NVL72 Contains 36 Vera CPUs
&lt;/h3&gt;&lt;p&gt;NVIDIA&amp;rsquo;s &lt;a class="link" href="https://www.nvidia.com/en-us/data-center/vera-rubin-nvl72/" target="_blank" rel="noopener"
 &gt;official Vera Rubin NVL72 page&lt;/a&gt; presents the following configuration:&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 Specification&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Rubin GPUs&lt;/td&gt;
 &lt;td style="text-align: right"&gt;72&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Vera CPUs&lt;/td&gt;
 &lt;td style="text-align: right"&gt;36&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CPU Memory&lt;/td&gt;
 &lt;td style="text-align: right"&gt;LPDDR5X 54TB&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Memory per Vera CPU&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1.5TB&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Vera CPU Cores&lt;/td&gt;
 &lt;td style="text-align: right"&gt;88&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The structure is not simply one CPU per GPU. In NVL72, one Vera CPU supports two GPUs. Nevertheless, a single 72-GPU rack requires 36 CPUs. This shows that the CPU and LPDDR layers have risen to major components in AI system cost structure.&lt;/p&gt;
&lt;h3 id="22-vera-sells-beyond-gpu-racks"&gt;2.2 Vera Sells Beyond GPU Racks
&lt;/h3&gt;&lt;p&gt;The more significant change is the standalone Vera CPU system. &lt;a class="link" href="https://developer.nvidia.com/blog/?p=114004" target="_blank" rel="noopener"
 &gt;NVIDIA&amp;rsquo;s technical blog&lt;/a&gt; explains that Vera CPU racks can hold up to 4 nodes per 1U tray and scale to 256 Vera CPUs per rack. Single-socket and dual-socket systems support up to 1.5TB LPDDR5X per socket.&lt;/p&gt;
&lt;p&gt;NVIDIA announced that Cisco, Dell, HPE, Lenovo, and Supermicro plan to launch Vera-based systems in the second half of 2026. If the schedule holds, Vera demand sources extend beyond NVL72 assembly volumes.&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;Vera Demand Sources
&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;CPU layer in Vera Rubin NVL72
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;+ Standalone Vera CPU servers
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;+ CPU racks for inference and agents
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;+ Single and dual-socket systems from major OEMs
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;This is why Simmtech&amp;rsquo;s investment logic does not depend on GPU rack production alone. Expansion of standalone CPU systems broadens the demand base for SoCAMM modules and PCBs.&lt;/p&gt;
&lt;h2 id="3-what-does-socamm-capacity-reduction-mean"&gt;3. What Does SoCAMM Capacity Reduction Mean?
&lt;/h2&gt;&lt;p&gt;&lt;a class="link" href="https://www.trendforce.com/presscenter/news/20260610-13090.html" target="_blank" rel="noopener"
 &gt;TrendForce&amp;rsquo;s June 10, 2026 analysis&lt;/a&gt; found that only about 60% of NVIDIA&amp;rsquo;s 2027 LPDRAM requirements could be met with reserved allocations. To address the supply shortage, NVIDIA is reported to have chosen the following adjustments:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Lower the memory capacity of each SoCAMM module for Vera.&lt;/li&gt;
&lt;li&gt;Increase total module shipments.&lt;/li&gt;
&lt;li&gt;Produce more Vera CPUs with the same LPDRAM supply.&lt;/li&gt;
&lt;li&gt;Raise market penetration of standalone Vera CPU racks.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;In this structure, two types of demand can move in different directions.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Demand Unit&lt;/th&gt;
 &lt;th&gt;Impact of Capacity Reduction&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM bits and GB&lt;/td&gt;
 &lt;td&gt;Per-CPU installed capacity may decline&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SoCAMM modules&lt;/td&gt;
 &lt;td&gt;May increase if CPU shipment growth exceeds capacity reduction&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Module PCBs&lt;/td&gt;
 &lt;td&gt;Proportional to modules, so may increase&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Sockets, connectors, module testing&lt;/td&gt;
 &lt;td&gt;Increase with module count&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Expressing this in Simmtech terms gives the following formula:&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;Simmtech SoCAMM PCB Volume
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;= Vera CPU Shipment
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;× SoCAMM modules per CPU
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;× PCBs per module
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;× Simmtech supply share
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;Conversely, memory vendors&amp;rsquo; SoCAMM revenue approximates CPU count × per-CPU installed GB × price per GB. The same news can have different impacts on DRAM vendors and module PCB vendors.&lt;/p&gt;
&lt;h3 id="mismatch-between-official-specifications-and-trendforce-reporting"&gt;Mismatch Between Official Specifications and TrendForce Reporting
&lt;/h3&gt;&lt;p&gt;NVIDIA&amp;rsquo;s current official page shows a maximum 1.5TB LPDDR5X per Vera CPU. TrendForce reported next-generation module capacity reduction. With both documents existing simultaneously, it is not yet certain which case applies:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;NVIDIA&amp;rsquo;s page continues to display maximum configuration or existing design&lt;/li&gt;
&lt;li&gt;Capacity reduction applies only to specific customers or product configurations&lt;/li&gt;
&lt;li&gt;Final production specifications are not yet finalized&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;This inconsistency is not noise to be hidden but a verification item for investment judgment. The actual per-module capacity, modules per CPU, and customer-specific configuration of production systems must be verified through launched products and company disclosures.&lt;/p&gt;
&lt;h2 id="4-what-does-simmtech-supply-in-socamm2"&gt;4. What Does Simmtech Supply in SoCAMM2?
&lt;/h2&gt;&lt;h3 id="41-what-the-company-officially-confirms"&gt;4.1 What the Company Officially Confirms
&lt;/h3&gt;&lt;p&gt;&lt;a class="link" href="https://www.simmtech.com/product/board06.aspx" target="_blank" rel="noopener"
 &gt;Simmtech&amp;rsquo;s official product page&lt;/a&gt; classifies SoCAMM PCB as a next-generation memory module product for AI. The key specifications presented by the company are as follows:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Simmtech Official Description&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Base Memory&lt;/td&gt;
 &lt;td&gt;LPDDR&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Application&lt;/td&gt;
 &lt;td&gt;Next-generation memory modules for AI&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Layer Count&lt;/td&gt;
 &lt;td&gt;10–12 layers&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Key Requirements&lt;/td&gt;
 &lt;td&gt;Low dielectric constant and loss, low thermal expansion coefficient, impedance control&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Surface Treatment&lt;/td&gt;
 &lt;td&gt;Selective ENIG&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Confirmation that SoCAMM PCB appears on Simmtech&amp;rsquo;s product list is established. The company also manufactures general memory module PCBs and memory package substrates. When AI server LPDDR demand grows in both modules and packages, the company has a product foundation to respond.&lt;/p&gt;
&lt;h3 id="42-what-is-only-confirmed-by-analyst-estimates"&gt;4.2 What Is Only Confirmed by Analyst Estimates
&lt;/h3&gt;&lt;p&gt;The claim that Simmtech participates in SoCAMM2 production from Samsung Electronics, SK Hynix, and Micron simultaneously and holds high market share is not a company disclosure. It is an analysis and market estimate in &lt;a class="link" href="https://www.simmtech.com/upload/media/file/639179687810021917.pdf" target="_blank" rel="noopener"
 &gt;the brokerage report posted on Simmtech&amp;rsquo;s homepage&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;Therefore, these expressions must be distinguished:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Expression&lt;/th&gt;
 &lt;th&gt;Evidence Level&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Simmtech possesses SoCAMM PCB products&lt;/td&gt;
 &lt;td&gt;Company official confirmation&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Simmtech participates in SoCAMM2 production from three memory vendors&lt;/td&gt;
 &lt;td&gt;Analyst estimate&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Simmtech holds #1 supply share&lt;/td&gt;
 &lt;td&gt;Analyst estimate&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Customer-specific revenue and actual market share&lt;/td&gt;
 &lt;td&gt;Undisclosed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Exact timing of SoCAMM contribution to Simmtech profit&lt;/td&gt;
 &lt;td&gt;Undisclosed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This distinction is necessary because customer qualification and production volume drive the economics of the substrate business. The ability to manufacture a product and maintaining high market share for repeat revenue are separated by stages of customer qualification, yield, unit price, and utilization rate.&lt;/p&gt;
&lt;h3 id="43-sk-hynixs-mass-production-confirms-platform-reality"&gt;4.3 SK Hynix&amp;rsquo;s Mass Production Confirms Platform Reality
&lt;/h3&gt;&lt;p&gt;According to &lt;a class="link" href="https://news.skhynix.com/mass-production-socamm2-192gb/" target="_blank" rel="noopener"
 &gt;SK Hynix&amp;rsquo;s official announcement&lt;/a&gt;, the company began mass production of 192GB SoCAMM2 using 1cnm LPDDR5X in April 2026. The product targets the NVIDIA Vera Rubin platform.&lt;/p&gt;
&lt;p&gt;This shows that SoCAMM2 is not a long-term research project but has entered actual mass production. However, SK Hynix&amp;rsquo;s module mass production does not automatically prove Simmtech&amp;rsquo;s customer-specific supply volumes. Simmtech demonstrating SoCAMM revenue, high-layer board mix, and utilization rate improvements in second and third quarter results is separately necessary.&lt;/p&gt;
&lt;h2 id="5-results-are-recovering-but-still-in-the-stage-of-proving-expectations"&gt;5. Results Are Recovering but Still in the Stage of Proving Expectations
&lt;/h2&gt;&lt;p&gt;Simmtech recorded 1.41 trillion KRW in full-year 2025 revenue and 11.9 billion KRW in operating profit, with a net loss of 164.2 billion KRW. In the first quarter 2026, consolidated revenue recovered to approximately 422.4 billion KRW and operating profit to approximately 13.7 billion KRW. Second quarter consensus expects faster recovery pace.&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 Actual&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2Q26 Consensus&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Sequential 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;~422.4 billion KRW&lt;/td&gt;
 &lt;td style="text-align: right"&gt;466.9 billion KRW&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~+10.5%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Operating Profit&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~13.7 billion KRW&lt;/td&gt;
 &lt;td style="text-align: right"&gt;47.3 billion KRW&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~+245%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Operating Margin&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~3.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~10.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+6.9 percentage points&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;First quarter actuals are from &lt;a class="link" href="https://kind.krx.co.kr/external/2026/05/14/000890/20260514001959/11013.htm" target="_blank" rel="noopener"
 &gt;Simmtech&amp;rsquo;s quarterly report&lt;/a&gt;; second quarter estimates are from Naver and WiseReport consensus collected by Research OS local database on July 16th.&lt;/p&gt;
&lt;p&gt;Operating profit growth far exceeds revenue growth. This indicates the need for simultaneous increases in high-margin product mix, factory utilization, and fixed cost absorption. Even if SoCAMM volume increases, profit margin may fall below expectations due to raw material and gold plating costs, initial yield, and customer-specific price negotiations.&lt;/p&gt;
&lt;h3 id="the-profit-path-demanded-by-annual-consensus"&gt;The Profit Path Demanded by Annual Consensus
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Year&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;Net Income&lt;/th&gt;
 &lt;th style="text-align: right"&gt;EPS&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2026E PER (July 16 close)&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;2026E&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1.932 trillion KRW&lt;/td&gt;
 &lt;td style="text-align: right"&gt;189.6 billion KRW&lt;/td&gt;
 &lt;td style="text-align: right"&gt;149.3 billion KRW&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3,968 KRW&lt;/td&gt;
 &lt;td style="text-align: right"&gt;27.1x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2027E&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2.289 trillion KRW&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Consensus not aggregated&lt;/td&gt;
 &lt;td style="text-align: right"&gt;251.0 billion KRW&lt;/td&gt;
 &lt;td style="text-align: right"&gt;6,655 KRW&lt;/td&gt;
 &lt;td style="text-align: right"&gt;16.1x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2028E&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2.566 trillion KRW&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Consensus not aggregated&lt;/td&gt;
 &lt;td style="text-align: right"&gt;322.8 billion KRW&lt;/td&gt;
 &lt;td style="text-align: right"&gt;8,588 KRW&lt;/td&gt;
 &lt;td style="text-align: right"&gt;12.5x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;For the current stock price to look cheap, 2027 and 2028 earnings must actually materialize. The current 2026E PER of 27.1x and PBR of 5.64x substantially reflects early recovery. An average price target of 171,250 KRW is 59.5% higher than current price, but when targets are rising rapidly, they are difficult to use as a safety margin.&lt;/p&gt;
&lt;h2 id="6-july-16th-stock-price-overheating-reduced-but-trend-recovery-not-yet-confirmed"&gt;6. July 16th Stock Price: Overheating Reduced but Trend Recovery Not Yet Confirmed
&lt;/h2&gt;&lt;p&gt;Simmtech closed at 107,400 KRW on July 16th, declining 11.68% that day with an intraday low of 103,700 KRW. This is 34.6% lower than the July 1st high of 164,200 KRW.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Period and Indicator&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;1-day return&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-11.68%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;5-day&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-8.91%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;10-day&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-13.87%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;20-day&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-15.76%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;60-day&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+40.94%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;120-day&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+109.77%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;vs. 5-day MA&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-5.42%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;vs. 20-day MA&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-12.28%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;vs. 60-day MA&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-4.48%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;vs. 120-day MA&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+27.39%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;RSI14&lt;/td&gt;
 &lt;td style="text-align: right"&gt;44.4&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;RSI has exited overbought territory. However, price remains below 5-day, 20-day, and 60-day moving averages. Since it is 27% above the 120-day line, all upside from the long-term move has not disappeared. Both &amp;ldquo;a sharp decline means cheap&amp;rdquo; and &amp;ldquo;the trend has recovered&amp;rdquo; cannot yet be confirmed.&lt;/p&gt;
&lt;p&gt;On a trading basis, July 14th low of 99,600 KRW and July 16th low of 103,700 KRW represent the first support zone. The 113,800 KRW level confirms whether broken short-term support recovers; 121,600 to 124,000 KRW is a resistance zone where July 15th close and the 20-day moving average overlap.&lt;/p&gt;
&lt;h2 id="7-supply-and-demand-institutions-bought-but-foreign-investors-and-programs-still-sell"&gt;7. Supply and Demand: Institutions Bought But Foreign Investors and Programs Still Sell
&lt;/h2&gt;&lt;p&gt;Cumulative net buying amounts through July 16th from Research OS local Kiwoom database are as follows. Units are in billions of KRW.&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;Foreign&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;th style="text-align: right"&gt;Real Money&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Programs&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;1 day&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-9.83&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-5.88&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+15.57&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-5.91&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-11.55&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;5 days&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-25.79&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+6.02&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+18.95&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+7.98&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-28.71&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;10 days&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-93.39&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+80.32&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+11.79&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+75.77&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-40.78&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;20 days&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-161.09&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+236.21&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-84.25&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+186.44&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-185.94&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Real money is the sum of pension funds, mutual funds, and insurance. The table reveals two things simultaneously.&lt;/p&gt;
&lt;p&gt;First, on a 20-day basis, institutional and real money buying is substantial. During the post-peak correction, domestically oriented long-term capital absorbed volume.&lt;/p&gt;
&lt;p&gt;Second, foreign investor and program selling has continued at similar scale. On the July 16th plunge, foreign investors and institutions sold together while individuals absorbed 15.57 billion KRW. It is difficult to judge the bottom confirmed by institutional buying alone.&lt;/p&gt;
&lt;p&gt;Short-selling transaction share also rose to 21.2% on July 16th. The 20-day average is 9.3%. Short-rebate inventory is approximately 5.48 million shares or 14.6% of issued shares, but not all short-rebate inventory is short-sale inventory. This figure should only be consulted as a reference for available selling volume.&lt;/p&gt;
&lt;h2 id="8-simmtech-revisited-through-p--q--c"&gt;8. Simmtech Revisited Through P × Q × C
&lt;/h2&gt;&lt;h3 id="p-unit-price-of-high-layer-socamm-pcb"&gt;P: Unit Price of High-Layer SoCAMM PCB
&lt;/h3&gt;&lt;p&gt;SoCAMM PCB requires 10–12 layers, low dielectric loss, low thermal expansion, and precise impedance control. Higher manufacturing difficulty compared to general memory module PCB can help improve product mix. However, customer-specific unit prices and price increase structures are undisclosed.&lt;/p&gt;
&lt;h3 id="q-vera-cpu-and-socamm-module-count"&gt;Q: Vera CPU and SoCAMM Module Count
&lt;/h3&gt;&lt;p&gt;Q is the largest upside variable. Beyond NVL72 mass production, standalone Vera CPU racks and major OEM servers are scheduled for the second half of 2026. If NVIDIA follows TrendForce&amp;rsquo;s report in lowering per-module capacity while increasing module shipments, Simmtech&amp;rsquo;s Q can grow faster than DRAM bit count.&lt;/p&gt;
&lt;h3 id="c-raw-materials-gold-plating-yield-utilization"&gt;C: Raw Materials, Gold Plating, Yield, Utilization
&lt;/h3&gt;&lt;p&gt;The item determining whether revenue growth converts to operating profit. PCBs are sensitive to gold and copper prices, BT materials, plating costs, and initial yield. To achieve 10.1% operating margin consensus for Q2, increased high-margin products and cost pass-through must be confirmed together.&lt;/p&gt;
&lt;h2 id="9-three-scenarios"&gt;9. Three Scenarios
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Scenario&lt;/th&gt;
 &lt;th&gt;Conditions&lt;/th&gt;
 &lt;th&gt;Earnings and Price Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Bullish&lt;/td&gt;
 &lt;td&gt;Vera Rubin and standalone CPU servers ship as scheduled; total SoCAMM module count increases; Simmtech maintains three-memory-vendor market share&lt;/td&gt;
 &lt;td&gt;2027E EPS of 6,655 KRW or higher becomes achievable and justifies leading PER of ~16x&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Base&lt;/td&gt;
 &lt;td&gt;Vera shipments grow but LPDDR shortage and customer-specific ramps misalign; Simmtech share and margin remain at consensus levels&lt;/td&gt;
 &lt;td&gt;Business growth remains valid but stock can fluctuate widely based on earnings and supply-demand&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Bearish&lt;/td&gt;
 &lt;td&gt;Vera or SoCAMM mass production delays; reduced per-module capacity does not lead to increased module count; Simmtech share or yield underperforms&lt;/td&gt;
 &lt;td&gt;27x 2026E PER becomes a burden and probability of revisiting July lows increases&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;In these scenarios, the most important variable is not the 170 billion dollar server CPU market. It is Simmtech&amp;rsquo;s actual shipment volume, customer-specific revenue, high-layer board mix, and operating margin. Even if large industry numbers prove correct, if the portion accruing to the company is small, the stock will struggle.&lt;/p&gt;
&lt;h2 id="10-investment-verdict-and-verification-sequence"&gt;10. Investment Verdict and Verification Sequence
&lt;/h2&gt;&lt;h3 id="verdict-conditional-buy"&gt;Verdict: Conditional Buy
&lt;/h3&gt;&lt;p&gt;Business logic has become clearer. NVIDIA official materials confirmed Vera&amp;rsquo;s standalone server expansion, and TrendForce&amp;rsquo;s report shows capacity reduction as a supply shortage response, not demand decline. Simmtech officially possesses SoCAMM PCB products.&lt;/p&gt;
&lt;p&gt;However, the stock remains under pressure from foreign investor and program selling, and it is not cheap on 2026 earnings basis. Therefore, it is better to consider two entry paths.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Path&lt;/th&gt;
 &lt;th&gt;Verification Condition&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Support Confirmation&lt;/td&gt;
 &lt;td&gt;Hold the low between 103,700–106,000 KRW and confirm easing of foreign investor and program selling&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Trend Confirmation&lt;/td&gt;
 &lt;td&gt;Recover through 113,800 KRW on increased volume with either institutions or foreign investors turning net buyers&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Recovery through 121,600–124,000 KRW would substantially reverse near-term trend damage. Conversely, breaking below 99,600 KRW on a closing basis without recovery would require reassessing price hypothesis.&lt;/p&gt;
&lt;h3 id="catalysts"&gt;Catalysts
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Achievement of 466.9 billion KRW revenue and 47.3 billion KRW operating profit in 2Q26 earnings&lt;/li&gt;
&lt;li&gt;Company commentary on SoCAMM revenue, high-layer board mix, and utilization rate&lt;/li&gt;
&lt;li&gt;Actual market launch of Vera-based OEM servers in second half 2026&lt;/li&gt;
&lt;li&gt;NVIDIA&amp;rsquo;s final production specifications and expansion of standalone Vera CPU rack shipments&lt;/li&gt;
&lt;li&gt;Mass production and customer qualification updates from SK Hynix, Samsung Electronics, and Micron for SoCAMM2&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="invalidation-conditions"&gt;Invalidation Conditions
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Repeat orders fail to follow initial SoCAMM volume&lt;/li&gt;
&lt;li&gt;Reduced per-module capacity does not translate to increased total module shipments&lt;/li&gt;
&lt;li&gt;Simmtech&amp;rsquo;s customer-specific market share confirms below analyst estimates&lt;/li&gt;
&lt;li&gt;Q2 operating profit materially misses consensus with recurring raw material and yield burdens&lt;/li&gt;
&lt;li&gt;Vera Rubin or standalone Vera CPU system launch delays significantly&lt;/li&gt;
&lt;li&gt;99,600 KRW breaks below support with concurrent expansion of foreign investor and program selling&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="11-final-conclusion"&gt;11. Final Conclusion
&lt;/h2&gt;&lt;p&gt;Vera CPU supply expansion reinforces Simmtech&amp;rsquo;s SoCAMM2 investment thesis. Particularly, the path of lowering module capacity due to LPDDR shortage while increasing total module and CPU shipments is difficult to characterize as unfavorable for substrate vendors. DRAM vendors sell bits and GB; Simmtech sells PCBs that go into each module. These are different units.&lt;/p&gt;
&lt;p&gt;However, many items remain unconfirmed. Vera shipment volumes, final modules per CPU, Simmtech&amp;rsquo;s actual customer-specific market share, SoCAMM revenue, and high-layer board profit margin cannot be finalized by public materials alone. A calculation like 44 million modules is a useful scenario but not official guidance.&lt;/p&gt;
&lt;p&gt;Current stock price has corrected nearly 35% from peak, but there is no evidence that foreign investor and program selling has ended. Therefore, this idea should be viewed as &amp;ldquo;a conditional buy candidate requiring confirmation of both earnings and supply-demand&amp;rdquo; rather than &amp;ldquo;a growth stock that has fallen sharply.&amp;rdquo; Q2 operating profit and SoCAMM revenue commentary is the first verification, and second-half 2026 Vera standalone server shipments represent the second verification.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="evidence-and-limitations"&gt;Evidence and Limitations
&lt;/h2&gt;&lt;h3 id="confirmed-facts"&gt;Confirmed Facts
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;NVIDIA official specifications present 36 Vera CPUs and 54TB LPDDR5X in NVL72.&lt;/li&gt;
&lt;li&gt;NVIDIA disclosed standalone CPU systems accommodating up to 256 Vera CPUs per rack and OEM launch plans for second half 2026.&lt;/li&gt;
&lt;li&gt;TrendForce reported adjustments to lower per-module capacity while increasing total module shipments and Vera CPU production due to LPDRAM shortage.&lt;/li&gt;
&lt;li&gt;SK Hynix announced mass production of 192GB SoCAMM2 for Vera Rubin.&lt;/li&gt;
&lt;li&gt;Simmtech lists SoCAMM PCB on its official product list.&lt;/li&gt;
&lt;li&gt;July 16th Simmtech close, supply-demand, and consensus have been reverified from local database.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="estimates"&gt;Estimates
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Annual calculation of 40 to 48 million modules using assumptions of 5 to 6 million Vera CPUs and 8 modules per CPU&lt;/li&gt;
&lt;li&gt;Claims that Simmtech supplies simultaneously to three memory vendors and maintains high market share&lt;/li&gt;
&lt;li&gt;Magnitude of capacity reduction impact on Simmtech revenue and profit following module count increase&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="items-not-confirmed-by-public-materials"&gt;Items Not Confirmed by Public Materials
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;NVIDIA&amp;rsquo;s official 2027 Vera CPU shipment guidance&lt;/li&gt;
&lt;li&gt;Final production system modules per CPU and capacity per module&lt;/li&gt;
&lt;li&gt;Simmtech&amp;rsquo;s customer-specific SoCAMM revenue, market share, unit price, and profit margin&lt;/li&gt;
&lt;li&gt;Q2 earnings announcement date and company SoCAMM revenue guidance&lt;/li&gt;
&lt;li&gt;Precise scope of application for TrendForce&amp;rsquo;s reported capacity reduction and NVIDIA&amp;rsquo;s official 1.5TB specification&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Informational analysis based on public materials and local market data; not investment advice reflecting individual investor timelines, prices, or risk tolerance.&lt;/p&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>Weekly Semiconductor Deep Dive: Strong Earnings, Compressed Multiples, and the 2028 Inflection</title><link>https://koreainvestinsights.com/post/weekly-semiconductor-deep-dive-strong-earnings-lower-multiple-2028-pivot-2026-07-19/</link><pubDate>Sun, 19 Jul 2026 15:53:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/weekly-semiconductor-deep-dive-strong-earnings-lower-multiple-2028-pivot-2026-07-19/</guid><description>&lt;p&gt;On July 13, KOSPI fell 9.05% and SK Hynix dropped 15.37%. Yet in the same week, TSMC reported Q2 revenue of $40.2 billion and a gross margin of 67.7%, while ASML raised its annual revenue outlook to €43–45 billion. Commodity DRAM and NAND contract prices also rose sharply. It was a week in which physical orders and stock prices moved in opposite directions.&lt;/p&gt;
&lt;p&gt;This divergence is less a signal that the semiconductor boom is over, and more a sign that the market&amp;rsquo;s question has changed. Investors are no longer looking only at how much will be earned in 2026 and 2027. They are first calculating how much supply will expand in 2028, whether customers can sustain high memory prices and data center capex, and how much inference efficiency gains will erode the scarcity premium of HBM.&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Connected Context
This post consolidates &lt;a class="link" href="https://koreainvestinsights.com/post/semiconductor-bull-bear-four-clocks-capital-intensity-cycle-2026-07-17/" &gt;Five Clocks of the Semiconductor Bull and Bear Case&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/memory-fair-value-fcfe-terminal-samsung-hynix-micron-2026-07-17/" &gt;Memory Fair Value via FCFE and Normalized Earnings&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/samsung-sk-hynix-2027-2028-integrated-scenarios-risk-adjusted-valuation-2026-07-13/" &gt;Samsung Electronics and SK Hynix: 2027–2028 Scenarios&lt;/a&gt;, and &lt;a class="link" href="https://koreainvestinsights.com/post/hbm-2030-supply-demand-267eb-demand-model-crosscheck-2026-07-13/" &gt;HBM 2030 Supply-Demand Model Cross-Check (26.7 EB)&lt;/a&gt; into a weekly synthesis updated with new data from the third week of July. Related posts are available at the &lt;a class="link" href="https://koreainvestinsights.com/page/korea-semiconductor-hbm-kospi-hub/" &gt;AI HBM Hub&lt;/a&gt; and the &lt;a class="link" href="https://koreainvestinsights.com/page/exclusive-analysis-hub/" &gt;Exclusive Analysis 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;Official data confirmed from July 13–19 does not support the claim that AI semiconductor demand has peaked. TSMC and ASML results and guidance, DRAM and NAND prices, and server memory demand all reinforce the earnings-strength thesis for 2026–2027.&lt;/li&gt;
&lt;li&gt;The same data also amplifies 2028 supply risk. ASML&amp;rsquo;s capacity expansion plans and TSMC&amp;rsquo;s heavy capex are evidence of current orders — and simultaneously a reservation of future supply and depreciation.&lt;/li&gt;
&lt;li&gt;HBM long-term contracts can raise the floor on volume and earnings, but may slow the pace at which spot price spikes translate into blended ASPs. Volume stability and a cap on price upside coexist.&lt;/li&gt;
&lt;li&gt;China plays three roles simultaneously: an end-demand destination, a competitor in commodity memory, and a market that could be politically decoupled. Its larger impact is on 2028 commodity DRAM and NAND price ceilings and global supply discipline, not on near-term HBM substitution.&lt;/li&gt;
&lt;li&gt;Efficiency gains from open models such as Kimi K3 do not simply eliminate semiconductor demand. They can shift the value of some HBM toward server DRAM, eSSD, networking, and power. The key question is not whether cost-per-task falls, but how much total workload expands and what share is deployed on self-built infrastructure.&lt;/li&gt;
&lt;li&gt;Current base-case scenario probabilities: Extended Scarcity 25%, Strong Earnings / Compressed Multiple 40%, Efficiency and Supply Normalization 25%, Financial / Demand Dislocation 10%. The most probable path is one where strong earnings in 2026–2027 and duration discounting for 2028 coexist.&lt;/li&gt;
&lt;/ul&gt;
&lt;hr&gt;
&lt;div class="thesis-callout"&gt;
 &lt;div class="thesis-callout__label"&gt;Key Statement of the Week&lt;/div&gt;
 &lt;div class="thesis-callout__body"&gt;
 The physical evidence underpinning the memory supercycle passed its first full-scale stress test. Stock prices, however, have begun asking not about the magnitude of scarcity but about how long excess earnings will last. 2026–2027 is the period for earnings delivery; 2028 is the period that simultaneously tests supply, efficiency, and customer profitability.
 &lt;/div&gt;
&lt;/div&gt;
&lt;hr&gt;
&lt;h2 id="1-scope-and-evidence-rules"&gt;1. Scope and Evidence Rules
&lt;/h2&gt;&lt;p&gt;The analysis covers July 13–19, 2026. Company IR materials, government and congressional filings, industry data, broker estimates, and market supply-demand figures drawn from two weekly analysis documents were re-categorized. Where multiple posts cited the same underlying source, the confirmation count was not incremented.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Grade&lt;/th&gt;
 &lt;th&gt;Source&lt;/th&gt;
 &lt;th&gt;How Used in This Report&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;A&lt;/td&gt;
 &lt;td&gt;Company IR, government / congressional / regulatory filings, SEC disclosures&lt;/td&gt;
 &lt;td&gt;Baseline for facts and official plans&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;B&lt;/td&gt;
 &lt;td&gt;Industry research such as TrendForce, primary broker reports&lt;/td&gt;
 &lt;td&gt;Supporting evidence for pricing, share, and company-level estimates&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;C&lt;/td&gt;
 &lt;td&gt;Cross-referenced press reports, channel check summaries&lt;/td&gt;
 &lt;td&gt;Used for directional confirmation only; not used as standalone investment signals&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;D&lt;/td&gt;
 &lt;td&gt;Social media, circulating summary images&lt;/td&gt;
 &lt;td&gt;Used only to identify questions requiring further verification&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Orders from TSMC, ASML, and memory manufacturers can reflect the same hyperscaler capex plan multiple times. Official plans signal demand intent but do not guarantee actual utilization or cash flow. The sharp stock price declines were not attributed solely to technical supply-demand dynamics. Flow imbalances can amplify drawdowns, but they cannot eliminate a fundamental re-rating of earnings duration.&lt;/p&gt;
&lt;h2 id="2-a-map-of-the-weeks-events"&gt;2. A Map of the Week&amp;rsquo;s Events
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Date&lt;/th&gt;
 &lt;th&gt;Confirmed Developments&lt;/th&gt;
 &lt;th&gt;Questions the Market Began Asking&lt;/th&gt;
 &lt;th&gt;Current Assessment&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;July 13&lt;/td&gt;
 &lt;td&gt;HBM 2030 demand model at 26.7 EB re-verified; SK Hynix Q2 earnings estimate cut; KOSPI sharp decline&lt;/td&gt;
 &lt;td&gt;Long-term shortage, or excessive extrapolation?&lt;/td&gt;
 &lt;td&gt;2027 shortage has strong support; 2030 shortage magnitude carries low conviction&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;July 14&lt;/td&gt;
 &lt;td&gt;SK Hynix estimates adjusted to reflect HBM long-term contracts; IBM and Ericsson flagging memory cost burden; reports of Apple testing CXMT chips&lt;/td&gt;
 &lt;td&gt;Are long-term contracts an earnings floor or a price ceiling?&lt;/td&gt;
 &lt;td&gt;Volume visibility is high, but ASP upside constraints and customer cost burden emerge alongside&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;July 15&lt;/td&gt;
 &lt;td&gt;ASML Q2 results and 2027–2028 capacity expansion plans&lt;/td&gt;
 &lt;td&gt;Strong orders, or the start of a supply response?&lt;/td&gt;
 &lt;td&gt;Both are correct — simultaneously reinforces current demand and amplifies 2028 supply risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;July 16&lt;/td&gt;
 &lt;td&gt;TSMC Q2 results and Q3 guidance; U.S. House letter requesting restrictions on Chinese memory procurement&lt;/td&gt;
 &lt;td&gt;AI demand is strong — why is memory stock price weak?&lt;/td&gt;
 &lt;td&gt;Debate centers on earnings attribution, supply, and cost of capital — not demand collapse&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;July 17&lt;/td&gt;
 &lt;td&gt;Kimi K3 launch; Chinese open-model efficiency gains and API pricing repricing&lt;/td&gt;
 &lt;td&gt;Does efficiency reduce semiconductor demand?&lt;/td&gt;
 &lt;td&gt;Cost-per-task falls, but aggregate usage and self-built deployments may expand&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;July 18&lt;/td&gt;
 &lt;td&gt;China AI API cost analysis; review of memory trio&amp;rsquo;s China exposure&lt;/td&gt;
 &lt;td&gt;Is China a customer or a competitor?&lt;/td&gt;
 &lt;td&gt;Both roles simultaneously; impacts 2028 normalized value more than 2026 EPS&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;July 19&lt;/td&gt;
 &lt;td&gt;Channel checks circulating claiming strong server DRAM and HBM4 orders&lt;/td&gt;
 &lt;td&gt;Did the sell-off price in a Q3 earnings collapse?&lt;/td&gt;
 &lt;td&gt;Probability appears lower, but primary source unconfirmed — maintained at Grade C&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The most significant shift in this table is not the direction of individual events, but the time horizon. Until early July, the debate centered on whether H2 2026 earnings would beat expectations. It has since shifted to whether 2028 earnings represent a normalized earnings level, and how long elevated prices and margins can be sustained.&lt;/p&gt;
&lt;h2 id="3-evidence-strengthened-this-week"&gt;3. Evidence Strengthened This Week
&lt;/h2&gt;&lt;h3 id="31-tsmc-and-asml-showed-that-ai-orders-have-translated-into-real-activity"&gt;3.1 TSMC and ASML Showed That AI Orders Have Translated into Real Activity
&lt;/h3&gt;&lt;p&gt;TSMC&amp;rsquo;s Q2 results demonstrate that AI demand has moved beyond the planning stage.&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;TSMC Q2 2026 and Q3 2026 Outlook&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Q2 revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$40.2 billion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q2 gross margin&lt;/td&gt;
 &lt;td style="text-align: right"&gt;67.7%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q2 operating margin&lt;/td&gt;
 &lt;td style="text-align: right"&gt;60.3%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q3 revenue guidance&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$44.6–$45.8 billion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q3 gross margin guidance&lt;/td&gt;
 &lt;td style="text-align: right"&gt;65–67%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The high gross margins and next-quarter growth guidance indicate that the quality of orders for leading-edge process nodes and advanced packaging remains intact. &lt;a class="link" href="https://investor.tsmc.com/english/quarterly-results/2026/q2" target="_blank" rel="noopener"
 &gt;TSMC Q2 2026 Official Results&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;ASML confirmed the same direction. Q2 revenue of €9.326 billion and net income of €2.918 billion were reported, and the 2026 revenue outlook was set at €43–45 billion. The capacity roadmap reveals the timeline at which current orders translate into future supply.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Equipment&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2026 Capacity&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2027 Plan&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2028 Under Review&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Low-NA EUV&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~65 units&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~30% expansion&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Additional ~30% expansion under review&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;DUV immersion&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~130 units&lt;/td&gt;
 &lt;td style="text-align: right"&gt;~30% expansion&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Additional ~30% expansion under review&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;ASML stated that AI investment is driving demand for leading-edge logic and memory, and that customer capacity commitments are converting into long-term demand visibility. These figures weaken the strong bear case that AI orders consist entirely of duplicate bookings. &lt;a class="link" href="https://www.asml.com/en/news/press-releases/2026/q2-2026-financial-results" target="_blank" rel="noopener"
 &gt;ASML Q2 2026 Official Results&lt;/a&gt;&lt;/p&gt;
&lt;h3 id="32-memory-prices-remain-elevated"&gt;3.2 Memory Prices Remain Elevated
&lt;/h3&gt;&lt;p&gt;TrendForce projected Q2 2026 commodity DRAM contract prices to rise 58–63% quarter-over-quarter, and NAND prices to rise 70–75%. The Q3 server DRAM price growth forecast was revised to 13–18%, but the upward trajectory is maintained. &lt;a class="link" href="https://www.trendforce.com/presscenter/news/20260331-12995.html" target="_blank" rel="noopener"
 &gt;Q2 DRAM/NAND Outlook&lt;/a&gt;, &lt;a class="link" href="https://www.trendforce.com/presscenter/news/20260709-13140.html" target="_blank" rel="noopener"
 &gt;Q3 Server DRAM Outlook&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;It is important to distinguish between a deceleration in the rate of price increases and an actual price decline. Given how rapidly prices rose in Q2, the quarter-over-quarter growth rate in Q3 may slow. Nonetheless, if absolute prices remain elevated, suppliers&amp;rsquo; operating income can remain strong. The risk the market is pricing is not a decline in earnings, but a deceleration in earnings growth.&lt;/p&gt;
&lt;h3 id="33-hbm-ramp-cannibalizes-commodity-dram-supply"&gt;3.3 HBM Ramp Cannibalizes Commodity DRAM Supply
&lt;/h3&gt;&lt;p&gt;HBM requires more wafer area, stacking, packaging, and testing capacity than commodity DRAM to produce the same number of bits. As the three memory manufacturers increase their HBM mix, HBM shipments expand while the effective bit supply of commodity DRAM can shrink.&lt;/p&gt;
&lt;p&gt;As a result, the bottleneck broadens in the following sequence:&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 / ASIC
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;-&amp;gt; HBM
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;-&amp;gt; Server DRAM
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;-&amp;gt; eSSD and NAND
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;-&amp;gt; Packaging / Substrate / Test
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;-&amp;gt; Networking / Power / Cooling
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;AI agents and long-context inference do not rely solely on HBM for model weights. They also consume server DRAM and eSSD for KV cache and data serving, and consume more networking to connect multiple accelerators and sparse mixture-of-expert models. This is why aggregate memory tier volumes can continue to expand even if HBM unit pricing comes under pressure.&lt;/p&gt;
&lt;h3 id="34-2027-shortage-is-a-strong-conclusion-25-shortage-by-2030-is-a-scenario"&gt;3.4 2027 Shortage Is a Strong Conclusion; 2.5× Shortage by 2030 Is a Scenario
&lt;/h3&gt;&lt;p&gt;Comparing 2030 HBM demand of 26.7 EB against supply of 10.6 EB yields demand at 2.52× supply. The arithmetic is reproducible. The issue lies not in the result but in whether the underlying assumptions hold simultaneously. The base model assumes 24× token growth, 5× model size expansion, 4× context length extension, and a 70% HBM retention rate for KV cache — all simultaneously.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Assumption Change&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2030 HBM Demand&lt;/th&gt;
 &lt;th style="text-align: right"&gt;vs. 10.6 EB Supply&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Base model&lt;/td&gt;
 &lt;td style="text-align: right"&gt;26.7 EB&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2.52×&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Model size 2×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;18.5 EB&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1.75×&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KV efficiency 6×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;22.3 EB&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2.10×&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM retention rate 50%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;22.9 EB&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2.16×&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Tokens 12×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;16.2 EB&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1.53×&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Composite bear assumptions&lt;/td&gt;
 &lt;td style="text-align: right"&gt;6.5 EB&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0.62×&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Reducing any single variable still leaves a shortage, but if multiple bearish assumptions hold simultaneously, oversupply becomes possible. The investable statement is therefore: &amp;ldquo;The market is likely to remain tight through 2027.&amp;rdquo; The statement &amp;ldquo;Supply will be definitively 2.5× short by 2030&amp;rdquo; cannot be directly plugged into a price target.&lt;/p&gt;
&lt;h2 id="4-risks-strengthened-this-week"&gt;4. Risks Strengthened This Week
&lt;/h2&gt;&lt;h3 id="41-earnings-growth-rate-has-become-more-important-than-strong-absolute-earnings"&gt;4.1 Earnings Growth Rate Has Become More Important Than Strong Absolute Earnings
&lt;/h3&gt;&lt;p&gt;The SK Hynix Q2 operating income estimate cut originated from adjustments to ASP assumptions rather than from a collapse in shipment volumes.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Report&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Q2 2026 Operating Income Estimate&lt;/th&gt;
 &lt;th&gt;Key Adjustment&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Mirae Asset Securities&lt;/td&gt;
 &lt;td style="text-align: right"&gt;From ₩70.7 trillion to ₩62.3 trillion&lt;/td&gt;
 &lt;td&gt;DRAM ASP from +40.6% to +32.9%; NAND from +55% to +50%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Korea Investment &amp;amp; Securities&lt;/td&gt;
 &lt;td style="text-align: right"&gt;₩60.4 trillion&lt;/td&gt;
 &lt;td&gt;DRAM ASP +28.9%; NAND +50.9%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;An operating income in the ₩60 trillion range is very large in absolute terms. However, stock prices responded more to the magnitude of the revision from prior estimates than to the absolute level of earnings. The 15.37% decline in SK Hynix on July 13 is better explained by the combination of shifting expectations and concentrated technical supply-demand dynamics than by any conclusion that the upcycle has ended.&lt;/p&gt;
&lt;p&gt;This is the central logic of the weekly analysis. Strong demand alone does not drive stock prices higher. Demand must beat market expectations, and evidence that elevated earnings persist into 2028 is required.&lt;/p&gt;
&lt;h3 id="42-long-term-contracts-create-both-an-earnings-floor-and-a-price-ceiling"&gt;4.2 Long-Term Contracts Create Both an Earnings Floor and a Price Ceiling
&lt;/h3&gt;&lt;p&gt;HBM long-term supply agreements provide volume visibility, customer lock-in, predictability of payback periods, and downside protection when spot prices fall. On the other side, the following concerns apply:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Even if spot prices spike, a company&amp;rsquo;s blended ASP may rise only gradually.&lt;/li&gt;
&lt;li&gt;Pricing formulas, renegotiation mechanisms, minimum purchase commitments, and cancellation terms are not disclosed.&lt;/li&gt;
&lt;li&gt;Customer compliance with minimum purchase obligations and credit risk remain with the supplier.&lt;/li&gt;
&lt;li&gt;Up-front payments and long-term volume commitments may not flow through to accounting earnings and operating cash flow at the same pace.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Interpreting long-term contracts as an unconditional reinforcement of pricing power is only half the picture. Contracts raise the earnings floor and reduce volatility, but they may not capture the full upside of a spot price spike. The SK Hynix Q2 estimate revision was the first instance of this gap appearing in reported numbers. The detailed figures are discussed in &lt;a class="link" href="https://koreainvestinsights.com/post/sk-hynix-2q-earnings-cut-mirae-kis-lta-target-price-2026-07-14/" &gt;SK Hynix Q2 Earnings Cut and Maintained Price Targets&lt;/a&gt;.&lt;/p&gt;
&lt;h3 id="43-equipment-orders-are-both-current-demand-and-future-supply"&gt;4.3 Equipment Orders Are Both Current Demand and Future Supply
&lt;/h3&gt;&lt;p&gt;Reading ASML&amp;rsquo;s capacity expansion solely as near-term bullish evidence misses the time dimension.&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;Low-NA EUV 2027 capacity ≈ 65 units × 1.30 = 84.5 units
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;DUV immersion 2027 capacity ≈ 130 units × 1.30 = 169 units
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;Equipment manufacturers recognize orders as revenue in 2026–2027 first. Actual bit production at memory manufacturers can increase from 2028 onward, after equipment installation, yield stabilization, and packaging capacity additions. The same capex that initially boosts equipment and materials earnings later pressures memory prices and margins.&lt;/p&gt;
&lt;p&gt;The metrics to watch for 2028 are therefore not the announced investment totals. They are equipment installation timing, new wafer starts, yield ramp, packaging throughput, the increase in depreciation, and actual bit output.&lt;/p&gt;
&lt;h3 id="44-high-memory-prices-can-erode-customer-demand"&gt;4.4 High Memory Prices Can Erode Customer Demand
&lt;/h3&gt;&lt;p&gt;When memory prices rise, customers can respond in four ways:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Pull forward inventory purchases before prices rise further.&lt;/li&gt;
&lt;li&gt;Raise server and end-product prices.&lt;/li&gt;
&lt;li&gt;Reduce memory content or downgrade specifications.&lt;/li&gt;
&lt;li&gt;Delay purchases and projects.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;The first behavior makes near-term orders appear stronger but simply pulls forward future demand. The other three represent pathways through which suppliers&amp;rsquo; pricing power ultimately destroys customer demand. The IBM and Ericsson examples showed that memory supply constraints have begun to affect end-customer budget allocation and margins. We are currently in a phase where pre-buying and specification adjustments are visible before large-scale order cancellations.&lt;/p&gt;
&lt;h3 id="45-ai-capex-bottleneck-is-shifting-to-cost-of-capital"&gt;4.5 AI Capex Bottleneck Is Shifting to Cost of Capital
&lt;/h3&gt;&lt;p&gt;Even if AI demand is real, if the investing entity&amp;rsquo;s cash flow cannot keep pace with depreciation and financing costs, the pace of investment will slow. Going forward, the following four items matter more than the absolute capex figure from hyperscalers:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Whether AI and cloud gross margin growth outpaces the increase in depreciation.&lt;/li&gt;
&lt;li&gt;Whether data center power delivery and GPU utilization ramp according to plan.&lt;/li&gt;
&lt;li&gt;Whether returns on new projects, including borrowing costs, exceed the cost of capital.&lt;/li&gt;
&lt;li&gt;Whether the credit quality and up-front payments from customers that have signed HBM long-term contracts remain stable.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The strength of orders at TSMC and ASML means they are not evidence of current demand collapse. However, in U.S. Big Tech earnings releases at the end of July, investors should examine AI gross margins, free cash flow, and the slope of 2027 investment guidance alongside — not instead of — total capex figures.&lt;/p&gt;
&lt;h2 id="5-was-the-share-price-plunge-a-supply-demand-shock-or-a-fundamental-repricing"&gt;5. Was the Share Price Plunge a Supply-Demand Shock or a Fundamental Repricing?
&lt;/h2&gt;&lt;p&gt;This week&amp;rsquo;s sharp decline in Korean semiconductor stocks bears the hallmarks of leverage unwinding, program selling, and a market-cap structure heavily concentrated in a handful of large-cap names — all of which amplified the drawdown. Simultaneous selling by foreign and institutional investors caused single-day price moves far in excess of any change in underlying business value.&lt;/p&gt;
&lt;p&gt;Yet concluding that &amp;ldquo;it was merely a supply-demand shock, so everything rebounds&amp;rdquo; carries its own danger. If relative strength fails to recover even after strong earnings and pricing data, it may signal that the market is actively lowering its estimates for 2028 EPS and normalized multiples.&lt;/p&gt;
&lt;p&gt;The diagnostic framework is straightforward.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Signals pointing toward a supply-demand shock&lt;/th&gt;
 &lt;th&gt;Signals pointing toward a fundamental repricing&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Foreign and program selling decelerates quickly&lt;/td&gt;
 &lt;td&gt;Selling continues despite positive news&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Relative strength recovers on volume&lt;/td&gt;
 &lt;td&gt;Semiconductors continue to underperform the broader market&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;12-month forward EPS revisions remain positive&lt;/td&gt;
 &lt;td&gt;Samsung Electronics, SK Hynix, and Micron EPS all revised down simultaneously&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Contract prices rise and inventories stabilize&lt;/td&gt;
 &lt;td&gt;Inventories build while contract prices decline simultaneously&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Big-Tech CAPEX and AI profitability both improve&lt;/td&gt;
 &lt;td&gt;CAPEX cuts or deteriorating AI margins&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This distinction is precisely why price and earnings estimates must be tracked together. Watching only supply-demand dynamics risks missing long-term risks; watching only results risks missing the additional downside created by crowded positioning.&lt;/p&gt;
&lt;h2 id="6-china-is-three-variables-at-once"&gt;6. China Is Three Variables at Once
&lt;/h2&gt;&lt;h3 id="61-china-is-a-major-demand-market"&gt;6.1 China Is a Major Demand Market
&lt;/h3&gt;&lt;p&gt;China exposure figures disclosed in company filings use different definitions, making direct comparisons difficult.&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;China Exposure Metric&lt;/th&gt;
 &lt;th&gt;Caveat&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;30.1%&lt;/td&gt;
 &lt;td&gt;Based on standalone consolidated revenue — not memory-only share&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1Q26 24.3%, full-year 2025 19.7%&lt;/td&gt;
 &lt;td&gt;Based on sales entity location&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron&lt;/td&gt;
 &lt;td style="text-align: right"&gt;FY25 China + Hong Kong 10.1%&lt;/td&gt;
 &lt;td&gt;Based on customer headquarters location&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Samsung and SK Hynix are directly exposed if Chinese demand contracts. Micron&amp;rsquo;s direct exposure is comparatively low, but it cannot escape the impact if global pricing declines. Revenue-share figures alone are insufficient to assess China risk.&lt;/p&gt;
&lt;h3 id="62-china-is-a-commodity-memory-competitor"&gt;6.2 China Is a Commodity Memory Competitor
&lt;/h3&gt;&lt;p&gt;CXMT has announced mass production of LPDDR5X at 8,533 Mbps and 9,600 Mbps, with its 10,667 Mbps product reportedly in customer qualification. Its estimated share of global DRAM revenue for Q1 2026 stands at approximately 8%. However, an Apple supply agreement, broad adoption in flagship global products, and HBM yield have not been confirmed. &lt;a class="link" href="https://www.cxmt.com/en/news/info_19.html" target="_blank" rel="noopener"
 &gt;CXMT LPDDR5X Official Announcement&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;The near-term threat is concentrated in two channels rather than HBM displacement:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Localization of commodity DRAM and NAND in Chinese smartphones and PCs.&lt;/li&gt;
&lt;li&gt;Using CXMT as a fourth-supplier card in price negotiations with the established memory trio.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;If volume displaced from China migrates to other markets, it could pull down global contract prices for Samsung, SK Hynix, and Micron. This price contagion may matter more as a second-order effect than a direct decline in China revenue. A detailed exposure breakdown is available in &lt;a class="link" href="https://koreainvestinsights.com/post/china-memory-localization-exposure-samsung-hynix-micron-2026-07-18/" &gt;China Memory Localization and the Big Three&lt;/a&gt;.&lt;/p&gt;
&lt;h3 id="63-china-can-be-segmented-by-policy"&gt;6.3 China Can Be Segmented by Policy
&lt;/h3&gt;&lt;p&gt;On July 16, members of the House Select Committee on China sent a letter to the Department of Commerce calling for tightened controls on YMTC, accelerated review of CXMT for Entity List inclusion, restrictions on Chinese DRAM and HBM procurement in AI systems, data centers, federal IT, and critical infrastructure, and coordination with South Korea, Japan, and the EU. This is a policy demand — not an implemented purchasing ban. &lt;a class="link" href="https://chinaselectcommittee.house.gov/media/letters/moolenaar-whitesides-to-secretary-lutnick-hold-firm-on-chinese-memory-chips-ban" target="_blank" rel="noopener"
 &gt;House Select Committee on China Official Letter&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;If enacted, the global memory market could bifurcate along the following lines:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Market&lt;/th&gt;
 &lt;th&gt;Possible Structure&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;United States and allies&lt;/td&gt;
 &lt;td&gt;Verified supply chains, higher prices, market-share defense by the memory trio&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Chinese domestic market&lt;/td&gt;
 &lt;td&gt;Localization, lower prices, medium-term oversupply risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Micron stands to benefit most directly from U.S. policy. Samsung and SK Hynix carry both the advantage of being allied-nation suppliers and the exposure of their China businesses. This policy factor is less a catalyst that directly lifts 2026 HBM earnings and more a variable that reduces the China-oversupply discount rate priced into 2028 valuations.&lt;/p&gt;
&lt;h2 id="7-kimi-k3-and-efficiency-demand-shifting-not-demand-destruction"&gt;7. Kimi K3 and Efficiency: Demand Shifting, Not Demand Destruction
&lt;/h2&gt;&lt;p&gt;According to Kimi&amp;rsquo;s official announcement, K3 has 2.8 trillion total parameters, activates 16 of 896 experts per token, and supports a 1 million-token context window. The sparse expert architecture reduces compute per request, but still requires large memory and high-bandwidth networking to store all weights and route expert selection. &lt;a class="link" href="https://www.kimi.com/blog/kimi-k3" target="_blank" rel="noopener"
 &gt;Kimi K3 Official Announcement&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;The semiconductor impact of efficiency gains is better understood through the following equation:&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;Total semiconductor demand = silicon cost per task × total number of tasks × fraction self-hosted
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;ul&gt;
&lt;li&gt;Sparse architectures, caching, and quantization reduce GPU time and HBM footprint per task.&lt;/li&gt;
&lt;li&gt;Falling API prices increase AI usage volumes and the number of agents.&lt;/li&gt;
&lt;li&gt;Open-weight models may increase self-hosted server deployments by enterprises and governments.&lt;/li&gt;
&lt;li&gt;Memory tiering shifts some HBM workload onto server DRAM, eSSD, and remote caches.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Current firm orders and CAPEX data suggest that volume growth is outpacing efficiency gains. Over a longer horizon, however, the volume elasticity of server DRAM, eSSD, networking, and power may prove more favorable than the pricing power of premium GPUs and HBM. Efficiency is less a variable that &amp;ldquo;reduces total semiconductor consumption&amp;rdquo; and more one that reshapes &amp;ldquo;which semiconductors capture the revenue.&amp;rdquo;&lt;/p&gt;
&lt;p&gt;Two caveats apply here. The KV-cache savings rates and theoretical throughput from Kimi&amp;rsquo;s linear research models cannot be directly applied to K3 in production. Additionally, self-reported performance benchmarks should not be weighted equally with independent verification. Weights, licensing terms, independent throughput results, and actual tokens and total cost per task require confirmation.&lt;/p&gt;
&lt;h2 id="8-five-clocks--when-they-are-watched-together-the-contradictions-resolve"&gt;8. Five Clocks — When They Are Watched Together, the Contradictions Resolve
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Clock&lt;/th&gt;
 &lt;th&gt;This Week&amp;rsquo;s Observation&lt;/th&gt;
 &lt;th&gt;Investment Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Pricing&lt;/td&gt;
 &lt;td&gt;Sharp 2Q DRAM and NAND price increases; 3Q pace of gains decelerating&lt;/td&gt;
 &lt;td&gt;2026 earnings look strong, but the rate of increase may slow&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Investment&lt;/td&gt;
 &lt;td&gt;TSMC expanding investment; ASML orders and capacity growing&lt;/td&gt;
 &lt;td&gt;Current demand confirmed while 2027–2028 supply response is booked&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Construction&lt;/td&gt;
 &lt;td&gt;Power, land, and packaging bottlenecks persist&lt;/td&gt;
 &lt;td&gt;Announced investment lags before translating into actual production ramp&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Monetization&lt;/td&gt;
 &lt;td&gt;AI usage rising; customer costs and financing charges also rising&lt;/td&gt;
 &lt;td&gt;AI gross margin and free cash flow require scrutiny alongside revenue&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Share price&lt;/td&gt;
 &lt;td&gt;Sluggish reaction to positive news; concentrated supply-demand shock in Korea&lt;/td&gt;
 &lt;td&gt;Market is discounting duration and positioning rather than current earnings&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The pricing and investment clocks are running fast, while the construction and monetization clocks are lagging. The share-price clock is the forward-looking reflection of whether that gap narrows or widens ahead. That is why strong results and raised CAPEX guidance can trigger selling on the same day.&lt;/p&gt;
&lt;p&gt;The &amp;ldquo;55% industry strength / valuation weakness coexistence&amp;rdquo; from the prior piece and &amp;ldquo;P2 at 40%&amp;rdquo; below are not two different readings of the same table. The 55% was a broad category in which multiples remain compressed even as the industry stays strong. The more granular distribution here splits that category further into the early phases of P2 and P3. Because new data from TSMC and ASML confirmed current demand while simultaneously expanding the implied supply response, a portion of the long-term scarcity probability has been shifted toward supply normalization.&lt;/p&gt;
&lt;h2 id="9-scenario-update"&gt;9. Scenario Update
&lt;/h2&gt;&lt;p&gt;The probabilities below are judgment values reflecting information confirmed as of July 19 — not historical base rates.&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&lt;/th&gt;
 &lt;th&gt;Conditions for Validity&lt;/th&gt;
 &lt;th&gt;Expected Market Behavior&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;P1. Sustained Scarcity&lt;/td&gt;
 &lt;td style="text-align: right"&gt;25%&lt;/td&gt;
 &lt;td&gt;AI usage, CAPEX, and utilization rates continue to overwhelm supply and efficiency gains; HBM4 pricing and orders hold&lt;/td&gt;
 &lt;td&gt;High-beta memory names such as SK Hynix and Micron, alongside packaging and equipment, outperform&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;P2. Strong Earnings, Compressed Multiples&lt;/td&gt;
 &lt;td style="text-align: right"&gt;40%&lt;/td&gt;
 &lt;td&gt;Earnings remain strong in 2026–2027 but concerns over 2028 supply and customer profitability persist&lt;/td&gt;
 &lt;td&gt;Range-bound trading and elevated volatility despite earnings growth; Samsung Electronics carries relative advantage&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;P3. Efficiency and Supply Normalization&lt;/td&gt;
 &lt;td style="text-align: right"&gt;25%&lt;/td&gt;
 &lt;td&gt;2027–2028 capacity expansion, Chinese commodity supply, customer price resistance, and declining AI costs converge&lt;/td&gt;
 &lt;td&gt;HBM scarcity premium narrows; equipment and materials orders peak before memory itself&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;P4. Financial or Demand Dislocation&lt;/td&gt;
 &lt;td style="text-align: right"&gt;10%&lt;/td&gt;
 &lt;td&gt;Big-Tech CAPEX cuts, a credit event, rising inventories, and simultaneous contract price and EPS downgrades&lt;/td&gt;
 &lt;td&gt;Sector-wide semiconductor decline; cash and non-semiconductor diversification become more important&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The previous detailed distribution was 35% / 40% / 15% / 10%. This revision moves 10 percentage points from P1 to P3. TSMC and ASML order data did not raise the probability of P4. Conversely, equipment capacity expansion and the pricing burden on customers have increased the probability of supply and efficiency normalization relative to prolonged scarcity.&lt;/p&gt;
&lt;h2 id="10-roles-by-stock-and-value-chain"&gt;10. Roles by Stock and Value Chain
&lt;/h2&gt;&lt;h3 id="samsung-electronics-risk-adjusted-core"&gt;Samsung Electronics: Risk-Adjusted Core
&lt;/h3&gt;&lt;p&gt;Samsung absorbs commodity DRAM and NAND price appreciation across a broad market share base while retaining the option value of catching up in HBM and recovering its foundry business. It also holds relative defensive characteristics should open models and low-cost inference proliferate, shifting some HBM workload toward commodity memory and storage.&lt;/p&gt;
&lt;p&gt;Key risks include exposure to China and commodity memory, uncertainty around HBM4 customer qualification, and the high capital burden of foundry operations. Samsung offers more defensive characteristics than SK Hynix in P2 and P3, but diversification does not equal profitability if HBM catch-up has not been validated in earnings.&lt;/p&gt;
&lt;h3 id="sk-hynix-pure-beta-on-sustained-scarcity"&gt;SK Hynix: Pure Beta on Sustained Scarcity
&lt;/h3&gt;&lt;p&gt;SK Hynix&amp;rsquo;s leading HBM market share, co-development relationships with customers, and long-term supply agreements give it the strongest earnings leverage under P1. Conversely, it also carries the highest sensitivity to earnings and multiple compression if supply expansion and efficiency gains accelerate under P3.&lt;/p&gt;
&lt;p&gt;Additional judgment requires tracking HBM4 and HBM4E customer qualifications, yield, average selling price terms embedded in long-term contracts, and foreign-investor and program-trading flows alongside relative strength. Business quality and price-embedded margin of safety must be assessed separately.&lt;/p&gt;
&lt;h3 id="micron-intersection-of-us-policy-and-pure-play-memory"&gt;Micron: Intersection of U.S. Policy and Pure-Play Memory
&lt;/h3&gt;&lt;p&gt;Micron is the most direct beneficiary of U.S.-based manufacturing, policy support, and restrictions on Chinese memory procurement. It offers a clean read-through to the HBM, DRAM, and NAND upcycle. However, its limited business diversification means it is also directly exposed to any normalization in global average selling prices. Even with a low China revenue share, Micron cannot easily avoid the impact if surplus Chinese-origin volumes depress global pricing.&lt;/p&gt;
&lt;h3 id="sandisk-tactical-beta-on-nand-and-essd"&gt;SanDisk: Tactical Beta on NAND and eSSD
&lt;/h3&gt;&lt;p&gt;SanDisk is directly exposed to NAND supply constraints and AI server eSSD demand. It stands to benefit most if low-cost inference and memory tiering shift incremental storage volumes away from HBM. At the same time, NAND exhibits greater price volatility than DRAM and carries more direct YMTC exposure, giving it a more tactical character relative to core positions.&lt;/p&gt;
&lt;h3 id="tsmc-and-asml-demand-validators-and-leading-indicators-of-future-supply"&gt;TSMC and ASML: Demand Validators and Leading Indicators of Future Supply
&lt;/h3&gt;&lt;p&gt;TSMC demonstrates the quality of AI demand and the depth of the leading-edge foundry moat. ASML is the earliest observable data point for customers&amp;rsquo; long-term capacity expansion intentions. Both companies&amp;rsquo; results serve as supporting evidence for the memory thesis, but they are not simple coincident indicators. High CAPEX and growing equipment supply signal not only future memory demand but also future supply expansion.&lt;/p&gt;
&lt;h3 id="korean-equipment-and-materials-actual-orders-and-recurring-revenue-over-published-plans"&gt;Korean Equipment and Materials: Actual Orders and Recurring Revenue Over Published Plans
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Category&lt;/th&gt;
 &lt;th&gt;Candidate Names&lt;/th&gt;
 &lt;th&gt;Data to Confirm&lt;/th&gt;
 &lt;th&gt;Current Role&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Recurring consumables&lt;/td&gt;
 &lt;td&gt;Hana Materials, Wooldex&lt;/td&gt;
 &lt;td&gt;Customer utilization rates, OEM channels, revenue and operating margin correlation&lt;/td&gt;
 &lt;td&gt;Priority research group&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Process equipment&lt;/td&gt;
 &lt;td&gt;KC Tech, Nextin&lt;/td&gt;
 &lt;td&gt;Actual orders, customer qualification, backlog&lt;/td&gt;
 &lt;td&gt;Conditional confirmation group&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM back-end&lt;/td&gt;
 &lt;td&gt;Techwing&lt;/td&gt;
 &lt;td&gt;Additional Cube Prober orders and production revenue&lt;/td&gt;
 &lt;td&gt;Event confirmation group&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Infrastructure construction&lt;/td&gt;
 &lt;td&gt;Hanyang E&amp;amp;E, others&lt;/td&gt;
 &lt;td&gt;Major contract wins, order-to-revenue ratio, margins&lt;/td&gt;
 &lt;td&gt;Order-event group rather than core&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The confirmation sequence for investments is: capacity announcement → actual order placement → backlog build → revenue recognition → recurring consumables. Equipment names may peak in earnings earlier than memory production itself. Rather than buying broadly on fab plans alone, the focus should narrow to companies where actual orders and margins are confirmed.&lt;/p&gt;
&lt;h2 id="11-counter-arguments-where-this-thesis-could-be-wrong"&gt;11. Counter-Arguments: Where This Thesis Could Be Wrong
&lt;/h2&gt;&lt;h3 id="demand-evidence-may-have-been-double-counted"&gt;Demand Evidence May Have Been Double-Counted
&lt;/h3&gt;&lt;p&gt;Orders from TSMC, ASML, and the memory trio may all originate from the same Big-Tech capital expenditure plans. Adding orders across multiple tiers of the supply chain as separate demand signals overstates actual end demand. Power connection timelines, GPU utilization, server shipments, and cloud AI gross margins are the terminal verification data points.&lt;/p&gt;
&lt;h3 id="the-2030-demand-model-may-underestimate-efficiency-gains"&gt;The 2030 Demand Model May Underestimate Efficiency Gains
&lt;/h3&gt;&lt;p&gt;If any one of — token count, model size, context length, or HBM retention rate — is reduced, a shortage persists. But if all four decline simultaneously, a 2030 oversupply scenario becomes plausible. The directional case for long-term shortage and the numerical magnitude of that shortage must be assessed separately.&lt;/p&gt;
&lt;h3 id="capex-may-grow-faster-than-returns"&gt;CAPEX May Grow Faster Than Returns
&lt;/h3&gt;&lt;p&gt;Suppliers must spend more on new fabs and packaging even when order books are full. Customers must absorb data center depreciation and interest expense. If revenue growth fails to translate into free cash flow growth, high earnings may attract depressed multiples.&lt;/p&gt;
&lt;h3 id="china-policy-may-not-be-implemented"&gt;China Policy May Not Be Implemented
&lt;/h3&gt;&lt;p&gt;The House Committee letter is a policy signal — not an executive order or coordinated allied action. If CXMT enters only a limited set of consumer products within China and enforcement remains loose, pricing leverage for the incumbents may erode before any meaningful market-share shift occurs.&lt;/p&gt;
&lt;h3 id="open-model-company-disclosures-may-not-be-independently-reproducible"&gt;Open-Model Company Disclosures May Not Be Independently Reproducible
&lt;/h3&gt;&lt;p&gt;Kimi K3&amp;rsquo;s official specifications are confirmed, but throughput, real-world token consumption, and total serving cost require independent verification. Conversely, if efficiency gains arrive faster than expected and usage elasticity is lower than assumed, the HBM scarcity premium could compress more severely.&lt;/p&gt;
&lt;h3 id="the-entire-price-decline-may-have-been-attributed-to-forced-liquidation"&gt;The Entire Price Decline May Have Been Attributed to Forced Liquidation
&lt;/h3&gt;&lt;p&gt;Leverage unwinding and program selling explain the magnitude of the drop. But if prices fail to recover even in the face of positive news, the market may be actively lowering its 2028 earnings and normalized-multiple assumptions. A supply-demand explanation cannot substitute for validating earnings durability.&lt;/p&gt;
&lt;h2 id="12-observable-conditions-that-would-change-the-thesis"&gt;12. Observable Conditions That Would Change the Thesis
&lt;/h2&gt;&lt;p&gt;Rather than any single article or a one-day bounce, the key is whether two independent data axes move together.&lt;/p&gt;
&lt;h3 id="conditions-that-raise-the-probability-of-a-bullish-outcome"&gt;Conditions That Raise the Probability of a Bullish Outcome
&lt;/h3&gt;&lt;ol&gt;
&lt;li&gt;Big-Tech CAPEX growth and improvement in AI and cloud gross margins are confirmed simultaneously.&lt;/li&gt;
&lt;li&gt;Data center power connections, GPU utilization rates, and server shipments all rise together.&lt;/li&gt;
&lt;li&gt;HBM4, server DRAM, and eSSD prices, along with 12-month forward EPS, continue to be revised upward.&lt;/li&gt;
&lt;li&gt;Samsung Electronics, SK Hynix, and Micron rally on positive news and recover relative strength versus the semiconductor index.&lt;/li&gt;
&lt;li&gt;Foreign-investor and program selling decelerates, and a trend recovery accompanied by volume emerges.&lt;/li&gt;
&lt;/ol&gt;
&lt;h3 id="conditions-that-raise-the-probability-of-p3-normalization"&gt;Conditions That Raise the Probability of P3 Normalization
&lt;/h3&gt;&lt;ol&gt;
&lt;li&gt;2027–2028 equipment delivery and new bit capacity come online faster than planned.&lt;/li&gt;
&lt;li&gt;HBM yields and packaging productivity improve sharply while supply from all three incumbents expands simultaneously.&lt;/li&gt;
&lt;li&gt;Following memory price increases, customers reduce their content per device and purchasing deferrals become widespread.&lt;/li&gt;
&lt;li&gt;CXMT DDR5 and LPDDR5X customer qualifications and sales outside China increase.&lt;/li&gt;
&lt;li&gt;Server DRAM and HBM contract price growth rates and 2028 EPS are simultaneously revised lower.&lt;/li&gt;
&lt;/ol&gt;
&lt;h3 id="p4-warning-conditions"&gt;P4 Warning Conditions
&lt;/h3&gt;&lt;ol&gt;
&lt;li&gt;U.S. Big-Tech companies cut CAPEX or cancel or delay data center projects.&lt;/li&gt;
&lt;li&gt;Credit spreads and refinancing risk for AI infrastructure customers spike sharply.&lt;/li&gt;
&lt;li&gt;DRAM and NAND inventory builds and contract price declines occur simultaneously.&lt;/li&gt;
&lt;li&gt;12-month forward EPS for Samsung Electronics, SK Hynix, and Micron are all revised downward together.&lt;/li&gt;
&lt;li&gt;Semiconductor relative strength and market breadth (advance-decline) both deteriorate simultaneously.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id="13-data-to-track-from-late-july-onward"&gt;13. Data to Track from Late July Onward
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Timing&lt;/th&gt;
 &lt;th&gt;Data to Track&lt;/th&gt;
 &lt;th&gt;Judgment That Changes&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Late July&lt;/td&gt;
 &lt;td&gt;U.S. Big-Tech earnings and conference calls&lt;/td&gt;
 &lt;td&gt;CAPEX, AI gross margin, free cash flow, memory price resistance&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Late July&lt;/td&gt;
 &lt;td&gt;Additional Kimi K3 technical disclosures and independent evaluations&lt;/td&gt;
 &lt;td&gt;Which is larger — efficiency gains or usage growth&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Late July&lt;/td&gt;
 &lt;td&gt;Samsung Electronics and SK Hynix detailed IR disclosures&lt;/td&gt;
 &lt;td&gt;HBM4, long-term contracts, DRAM and NAND ASP, shipments, CAPEX&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;August onward&lt;/td&gt;
 &lt;td&gt;3Q contract pricing and distribution inventory levels&lt;/td&gt;
 &lt;td&gt;Whether the pace of price gains has peaked and pre-buying is normalizing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Quarterly&lt;/td&gt;
 &lt;td&gt;ASML bookings, TSMC CAPEX, memory equipment orders&lt;/td&gt;
 &lt;td&gt;Speed of 2027–2028 supply response&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Ongoing&lt;/td&gt;
 &lt;td&gt;CXMT and YMTC regulatory actions and customer qualifications&lt;/td&gt;
 &lt;td&gt;Chinese supply&amp;rsquo;s price contagion to global markets and market bifurcation&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="14-final-judgment"&gt;14. Final Judgment
&lt;/h2&gt;&lt;p&gt;This week served as the first comprehensive stress test of the memory supercycle narrative. Share prices moved sharply, but TSMC and ASML results, DRAM and NAND pricing, and server memory demand all tilted the weight of evidence toward 2026–2027 physical shortages remaining intact.&lt;/p&gt;
&lt;p&gt;At the same time, the same data made 2028 risks more clearly defined. Equipment capacity and CAPEX are expanding, long-term HBM supply agreements slow the pass-through of price increases, customers must absorb elevated memory prices alongside data center financing costs, and Chinese commodity memory and open-model efficiency gains will exert pressure on 2028 pricing and multiples.&lt;/p&gt;
&lt;p&gt;The choice between &amp;ldquo;the boom is over&amp;rdquo; and &amp;ldquo;this is a permanent supercycle&amp;rdquo; is therefore a false one. The current base-case path is as follows:&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;2026–2027: Physical shortage and high earnings
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;-&amp;gt; 2027–2028: Equipment, capacity, and yield response
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;-&amp;gt; Pace of price increases decelerates
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;-&amp;gt; Absolute earnings remain above historical levels, but multiples compress
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;-&amp;gt; Return dispersion across names driven by product mix, contract structure, and China exposure
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;From this point, the central task is not to re-prove that AI demand exists. It is to measure simultaneously the speed of the supply response, the monetization trajectory for customers, contract quality, 2028 earnings estimates, and how share prices react to positive news. Earnings today are speaking to 2026 and 2027. The market is already asking about 2028.&lt;/p&gt;
&lt;hr&gt;
&lt;h2 id="key-sources"&gt;Key Sources
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://investor.tsmc.com/english/quarterly-results/2026/q2" target="_blank" rel="noopener"
 &gt;TSMC Q2 2026 Official Earnings&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.asml.com/en/news/press-releases/2026/q2-2026-financial-results" target="_blank" rel="noopener"
 &gt;ASML Q2 2026 Official Earnings and Capacity Plans&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.trendforce.com/presscenter/news/20260331-12995.html" target="_blank" rel="noopener"
 &gt;TrendForce Q2 2026 DRAM and NAND Price Outlook&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.trendforce.com/presscenter/news/20260709-13140.html" target="_blank" rel="noopener"
 &gt;TrendForce Q3 2026 Server DRAM Outlook&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://chinaselectcommittee.house.gov/media/letters/moolenaar-whitesides-to-secretary-lutnick-hold-firm-on-chinese-memory-chips-ban" target="_blank" rel="noopener"
 &gt;U.S. House Select Committee on China — Letter Calling for Restrictions on Chinese Memory Procurement&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.bis.gov/ear/title-15/subtitle-b/chapter-vii/subchapter-c/part-744/ss-74416-entity-list" target="_blank" rel="noopener"
 &gt;U.S. Department of Commerce BIS Entity List Regulations&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.kimi.com/blog/kimi-k3" target="_blank" rel="noopener"
 &gt;Kimi K3 Official Announcement&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.cxmt.com/en/news/info_19.html" target="_blank" rel="noopener"
 &gt;CXMT LPDDR5X Official Announcement&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="limitations-of-public-data"&gt;Limitations of Public Data
&lt;/h2&gt;&lt;ol&gt;
&lt;li&gt;The long-term contract pricing formulas for HBM from Samsung Electronics, SK Hynix, and Micron — including floor and ceiling prices, minimum purchase volumes, cancellation clauses, and customer-specific credit terms — have not been disclosed publicly.&lt;/li&gt;
&lt;li&gt;There are no reliable official figures for CXMT&amp;rsquo;s HBM yield or its progress toward Apple qualification.&lt;/li&gt;
&lt;li&gt;Full unit cost by model and actual per-accelerator serving costs for Chinese AI API providers are not publicly reported.&lt;/li&gt;
&lt;li&gt;The 26.7 EB HBM demand figure for 2030 is not a consensus forecast but a scenario constructed by combining several bullish assumptions.&lt;/li&gt;
&lt;li&gt;The July 19 channel checks on server DRAM and HBM4 could not be independently sourced from primary documents and are used solely for directional reference.&lt;/li&gt;
&lt;li&gt;The information cutoff date is July 19, 2026. Scenarios should be revisited following earnings releases from U.S. large-cap tech companies and memory makers expected in late July.&lt;/li&gt;
&lt;/ol&gt;

 &lt;blockquote&gt;
 &lt;p&gt;This article is an informational analysis based on publicly available sources and does not constitute a recommendation to buy or sell any specific security. Prices, supply-demand dynamics, earnings estimates, and policy conditions are subject to change; readers should verify the latest disclosures and assess findings against their own risk tolerance.&lt;/p&gt;

 &lt;/blockquote&gt;</description></item><item><title>U.S. House Letter Seeking Chinese Memory Procurement Restrictions: CXMT, YMTC, and the 2028 Cycle</title><link>https://koreainvestinsights.com/post/us-house-chinese-memory-procurement-ban-cxmt-ymtc-cycle-impact-2026-07-17/</link><pubDate>Fri, 17 Jul 2026 20:45:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/us-house-chinese-memory-procurement-ban-cxmt-ymtc-cycle-impact-2026-07-17/</guid><description>&lt;p&gt;On July 16, 2026, House Select Committee on China Chairman John Moolenaar and Ranking Member George Whitesides sent a letter to the Secretary of Commerce. The two lawmakers called for tighter export controls on CXMT and YMTC, a prohibition on U.S. companies procuring their memory for AI systems and data centers, and allied participation — from South Korea, Japan, and the EU — in the same supply-chain rules.&lt;/p&gt;
&lt;p&gt;The letter is real. Its language is strong. It is not, however, a regulation, an executive order, or a statute. &lt;strong&gt;This is a bipartisan policy demand addressed to the executive branch, not an enacted Chinese memory embargo.&lt;/strong&gt; Conflating the two risks materially overstating the near-term significance for equity prices.&lt;/p&gt;
&lt;p&gt;That does not mean the letter is inconsequential. If these demands translate into actual policy, the United States and its allies will find it harder to fill memory shortfalls with Chinese supply. This is not an event that immediately increases HBM shipments, but it does reduce the concern that CXMT and YMTC capacity additions in 2027–2028 will collapse global pricing. The China oversupply discount that investors have been applying to normalized earnings and target multiples for Samsung, SK Hynix, and Micron could narrow.&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Context: This article follows &lt;a class="link" href="https://koreainvestinsights.com/post/apple-cxmt-dram-china-test-memory-pricing-korea-2026-07-14/" &gt;Will Apple Really Use CXMT?&lt;/a&gt; and &lt;a class="link" href="https://koreainvestinsights.com/post/cxmt-ipo-memory-price-risk-hbm-client-dram-2026-06-21/" &gt;CXMT IPO and Memory Price Risk&lt;/a&gt;. Those earlier pieces addressed Chinese memory supply expansion and Apple&amp;rsquo;s use of alternative suppliers as a negotiating lever. This piece examines how the memory cycle changes if the United States moves to close that supply channel through policy. Related material is available at the &lt;a class="link" href="https://koreainvestinsights.com/page/korea-semiconductor-hbm-kospi-hub/" &gt;AI HBM Hub&lt;/a&gt; and the &lt;a class="link" href="https://koreainvestinsights.com/page/exclusive-analysis-hub/" &gt;Exclusive Analysis 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;The July 16 House Select Committee letter has been officially confirmed. It is, however, a policy demand from two members of Congress addressed to the Commerce Department — not an enacted regulation.&lt;/li&gt;
&lt;li&gt;Three demands are on the table: strengthen controls on YMTC and accelerate CXMT&amp;rsquo;s Entity List review; prohibit U.S. persons and entities from procuring Chinese DRAM, HBM, and other memory for AI systems, data centers, federal IT, and critical infrastructure; coordinate with South Korea, Japan, and the EU on equivalent standards.&lt;/li&gt;
&lt;li&gt;The Entity List restricts the export, re-export, and in-country transfer of items subject to U.S. Export Administration Regulations when a listed entity is a party to the transaction. It does not automatically prohibit U.S. companies from purchasing finished Chinese memory products domestically. A purchase ban requires a separate executive order, agency directive, or legislation.&lt;/li&gt;
&lt;li&gt;If enacted, the policy is positive for market share and pricing power at Samsung, SK Hynix, and Micron. SanDisk and Kioxia also benefit as YMTC&amp;rsquo;s path into global NAND and enterprise SSD markets becomes more constrained.&lt;/li&gt;
&lt;li&gt;Product-level directness varies. Mobile and PC DRAM and NAND could feel the impact immediately; advanced HBM faces limited near-term supply alternatives regardless. This is more relevant to defending post-2028 competitive risk and multiples for SK Hynix than to 2026–2027 HBM EPS.&lt;/li&gt;
&lt;li&gt;The latest pricing outlook already signals a deceleration. TrendForce projected Q2 commodity DRAM contract prices up 58–63% quarter-over-quarter and NAND up 70–75%, but trimmed Q3 server DRAM growth to 13–18%. This letter is a policy signal that increases the plausibility of a sustained supply shortage into 2027 rather than a catalyst for another sharp near-term price spike.&lt;/li&gt;
&lt;li&gt;Our current stance is &lt;strong&gt;Watchlist — incorporate upon policy confirmation&lt;/strong&gt;. Chasing memory equities on the letter alone is premature; the correct sequence is to watch for a CXMT Entity List notice, a U.S. corporate procurement restriction, allied coordination, Apple&amp;rsquo;s sourcing decisions, and Q3 contract pricing.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;img alt="Transmission channels from the U.S. congressional Chinese memory policy letter" loading="lazy" sizes="(max-width: 767px) calc(100vw - 30px), (max-width: 1023px) 700px, (max-width: 1279px) 950px, 1232px" src="https://koreainvestinsights.com/images/posts/us-house-chinese-memory-policy-impact-2026-07-17.png"&gt;
&lt;/p&gt;
&lt;div class="thesis-callout"&gt;
 &lt;div class="thesis-callout__label"&gt;One-line view&lt;/div&gt;
 &lt;div class="thesis-callout__body"&gt;
 This letter is not a HBM demand shock. It is a policy signal aimed at reducing the risk that Chinese-origin oversupply in 2027–2028 destroys global memory profitability.
 &lt;/div&gt;
&lt;/div&gt;
&lt;h2 id="1-three-demands-confirmed-in-the-letter"&gt;1. Three Demands Confirmed in the Letter
&lt;/h2&gt;&lt;p&gt;The signatories are Republican John Moolenaar, Chairman of the House Select Committee on China, and Democrat George Whitesides. The letter is addressed to Commerce Secretary Howard Lutnick. The bipartisan signal is clear given that members of opposing parties co-signed. It is not, however, a full congressional vote or a Commerce Department decision.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Demand&lt;/th&gt;
 &lt;th&gt;Actual Content&lt;/th&gt;
 &lt;th&gt;Current Status&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Tighten supply controls&lt;/td&gt;
 &lt;td&gt;Strengthen Entity List restrictions on YMTC; expedite formal Entity List review for CXMT&lt;/td&gt;
 &lt;td&gt;Policy request, not finalized&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Create demand-side restrictions&lt;/td&gt;
 &lt;td&gt;Prohibit U.S. persons and entities from procuring DRAM, HBM, and other memory from YMTC, CXMT, and similar firms for AI systems, data centers, federal IT, and critical infrastructure&lt;/td&gt;
 &lt;td&gt;Requires separate executive order or agency directive&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Allied coordination&lt;/td&gt;
 &lt;td&gt;Work with South Korea, Japan, and the EU to prevent CXMT and YMTC from entering allied supply chains&lt;/td&gt;
 &lt;td&gt;Diplomatic and policy coordination request&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The official letter also cited concerns that Apple and other U.S. technology companies are exploring purchases of Chinese memory. The demand is therefore not an abstract security slogan — it is an attempt to foreclose demand channels before CXMT and YMTC can achieve customer qualification and meaningful entry into global supply chains. &lt;a class="link" href="https://chinaselectcommittee.house.gov/media/letters/moolenaar-whitesides-to-secretary-lutnick-hold-firm-on-chinese-memory-chips-ban" target="_blank" rel="noopener"
 &gt;Official House Select Committee Letter&lt;/a&gt;&lt;/p&gt;
&lt;h3 id="separating-fact-from-overstatement"&gt;Separating Fact from Overstatement
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Statement&lt;/th&gt;
 &lt;th&gt;Verdict&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;U.S. lawmakers demanded restrictions on Chinese memory procurement&lt;/td&gt;
 &lt;td&gt;Fact&lt;/td&gt;
 &lt;td&gt;Official letter dated July 16 confirmed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;A full Chinese memory embargo has been finalized&lt;/td&gt;
 &lt;td&gt;Overstatement&lt;/td&gt;
 &lt;td&gt;No executive branch decision or implementing regulations&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CXMT&amp;rsquo;s Entity List designation has been confirmed&lt;/td&gt;
 &lt;td&gt;Unconfirmed&lt;/td&gt;
 &lt;td&gt;Only an accelerated review was requested&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Apple&amp;rsquo;s devices sold in China are automatically prohibited&lt;/td&gt;
 &lt;td&gt;Overstatement&lt;/td&gt;
 &lt;td&gt;The letter&amp;rsquo;s explicit scope covers AI, data centers, federal IT, and critical infrastructure&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;South Korea and Japan have banned Chinese memory&lt;/td&gt;
 &lt;td&gt;False&lt;/td&gt;
 &lt;td&gt;Coordination was requested; no national measures have been announced&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The policy direction is forceful, but multiple implementation thresholds remain. Equity analysis should therefore start from &amp;ldquo;which discount disappears if this is implemented&amp;rdquo; rather than &amp;ldquo;the ban is already in effect.&amp;rdquo;&lt;/p&gt;
&lt;h2 id="2-the-entity-list-and-a-purchase-prohibition-are-distinct-instruments"&gt;2. The Entity List and a Purchase Prohibition Are Distinct Instruments
&lt;/h2&gt;&lt;p&gt;There is a reason the letter requests supply-side and demand-side controls separately: the two instruments operate differently under U.S. law.&lt;/p&gt;
&lt;p&gt;The Bureau of Industry and Security&amp;rsquo;s Entity List imposes license requirements on the export, re-export, or in-country transfer of items subject to the Export Administration Regulations when a listed entity is a party to the transaction. The covered items include U.S.-origin semiconductor equipment, components, design tools, software, and process support. &lt;a class="link" href="https://www.bis.gov/ear/title-15/subtitle-b/chapter-vii/subchapter-c/part-744/ss-74416-entity-list" target="_blank" rel="noopener"
 &gt;BIS Entity List Regulations&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;This mechanism pressures CXMT and YMTC on capacity and technology-node transitions. If equipment parts cannot be replaced on schedule and process support is unavailable, yields, wafer starts, and the pace of leading-edge conversion can slow.&lt;/p&gt;
&lt;p&gt;It does not, however, automatically prohibit U.S. companies from purchasing finished DRAM or NAND chips that were manufactured in China. That is precisely why the lawmakers included a separate procurement prohibition as their second demand. The structure is as follows:&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;Entity List tightening
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Restricts the path by which U.S.-origin equipment and technology reach Chinese memory makers
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Pressures capacity, yield, and technology-node transition
&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;Separate procurement ban
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Restricts the path by which U.S. companies purchase finished Chinese memory products
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Pressures customer qualification, revenue, and global market share
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;The maximum effect requires both instruments operating simultaneously — making production harder at the top of the supply chain while cutting off customers at the bottom. If only the Entity List is tightened without a purchase restriction, Chinese firms can route around it using non-U.S. equipment and domestic customers. If only a procurement ban is imposed without strong equipment controls, Chinese domestic capacity can continue to expand.&lt;/p&gt;
&lt;h3 id="how-far-does-this-reach-for-apple"&gt;How Far Does This Reach for Apple?
&lt;/h3&gt;&lt;p&gt;The letter&amp;rsquo;s explicitly enumerated scope covers AI systems, data centers, federal IT, and critical infrastructure. Extending the prohibition to cover standard consumer iPhones and iPads sold within China would require either a broader definition in the implementing rules or a separate action.&lt;/p&gt;
&lt;p&gt;It is therefore premature to conclude that &lt;a class="link" href="https://koreainvestinsights.com/post/apple-cxmt-dram-china-test-memory-pricing-korea-2026-07-14/" &gt;Apple&amp;rsquo;s CXMT testing&lt;/a&gt; has been made immediately impossible. What has changed materially, however, is the regulatory stability Apple needs to use CXMT as a negotiating lever. If a supplier could be added to the Entity List at any time, and if certain products could become subject to procurement restrictions without advance notice, proceeding with volume qualification and long-term contracts becomes much more difficult.&lt;/p&gt;
&lt;h2 id="3-the-transmission-path-to-the-memory-cycle"&gt;3. The Transmission Path to the Memory Cycle
&lt;/h2&gt;&lt;p&gt;The core transmission channel is straightforward.&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;Procurement restrictions on Chinese memory
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Narrowed customer qualification and sales channels for CXMT and YMTC globally
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Market share defense at Samsung, SK Hynix, Micron, SanDisk, and Kioxia
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Delayed inflection point for DRAM and NAND price declines
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Higher expected normalized earnings and cash flows in 2027–2028
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Compression of the China oversupply discount
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;The memory shortage itself is not new information. TrendForce projected Q2 2026 commodity DRAM contract prices up 58–63% quarter-over-quarter and NAND contract prices up 70–75%, driven by capacity prioritization toward HBM, server DRAM, and enterprise SSD at the expense of standard DRAM and NAND. &lt;a class="link" href="https://www.trendforce.com/presscenter/news/20260331-12995.html" target="_blank" rel="noopener"
 &gt;TrendForce Q2 Outlook&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;The latest forecast also shows a deceleration in the rate of price increase. On July 9, TrendForce projected Q3 server DRAM contract prices rising 13–18%. It noted that long-term supply agreements are capping some price upside, but also that server DRAM tightness could extend into 2027 — with RDIMM bit supply growth of 15–20% potentially failing to keep pace with server CPU shipment growth. Prices could continue to rise quarter-over-quarter from H2 2026 through H2 2027, but at a more gradual pace. &lt;a class="link" href="https://www.trendforce.com/presscenter/news/20260709-13140.html" target="_blank" rel="noopener"
 &gt;TrendForce Q3 Outlook&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;Two conclusions follow from this data.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;The near-term price spike is already in the market. The letter alone is insufficient to restart Q2-style rates of increase.&lt;/li&gt;
&lt;li&gt;The more important question is how quickly Chinese capacity additions end the 2027 supply shortage. If the United States and its allies exclude Chinese memory, the duration of the shortage extends.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id="4-product-level-impact-is-not-uniform"&gt;4. Product-Level Impact Is Not Uniform
&lt;/h2&gt;&lt;p&gt;Treating CXMT and YMTC as a single &amp;ldquo;Chinese memory&amp;rdquo; block obscures the actual impact. CXMT&amp;rsquo;s core product is DRAM; YMTC&amp;rsquo;s is NAND. Customer qualification timelines and supply bottlenecks also differ by product.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Product&lt;/th&gt;
 &lt;th&gt;Policy Directness&lt;/th&gt;
 &lt;th&gt;Impact Channel&lt;/th&gt;
 &lt;th&gt;Primary Beneficiaries&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Mobile and PC DRAM&lt;/td&gt;
 &lt;td&gt;High&lt;/td&gt;
 &lt;td&gt;Limits CXMT&amp;rsquo;s qualification with Apple and PC customers and global volume expansion&lt;/td&gt;
 &lt;td&gt;Samsung, Micron, some SK Hynix&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Server DRAM&lt;/td&gt;
 &lt;td&gt;Medium-High&lt;/td&gt;
 &lt;td&gt;Data center procurement restrictions block Chinese server memory entry&lt;/td&gt;
 &lt;td&gt;Samsung, SK Hynix, Micron&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM&lt;/td&gt;
 &lt;td&gt;Low near-term, Medium-High long-term&lt;/td&gt;
 &lt;td&gt;Limited near-term supply alternatives; reduces China&amp;rsquo;s post-2028 catch-up risk&lt;/td&gt;
 &lt;td&gt;SK Hynix, Samsung, Micron&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Consumer NAND&lt;/td&gt;
 &lt;td&gt;High&lt;/td&gt;
 &lt;td&gt;Limits YMTC&amp;rsquo;s entry into global smartphone and PC storage&lt;/td&gt;
 &lt;td&gt;SanDisk, Kioxia, Samsung&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Enterprise SSD&lt;/td&gt;
 &lt;td&gt;High&lt;/td&gt;
 &lt;td&gt;Likely falls directly within the U.S. data center procurement scope&lt;/td&gt;
 &lt;td&gt;SanDisk, Kioxia, Samsung, SK Hynix/Solidigm&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h3 id="hbm-multiple-rather-than-near-term-eps"&gt;HBM: Multiple Rather Than Near-Term EPS
&lt;/h3&gt;&lt;p&gt;CXMT is not currently a viable near-term substitute for leading-edge HBM from Samsung, SK Hynix, or Micron. Customer qualification, stacking yields, advanced packaging, interposers, and long-term supply agreements constitute high barriers to entry. Even if U.S. data centers were prohibited from purchasing CXMT HBM today, SK Hynix&amp;rsquo;s 2026 shipments would not increase immediately as a result.&lt;/p&gt;
&lt;p&gt;The calculus changes when projecting past 2028. The market discounts the risk that Chinese firms deploy cash earned from commodity DRAM and state subsidies to pursue HBM development and close the gap with leading players. If equipment controls and customer exclusion operate simultaneously, the pace of that catch-up and the scale at which it can be commercialized are reduced. This effect appears first in normalized P/E and terminal value rather than current-year EPS.&lt;/p&gt;
&lt;h3 id="more-direct-impact-on-mobile-dram-and-nand"&gt;More Direct Impact on Mobile DRAM and NAND
&lt;/h3&gt;&lt;p&gt;The Chinese alternatives Apple tested are closer to mobile and consumer memory. If CXMT establishes itself as a qualified supplier to Apple or global PC makers, the pricing leverage of the incumbent three erodes. If YMTC enters U.S. data center SSD and consumer device markets, the ceiling on NAND pricing falls as well.&lt;/p&gt;
&lt;p&gt;The letter is designed to block precisely these qualification pathways. Accordingly, by product portfolio, Samsung and Micron are more direct beneficiaries on DRAM, while SanDisk and Kioxia benefit more directly on NAND.&lt;/p&gt;
&lt;h2 id="5-impact-by-listed-company"&gt;5. Impact by Listed Company
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Company / Group&lt;/th&gt;
 &lt;th&gt;Impact&lt;/th&gt;
 &lt;th&gt;Investment Interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron &lt;code&gt;MU&lt;/code&gt;&lt;/td&gt;
 &lt;td&gt;Strongly positive&lt;/td&gt;
 &lt;td&gt;The only large-scale U.S. memory maker. Full coverage across DRAM, HBM, and NAND makes Micron the most direct beneficiary of security-driven procurement and commercial customer diversification. Whether elevated policy expectations are already in the valuation requires a separate check.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics &lt;code&gt;005930&lt;/code&gt;&lt;/td&gt;
 &lt;td&gt;Positive&lt;/td&gt;
 &lt;td&gt;Slower Chinese entry into mobile DRAM, commodity DRAM, and NAND is directly beneficial. Combined with an HBM recovery, earnings defensibility across the entire product portfolio improves.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix &lt;code&gt;000660&lt;/code&gt;&lt;/td&gt;
 &lt;td&gt;Positive&lt;/td&gt;
 &lt;td&gt;Long-term competitive risk in server DRAM and HBM diminishes. Near-term HBM volume upside is limited; the more significant effect is multiple defense for post-2028 earnings durability. Enterprise SSD exposure through Solidigm is also relevant.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SanDisk &lt;code&gt;SNDK&lt;/code&gt; / Kioxia&lt;/td&gt;
 &lt;td&gt;Positive&lt;/td&gt;
 &lt;td&gt;YMTC&amp;rsquo;s entry into U.S. and allied consumer NAND and enterprise SSD supply chains becomes more constrained. Favorable for NAND pricing and customer share retention.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Apple, PC OEMs, server OEMs&lt;/td&gt;
 &lt;td&gt;Negative&lt;/td&gt;
 &lt;td&gt;Supplier optionality and pricing leverage are reduced. Higher component costs must be absorbed through product pricing, specifications, or margins.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Applied Materials, Lam Research, KLA&lt;/td&gt;
 &lt;td&gt;Conditionally negative&lt;/td&gt;
 &lt;td&gt;Tighter equipment controls on CXMT and YMTC create headwinds for China revenue. Incremental capacity buildout in the U.S., South Korea, and Japan could offset some demand.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CXMT / YMTC&lt;/td&gt;
 &lt;td&gt;Negative abroad, mixed domestically&lt;/td&gt;
 &lt;td&gt;International qualification and customer acquisition become harder. Domestic Chinese substitution, state subsidies, and localized equipment and materials transitions could accelerate.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h3 id="differentiating-samsung-and-sk-hynix"&gt;Differentiating Samsung and SK Hynix
&lt;/h3&gt;&lt;p&gt;This policy is marginally more direct for Samsung. Samsung&amp;rsquo;s exposure surface to Chinese competitors spans mobile DRAM, commodity DRAM, and NAND simultaneously. Blocking Chinese suppliers from global market entry defends both pricing and share across those product lines.&lt;/p&gt;
&lt;p&gt;SK Hynix also benefits. Its current earnings, however, are concentrated in HBM and server DRAM, where CXMT is not yet a credible near-term threat. Near-term consensus estimates do not change materially. Instead, the reduction in post-2028 HBM catch-up risk and commodity DRAM oversupply concern strengthens the case that elevated profitability can persist longer — an argument that applies more to the duration of normalized earnings than to the next quarter&amp;rsquo;s EPS.&lt;/p&gt;
&lt;p&gt;The letter alone is not sufficient grounds to reverse the relative priority between Samsung and SK Hynix. By product portfolio, it is a more direct earnings-defense catalyst for Samsung and more of a structural moat extension for SK Hynix&amp;rsquo;s HBM competitive position.&lt;/p&gt;
&lt;h2 id="6-the-global-memory-market-could-bifurcate"&gt;6. The Global Memory Market Could Bifurcate
&lt;/h2&gt;&lt;p&gt;Chinese production capacity does not disappear. What changes is the geography of permissible sales.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Market&lt;/th&gt;
 &lt;th&gt;Expected Structure&lt;/th&gt;
 &lt;th&gt;Pricing and Earnings Impact&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;United States and allies&lt;/td&gt;
 &lt;td&gt;Centered on qualified non-Chinese suppliers; stricter security procurement and origin rules&lt;/td&gt;
 &lt;td&gt;Shortage conditions persist; higher ASPs and margins become sustainable&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;China domestic&lt;/td&gt;
 &lt;td&gt;CXMT and YMTC localization accelerates; subsidies and transition to local equipment and materials&lt;/td&gt;
 &lt;td&gt;Market share expansion possible; medium-term oversupply and price competition risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Emerging markets&lt;/td&gt;
 &lt;td&gt;Potential acceptance of Chinese memory as cost and regulation trade-offs are weighed&lt;/td&gt;
 &lt;td&gt;Could become a bypass sales channel for Chinese producers&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;This bifurcation is not uniformly positive for Samsung, SK Hynix, and Micron. In the U.S. and allied markets, pricing and margins may improve. In Chinese customer markets, however, the transition to domestic suppliers could accelerate. The net effect is a &lt;strong&gt;trade-off between higher global margins and a smaller China addressable market&lt;/strong&gt;.&lt;/p&gt;
&lt;p&gt;If Chinese volumes concentrate domestically and in emerging markets, regional price divergence for the same DRAM and NAND products could widen. U.S. and allied buyers pay a premium for qualified suppliers, while China&amp;rsquo;s domestic capacity additions drive price competition internally. The premise that global memory prices move as one unified market weakens.&lt;/p&gt;
&lt;h2 id="7-share-prices-may-reflect-this-through-discount-rates-before-eps"&gt;7. Share Prices May Reflect This Through Discount Rates Before EPS
&lt;/h2&gt;&lt;p&gt;Memory equity valuations are not determined by the next quarter&amp;rsquo;s earnings alone. Investors simultaneously assess how long cyclical peak earnings persist, how quickly new supply arrives, and where normalized margins ultimately settle.&lt;/p&gt;
&lt;p&gt;Chinese capacity additions compress valuations through three channels.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;They create concern that commodity DRAM and NAND prices could fall sharply in 2027–2028.&lt;/li&gt;
&lt;li&gt;They raise the probability that state-backed late entrants can challenge incumbents in HBM.&lt;/li&gt;
&lt;li&gt;They give customers leverage to use a fourth supplier as a pricing card in negotiations.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;If procurement restrictions and equipment controls are actually implemented, all three risks diminish. The effect may therefore appear in the duration of normalized earnings and target multiples before it shows up in next-quarter EPS revisions.&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;Policy not enacted
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Chinese capacity additions flow into global supply
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ 2027–2028 price decline risk persists
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Normalized earnings and P/E discounted
&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;Policy enacted
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Barrier between Chinese capacity and global customers
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Speed of oversupply transmission slows
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Earnings duration revised upward
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ China oversupply discount compresses
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;Chinese production capacity does not vanish entirely. If China&amp;rsquo;s domestic price declines spill over into emerging markets, or if non-U.S. equipment ecosystems mature faster than expected, discount rates could widen again.&lt;/p&gt;
&lt;h2 id="8-three-policy-scenarios"&gt;8. Three Policy Scenarios
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Scenario&lt;/th&gt;
 &lt;th&gt;Development&lt;/th&gt;
 &lt;th&gt;Impact on Memory Equities&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;1. Letter only&lt;/td&gt;
 &lt;td&gt;CXMT review delayed; no separate purchase ban; no allied coordination&lt;/td&gt;
 &lt;td&gt;Near-term news effect fades. Existing China oversupply risk intact.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2. U.S.-only partial implementation&lt;/td&gt;
 &lt;td&gt;CXMT added to Entity List; procurement restrictions applied to AI, data centers, and federal IT&lt;/td&gt;
 &lt;td&gt;Direct benefit for Micron. Korean companies also defend U.S. customer share and pricing.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;3. Allied joint implementation&lt;/td&gt;
 &lt;td&gt;U.S. actions joined by South Korea, Japan, and EU; customer qualification and equipment/materials channels blocked simultaneously&lt;/td&gt;
 &lt;td&gt;Deep global supply chain bifurcation. Most positive for long-term multiples at the incumbent three and leading NAND players.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The third scenario is the most consequential variable. If only the United States restricts access while China successfully enters South Korean, Japanese, and European supply chains, the global pricing defense is limited. Conversely, if allied nations apply equivalent origin and security standards, CXMT and YMTC capacity additions — however large — struggle to translate into global customer revenue.&lt;/p&gt;
&lt;h2 id="9-catalysts-to-monitor"&gt;9. Catalysts to Monitor
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item to Confirm&lt;/th&gt;
 &lt;th&gt;Significance&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;BIS CXMT Entity List review or Federal Register notice&lt;/td&gt;
 &lt;td&gt;Confirms whether supply-side controls are actually initiated&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Commerce Department or White House U.S. corporate purchase restriction&lt;/td&gt;
 &lt;td&gt;Confirms whether demand channels for finished Chinese memory products are closed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Apple&amp;rsquo;s CXMT or YMTC qualification or procurement halt&lt;/td&gt;
 &lt;td&gt;Confirms whether policy pressure reaches consumer electronics at scale&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;South Korea, Japan, EU coordination announcements&lt;/td&gt;
 &lt;td&gt;Confirms whether Chinese capacity is actually blocked from flowing into global pricing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Q3 DRAM and NAND contract prices&lt;/td&gt;
 &lt;td&gt;Confirms whether fundamentals — not just policy expectations — support earnings estimates&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CXMT and YMTC domestic capacity additions and local equipment transition&lt;/td&gt;
 &lt;td&gt;Indicates how long equipment controls suppress supply&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Our current stance is &lt;strong&gt;Watchlist — incorporate upon policy confirmation&lt;/strong&gt;. The letter alone does not justify chasing memory equities. Until Commerce Department actions materialize, this is more appropriately used as a basis for reweighting risk scenarios than for revising EPS models upward.&lt;/p&gt;
&lt;h2 id="10-bear-case-and-invalidation-conditions"&gt;10. Bear Case and Invalidation Conditions
&lt;/h2&gt;&lt;h3 id="macro-failure-conditions"&gt;Macro Failure Conditions
&lt;/h3&gt;&lt;p&gt;If memory prices rise too quickly, smartphone, PC, and server customers may reduce shipments or trade down on specifications. A policy that protects suppliers could produce demand destruction, with the impact of declining unit volumes exceeding the benefit of higher ASPs. An AI investment slowdown combined with higher interest rates would not spare server memory.&lt;/p&gt;
&lt;h3 id="company-level-failure-conditions"&gt;Company-Level Failure Conditions
&lt;/h3&gt;&lt;p&gt;If Chinese memory concentrates in China and emerging markets rather than the United States, it could accelerate the erosion of Samsung, SK Hynix, and Micron&amp;rsquo;s China customer base. If domestic Chinese demand is large enough and the local equipment and materials ecosystem matures, the supply-suppression effect of the Entity List could prove shorter-lived than expected.&lt;/p&gt;
&lt;h3 id="policy-invalidation-conditions"&gt;Policy Invalidation Conditions
&lt;/h3&gt;&lt;p&gt;The thesis weakens materially if the following combination emerges:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;CXMT&amp;rsquo;s Entity List designation is delayed indefinitely.&lt;/li&gt;
&lt;li&gt;The United States grants broad exceptions for Apple&amp;rsquo;s China-market consumer products.&lt;/li&gt;
&lt;li&gt;The procurement ban remains limited to federal agencies and does not apply to private AI and data center operators.&lt;/li&gt;
&lt;li&gt;South Korea, Japan, and the EU do not coordinate.&lt;/li&gt;
&lt;li&gt;CXMT and YMTC transition rapidly to non-U.S. equipment, sustaining volume and yields.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Conversely, if CXMT&amp;rsquo;s designation, private-sector data center procurement restrictions, and allied coordination materialize together, this letter graduates from political signaling to a structural variable in the memory cycle.&lt;/p&gt;
&lt;h2 id="11-final-assessment"&gt;11. Final Assessment
&lt;/h2&gt;&lt;p&gt;The July 16 letter is an official document with specific demands. It is not, however, an enacted Chinese memory embargo. Near-term equity catalysts therefore require policy confirmation.&lt;/p&gt;
&lt;p&gt;The investment case is clear. If the United States and its allies choose not to fill memory shortfalls with Chinese supply, Samsung, SK Hynix, and Micron can defend market share and pricing leverage for longer. SanDisk and Kioxia also benefit directly from YMTC&amp;rsquo;s restricted path into NAND and enterprise SSD.&lt;/p&gt;
&lt;p&gt;By product, the impact is more direct for Samsung&amp;rsquo;s mobile DRAM, commodity DRAM, and NAND; for SK Hynix the more significant effect is an extended duration for its HBM competitive advantage. Micron, as the only large-scale U.S. memory maker, is the clearest and most straightforward policy beneficiary.&lt;/p&gt;
&lt;p&gt;In one sentence: &lt;strong&gt;This letter is not news that raises 2026 HBM shipments — it is a policy signal that reduces the risk of Chinese-origin oversupply eroding global memory profitability in 2027–2028.&lt;/strong&gt; Until concrete regulations emerge, it is more accurately read as a basis for recalibrating discount rates than as a catalyst to chase.&lt;/p&gt;
&lt;h2 id="evidence-classification"&gt;Evidence Classification
&lt;/h2&gt;&lt;h3 id="confirmed-facts"&gt;Confirmed Facts
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;On July 16, 2026, Representatives John Moolenaar and George Whitesides sent an official letter to the Secretary of Commerce.&lt;/li&gt;
&lt;li&gt;The letter calls for strengthened controls on YMTC, accelerated Entity List review for CXMT, specific U.S. procurement restrictions, and coordination with South Korea, Japan, and the EU.&lt;/li&gt;
&lt;li&gt;The Entity List restricts the export, re-export, and in-country transfer of items subject to U.S. Export Administration Regulations.&lt;/li&gt;
&lt;li&gt;TrendForce projected Q2 commodity DRAM contract prices up 58–63% and NAND up 70–75% quarter-over-quarter.&lt;/li&gt;
&lt;li&gt;TrendForce projected Q3 server DRAM up 13–18% and flagged potential supply tightness into 2027.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="analysis-and-inference"&gt;Analysis and Inference
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;If enacted, the greatest effect is a reduction in the 2027–2028 China oversupply risk and a compression of the normalized earnings discount rate.&lt;/li&gt;
&lt;li&gt;Samsung faces more direct exposure on mobile DRAM, commodity DRAM, and NAND; for SK Hynix the more important consideration is reduced long-term competitive risk in HBM.&lt;/li&gt;
&lt;li&gt;A bifurcated market structure — high prices in the U.S. and allied markets alongside domestic price competition within China — may emerge.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="unconfirmed-items"&gt;Unconfirmed Items
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Whether and when CXMT will be added to the Entity List&lt;/li&gt;
&lt;li&gt;The legal form, scope, and exceptions of any separate purchase prohibition&lt;/li&gt;
&lt;li&gt;Apple&amp;rsquo;s actual order volume from CXMT and YMTC and allocation by supplier&lt;/li&gt;
&lt;li&gt;Whether South Korea, Japan, and the EU will coordinate&lt;/li&gt;
&lt;li&gt;The specific EPS uplift for Samsung, SK Hynix, and Micron if restrictions are implemented&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="key-sources"&gt;Key Sources
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://chinaselectcommittee.house.gov/media/letters/moolenaar-whitesides-to-secretary-lutnick-hold-firm-on-chinese-memory-chips-ban" target="_blank" rel="noopener"
 &gt;House Select Committee on China: Official Moolenaar–Whitesides Letter&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.bis.gov/ear/title-15/subtitle-b/chapter-vii/subchapter-c/part-744/ss-74416-entity-list" target="_blank" rel="noopener"
 &gt;U.S. Commerce Department BIS: Entity List Regulations&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.trendforce.com/presscenter/news/20260331-12995.html" target="_blank" rel="noopener"
 &gt;TrendForce: Q2 2026 DRAM and NAND Price Outlook&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.trendforce.com/presscenter/news/20260709-13140.html" target="_blank" rel="noopener"
 &gt;TrendForce: Q3 2026 Server DRAM Outlook&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;em&gt;This article is provided for informational purposes based on publicly available sources and does not constitute a recommendation to buy or sell any specific security.&lt;/em&gt;&lt;/p&gt;</description></item><item><title>Will Apple Really Use CXMT? China-Only Memory Testing and the Pricing Risk for Korean DRAM Makers</title><link>https://koreainvestinsights.com/post/apple-cxmt-dram-china-test-memory-pricing-korea-2026-07-14/</link><pubDate>Tue, 14 Jul 2026 22:10:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/apple-cxmt-dram-china-test-memory-pricing-korea-2026-07-14/</guid><description>&lt;p&gt;Apple is reportedly testing DRAM made by ChangXin Memory Technologies, or CXMT, for devices sold in China. No supply contract has been announced. Those two facts must be read together: CXMT has entered Apple&amp;rsquo;s qualification process, but it has not become the main memory supplier for global iPhones or Macs.&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Context: This is a follow-up to &lt;a class="link" href="https://koreainvestinsights.com/post/cxmt-ipo-memory-price-risk-hbm-client-dram-2026-06-21/" &gt;CXMT&amp;rsquo;s IPO and memory price risk&lt;/a&gt;. That report argued that CXMT capacity is more likely to cap client DRAM and LPDDR pricing before it disrupts HBM. This article examines how a buyer as large as Apple could use that emerging supply. Related collections are the &lt;a class="link" href="https://koreainvestinsights.com/page/korea-semiconductor-hbm-kospi-hub/" &gt;AI HBM Hub&lt;/a&gt; and the &lt;a class="link" href="https://koreainvestinsights.com/page/exclusive-analysis-hub/" &gt;Exclusive Analysis 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;Apple could eventually qualify CXMT as a commercial supplier. The most plausible first step is limited volume in standard devices sold in China. A rapid move into global flagship products remains unlikely.&lt;/li&gt;
&lt;li&gt;CXMT&amp;rsquo;s public LPDDR5X specifications are sufficient to enter qualification. Its 8,533Mbps and 9,600Mbps products are in mass production, while the 10,667Mbps version remains in customer sampling. Apple-grade power, thermals, package thickness, defect rates, long-term reliability, and high-volume yield remain undisclosed.&lt;/li&gt;
&lt;li&gt;A reported 5% to 10% average selling-price discount does not settle the economics. That estimate covers CXMT&amp;rsquo;s mixed DRAM portfolio, not an Apple-specific LPDDR5X quote. Dual bills of materials, inventory separation, qualification, and regulatory replacement risk reduce the net savings.&lt;/li&gt;
&lt;li&gt;Negotiating leverage may matter more than direct units. A fourth supplier capable of filling a commercial order could pressure Samsung, SK Hynix, and Micron across Apple&amp;rsquo;s much larger DRAM procurement book.&lt;/li&gt;
&lt;li&gt;CXMT&amp;rsquo;s near-term competitive pressure is concentrated in mobile and client DRAM, not leading-edge HBM. Treating the report as a blanket negative for every memory stock ignores product mix.&lt;/li&gt;
&lt;li&gt;Event-only ratings are Watchlist for SK Hynix, Wait for Samsung Electronics, and Wait for Apple. The evidence currently supports testing and lobbying, not a disclosed contract, product, volume, or duration.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="1-what-is-confirmed-and-what-is-not"&gt;1. What Is Confirmed and What Is Not
&lt;/h2&gt;&lt;h3 id="11-testing-is-reported-a-contract-is-not"&gt;1.1 Testing is reported; a contract is not
&lt;/h3&gt;&lt;p&gt;The Financial Times reported that Apple is testing CXMT DRAM for devices sold in China. Earlier reporting said Apple had sought regulatory clarity from the U.S. government. Apple, the White House, and CXMT did not respond to Reuters requests for comment at the time.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Question&lt;/th&gt;
 &lt;th&gt;Assessment as of July 14, 2026&lt;/th&gt;
 &lt;th&gt;Evidence level&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Is Apple testing CXMT DRAM?&lt;/td&gt;
 &lt;td&gt;Reported for China-market devices&lt;/td&gt;
 &lt;td&gt;Credible secondary reporting&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Has a supply contract been signed?&lt;/td&gt;
 &lt;td&gt;Not confirmed&lt;/td&gt;
 &lt;td&gt;Blocked&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Which product is involved?&lt;/td&gt;
 &lt;td&gt;iPhone, iPad, or Mac cannot be identified&lt;/td&gt;
 &lt;td&gt;Blocked&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Will the chips be used outside China?&lt;/td&gt;
 &lt;td&gt;No supporting evidence&lt;/td&gt;
 &lt;td&gt;Do not infer&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Has the U.S. government approved use?&lt;/td&gt;
 &lt;td&gt;Regulatory certainty was reportedly requested&lt;/td&gt;
 &lt;td&gt;No official decision&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Have incumbent allocations been reduced?&lt;/td&gt;
 &lt;td&gt;Not confirmed&lt;/td&gt;
 &lt;td&gt;Blocked&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The accurate wording is that Apple is qualifying CXMT as a potential supplier, not that Apple has already adopted it. &lt;a class="link" href="https://www.marketscreener.com/news/apple-seeks-approval-to-buy-chips-from-blacklisted-chinese-company-ft-reports-ce7f5fd9d08df72d" target="_blank" rel="noopener"
 &gt;Reuters coverage&lt;/a&gt;, &lt;a class="link" href="https://www.macrumors.com/2026/07/08/apple-begins-testing-controversial-chinese-memory/" target="_blank" rel="noopener"
 &gt;testing report summary&lt;/a&gt;&lt;/p&gt;
&lt;h3 id="12-cxmt-is-no-longer-an-experimental-producer"&gt;1.2 CXMT is no longer an experimental producer
&lt;/h3&gt;&lt;p&gt;Counterpoint Research estimated CXMT&amp;rsquo;s global DRAM revenue share at roughly 8% in 1Q26, placing it fourth behind Samsung, SK Hynix, and Micron. Market-share figures differ depending on whether the denominator is revenue, bit shipments, or wafer input, but the direction is clear: CXMT is no longer a minor laboratory-scale supplier. &lt;a class="link" href="https://counterpointresearch.com/en/insights/global-dram-revenue-surges-to-near-dollar-100-billion-mark-in-q1-2026" target="_blank" rel="noopener"
 &gt;Counterpoint Research&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;Scale does not prove Apple-grade maturity. Supplying commodity DRAM to large Chinese customers does not establish the power, reliability, and yield required in Apple&amp;rsquo;s premium devices.&lt;/p&gt;
&lt;h2 id="2-performance-qualification-is-harder-than-reaching-a-headline-speed"&gt;2. Performance: Qualification Is Harder Than Reaching a Headline Speed
&lt;/h2&gt;&lt;p&gt;CXMT says its LPDDR5X portfolio includes 12Gb and 16Gb dies, 12GB, 16GB, and 24GB packages, and data rates up to 10,667Mbps. Its 8,533Mbps and 9,600Mbps products entered mass production in May 2025; the 10,667Mbps version remains in customer sampling. &lt;a class="link" href="https://www.cxmt.com/en/news/info_19.html" target="_blank" rel="noopener"
 &gt;CXMT announcement&lt;/a&gt;&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Confirmed status&lt;/th&gt;
 &lt;th&gt;Investment interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Die density&lt;/td&gt;
 &lt;td&gt;12Gb and 16Gb&lt;/td&gt;
 &lt;td&gt;Meets an entry-level mobile and PC requirement&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Package capacity&lt;/td&gt;
 &lt;td&gt;12GB, 16GB, and 24GB&lt;/td&gt;
 &lt;td&gt;Supports standard device configurations&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Data rate&lt;/td&gt;
 &lt;td&gt;8,533, 9,600, and 10,667Mbps&lt;/td&gt;
 &lt;td&gt;The highest-speed part is still sampling&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Power&lt;/td&gt;
 &lt;td&gt;CXMT claims 30% savings versus its LPDDR5&lt;/td&gt;
 &lt;td&gt;Not a like-for-like comparison with the big three&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Apple qualification&lt;/td&gt;
 &lt;td&gt;Undisclosed&lt;/td&gt;
 &lt;td&gt;Commercial adoption cannot be established&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;High-volume yield&lt;/td&gt;
 &lt;td&gt;Undisclosed&lt;/td&gt;
 &lt;td&gt;Tens-of-millions-unit supply cannot be established&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Public specifications refute the claim that Chinese DRAM is automatically too slow for Apple. The real bottlenecks are standby and refresh power, thermal stability, package thickness, defect rates, lot-to-lot variation, long-term reliability, high-volume yield, and integration with Apple&amp;rsquo;s processors and power-management firmware.&lt;/p&gt;
&lt;p&gt;CXMT has earned a seat in qualification. Public evidence does not show that it can maintain Apple-grade quality at scale.&lt;/p&gt;
&lt;h2 id="3-price-a-5-to-10-discount-does-not-settle-the-economics"&gt;3. Price: A 5% to 10% Discount Does Not Settle the Economics
&lt;/h2&gt;&lt;p&gt;SemiAnalysis estimated that CXMT&amp;rsquo;s blended DRAM average selling price in 1Q26 was roughly 5% to 10% below the incumbent trio, while its DDR5 cost per bit could be more than 30% higher. These are company-wide mixed-product estimates, not Apple-specific LPDDR5X bids. &lt;a class="link" href="https://newsletter.semianalysis.com/p/chinas-cxmt-is-set-to-challenge-dram" target="_blank" rel="noopener"
 &gt;SemiAnalysis&lt;/a&gt;&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Additional cost&lt;/th&gt;
 &lt;th&gt;Burden on Apple&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;China and global supply-chain separation&lt;/td&gt;
 &lt;td&gt;Separate part numbers, assembly planning, and logistics&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Dual inventory&lt;/td&gt;
 &lt;td&gt;Regional demand mismatch and slower turns&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Qualification engineering&lt;/td&gt;
 &lt;td&gt;Repeated processor, firmware, and thermal validation&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Regulatory replacement risk&lt;/td&gt;
 &lt;td&gt;Emergency supplier switch and requalification&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Quality failure&lt;/td&gt;
 &lt;td&gt;Warranty, recall, and brand damage&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;A nominal discount shrinks after these costs. CXMT&amp;rsquo;s pricing also should not be read as proof of structural cost leadership. State support, lower margin requirements, product mix, and domestic-market strategy may contribute.&lt;/p&gt;
&lt;h2 id="4-negotiating-leverage-may-matter-more-than-units"&gt;4. Negotiating Leverage May Matter More Than Units
&lt;/h2&gt;&lt;p&gt;Apple&amp;rsquo;s economic benefit is not limited to the discount on chips purchased from CXMT. Qualifying a fourth supplier capable of filling a real commercial order could influence the price of Apple&amp;rsquo;s entire DRAM procurement book.&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;Maintain CXMT qualification
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ enable a small production order
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ demonstrate credible switching capacity
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ limit contract-price increases from the incumbent trio
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;This effect can appear before CXMT gains visible market share. A small allocation removes the incumbents&amp;rsquo; assumption that Apple has no alternative.&lt;/p&gt;
&lt;p&gt;Samples alone are not enough. The option becomes credible only if CXMT can deliver a minimum commercial volume into a shipping product. Without a contract and a willingness to switch, the incumbents have little reason to change their pricing.&lt;/p&gt;
&lt;h2 id="5-regulation-legal-purchasability-and-multi-year-supply-security-are-different"&gt;5. Regulation: Legal Purchasability and Multi-Year Supply Security Are Different
&lt;/h2&gt;&lt;h3 id="51-a-section-1260h-designation-does-not-automatically-ban-ordinary-commercial-purchases"&gt;5.1 A Section 1260H designation does not automatically ban ordinary commercial purchases
&lt;/h3&gt;&lt;p&gt;The U.S. Department of Defense included CXMT on its June 8, 2026 Section 1260H list of Chinese military companies. The designation creates government-contracting and national-security risk, but it is not a blanket prohibition on every private commercial transaction. &lt;a class="link" href="https://www.war.gov/News/Releases/Release/Article/4511232/dow-releases-list-of-chinese-military-companies-in-accordance-with-section-1260/" target="_blank" rel="noopener"
 &gt;U.S. Department of Defense release&lt;/a&gt;&lt;/p&gt;
&lt;h3 id="52-entity-list-risk-matters-most-through-production-continuity"&gt;5.2 Entity List risk matters most through production continuity
&lt;/h3&gt;&lt;p&gt;Reuters reported that an interagency committee had approved CXMT for Entity List addition, but publication remained on hold as of mid-June 2026. An Entity List designation would impose severe licensing restrictions on exports, reexports, or transfers of U.S. equipment, software, technology, and support to CXMT. Buying a finished DRAM chip is legally distinct from exporting U.S. technology to the producer.&lt;/p&gt;
&lt;p&gt;The commercial consequence is still large. Restricted spare parts, software updates, and engineering support could impair yield, capacity, and process migration. Apple needs multi-year supply continuity, not merely the ability to execute one purchase today. &lt;a class="link" href="https://www.investing.com/news/stock-market-news/us-holds-off-on-adding-deepseek-cxmt-to-trade-blacklist-reuters-reports-4746250" target="_blank" rel="noopener"
 &gt;Reuters report on the delay&lt;/a&gt;&lt;/p&gt;
&lt;h3 id="53-the-2027-federal-procurement-restriction-is-not-a-consumer-ban"&gt;5.3 The 2027 federal procurement restriction is not a consumer ban
&lt;/h3&gt;&lt;p&gt;A proposed Federal Acquisition Regulation would implement restrictions from December 23, 2027 on federal-agency procurement of electronic products and services containing covered chips from CXMT, YMTC, and SMIC. It does not ban ordinary consumers from buying iPhones. It could, however, force Apple to maintain separate non-CXMT configurations for government and some enterprise channels. The proposal includes an exception through December 23, 2028 for commercial products without alternatives. &lt;a class="link" href="https://public-inspection.federalregister.gov/2026-03065.pdf" target="_blank" rel="noopener"
 &gt;Proposed FAR rule&lt;/a&gt;&lt;/p&gt;
&lt;h3 id="54-ymtc-is-the-relevant-political-precedent"&gt;5.4 YMTC is the relevant political precedent
&lt;/h3&gt;&lt;p&gt;Apple explored using YMTC NAND in 2022 but shelved the plan after congressional opposition and tighter restrictions. CXMT is not in the same regulatory position today, but the episode shows that political pressure can interrupt a technically plausible sourcing plan.&lt;/p&gt;
&lt;h2 id="6-the-most-plausible-path-limited-use-in-standard-devices-sold-in-china"&gt;6. The Most Plausible Path: Limited Use in Standard Devices Sold in China
&lt;/h2&gt;&lt;p&gt;China-only standard models offer a compromise among engineering, economics, and politics.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Variable&lt;/th&gt;
 &lt;th&gt;Limited China-market adoption&lt;/th&gt;
 &lt;th&gt;Global primary adoption&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Qualification&lt;/td&gt;
 &lt;td&gt;Standard products can be tested first&lt;/td&gt;
 &lt;td&gt;Long validation across premium products&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Regulatory exposure&lt;/td&gt;
 &lt;td&gt;Can be separated from U.S. government channels&lt;/td&gt;
 &lt;td&gt;Conflicts across government and enterprise sales&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Supply-chain operations&lt;/td&gt;
 &lt;td&gt;Separate China bill of materials&lt;/td&gt;
 &lt;td&gt;Larger loss of global commonality&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Volume&lt;/td&gt;
 &lt;td&gt;Limited&lt;/td&gt;
 &lt;td&gt;Tens of millions of devices&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Political burden&lt;/td&gt;
 &lt;td&gt;Potentially manageable&lt;/td&gt;
 &lt;td&gt;Much greater U.S. opposition and sanction risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Current probability&lt;/td&gt;
 &lt;td&gt;Conditional&lt;/td&gt;
 &lt;td&gt;Low&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Four conditions must hold: Apple reliability qualification, sufficient Chinese yield and volume, regulatory continuity over the contract, and net benefits larger than the cost of dual sourcing.&lt;/p&gt;
&lt;h2 id="7-capital-flow-and-choke-points"&gt;7. Capital Flow and Choke Points
&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-text" data-lang="text"&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;AI data-center investment rises
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ HBM and server DRAM receive priority
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ mobile and PC DRAM supply becomes less elastic
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Apple&amp;#39;s memory bill rises
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ the incentive to qualify CXMT increases
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;The chain runs from equipment and materials through CXMT wafer production, LPDDR5X packaging and testing, Apple qualification, device assembly, and China distribution. The four choke points are Apple-grade quality and yield, regulatory durability, the cost of separate regional configurations, and CXMT&amp;rsquo;s access to U.S. equipment and technical support.&lt;/p&gt;
&lt;h2 id="8-memory-sector-impact-separate-hbm-from-mobile-dram"&gt;8. Memory-Sector Impact: Separate HBM from Mobile DRAM
&lt;/h2&gt;&lt;p&gt;CXMT&amp;rsquo;s current competitive position is closer to DDR5 and LPDDR than to leading-edge HBM. Extrapolating Apple&amp;rsquo;s test to a collapse in the entire DRAM profit pool ignores product economics.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Company&lt;/th&gt;
 &lt;th&gt;Direct CXMT exposure&lt;/th&gt;
 &lt;th&gt;More important variable&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;Potential LPDDR pricing and share pressure at Apple&lt;/td&gt;
 &lt;td&gt;HBM qualification, commodity mix, foundry losses&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td&gt;Some mobile exposure, but a larger share of the core thesis comes from HBM&lt;/td&gt;
 &lt;td&gt;HBM4 yield, share, ASP, and AI capex&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Micron&lt;/td&gt;
 &lt;td&gt;Potential mobile and PC DRAM pressure&lt;/td&gt;
 &lt;td&gt;Server DRAM, HBM ramp, long-term agreements&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Apple&lt;/td&gt;
 &lt;td&gt;Lower procurement cost and supplier diversification&lt;/td&gt;
 &lt;td&gt;Dual-BOM cost, regulation, total gross margin&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CXMT&lt;/td&gt;
 &lt;td&gt;Apple qualification would improve global credibility&lt;/td&gt;
 &lt;td&gt;Yield, cost, equipment access, regulatory security&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;More CXMT mobile capacity could encourage the incumbent trio to allocate still more capital to HBM and server DRAM. That would pressure commodity pricing without eliminating HBM&amp;rsquo;s qualification, stacking-yield, packaging, and customer-roadmap barriers.&lt;/p&gt;
&lt;p&gt;The long-term risk is real. If CXMT rapidly improves advanced DDR5 and LPDDR5X yield and absorbs export-control shocks through a domestic equipment ecosystem, the incumbents&amp;rsquo; commodity cash flow could weaken. If it later achieves commercial HBM yield, today&amp;rsquo;s product boundary would erode. Public evidence does not show that outcome yet.&lt;/p&gt;
&lt;h2 id="9-three-investment-frames-and-their-red-teams"&gt;9. Three Investment Frames and Their Red Teams
&lt;/h2&gt;&lt;h3 id="frame-1-limited-china-market-adoption"&gt;Frame 1: Limited China-market adoption
&lt;/h3&gt;&lt;p&gt;Classification: idiosyncratic event alpha&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Price: a lower CXMT supply price&lt;/li&gt;
&lt;li&gt;Quantity: selected devices sold in China&lt;/li&gt;
&lt;li&gt;Cost: qualification, dual inventory, regulation, and warranty risk&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Beneficiaries include Apple&amp;rsquo;s procurement organization, CXMT, and China&amp;rsquo;s packaging and materials chain. Incumbent Apple mobile-DRAM pricing would face pressure.&lt;/p&gt;
&lt;p&gt;Red team: A durable U.S.-China agreement, regulatory safe harbor, and premium-device qualification could break the China-only assumption. A failed reliability test would invalidate even limited adoption.&lt;/p&gt;
&lt;h3 id="frame-2-negotiating-impact-appears-before-market-share-impact"&gt;Frame 2: Negotiating impact appears before market-share impact
&lt;/h3&gt;&lt;p&gt;Classification: idiosyncratic event alpha&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Price: Apple contract prices with the incumbent trio&lt;/li&gt;
&lt;li&gt;Quantity: Apple&amp;rsquo;s full DRAM procurement book&lt;/li&gt;
&lt;li&gt;Cost: maintaining CXMT qualification&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Investors can observe supplier share, but the marginal supplier may influence contract pricing long before it wins large share.&lt;/p&gt;
&lt;p&gt;Red team: A memory surplus would force incumbents to cut prices without CXMT. A failed Apple qualification would also make the option non-credible.&lt;/p&gt;
&lt;h3 id="frame-3-cxmt-risk-is-concentrated-in-mobile-and-client-dram"&gt;Frame 3: CXMT risk is concentrated in mobile and client DRAM
&lt;/h3&gt;&lt;p&gt;Classification: product-mix trade&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Price: the ceiling on mobile and PC DRAM pricing&lt;/li&gt;
&lt;li&gt;Quantity: CXMT capacity and domestic shipments&lt;/li&gt;
&lt;li&gt;Cost: advanced-node yield and U.S. equipment restrictions&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Mobile DRAM buyers and HBM-heavy suppliers are relative beneficiaries. Suppliers dependent on Apple mobile DRAM or commodity price increases face more risk.&lt;/p&gt;
&lt;p&gt;Red team: Rapid progress in advanced-node yield and commercial HBM would weaken the product separation. A collapse in AI data-center investment would also remove HBM&amp;rsquo;s relative defense.&lt;/p&gt;
&lt;h2 id="10-stock-specific-monitoring-gates"&gt;10. Stock-Specific Monitoring Gates
&lt;/h2&gt;&lt;p&gt;These are conditional event assessments, not stand-alone buy recommendations.&lt;/p&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 style="text-align: right"&gt;Displayed P/E on July 14, 2026&lt;/th&gt;
 &lt;th&gt;Event interpretation&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;Watchlist&lt;/td&gt;
 &lt;td style="text-align: right"&gt;16.97x&lt;/td&gt;
 &lt;td&gt;HBM4 and AI investment matter more than direct CXMT risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td&gt;Wait&lt;/td&gt;
 &lt;td style="text-align: right"&gt;20.99x&lt;/td&gt;
 &lt;td&gt;Mobile pricing pressure must be weighed against HBM recovery&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Apple&lt;/td&gt;
 &lt;td&gt;Wait&lt;/td&gt;
 &lt;td style="text-align: right"&gt;38.38x&lt;/td&gt;
 &lt;td&gt;iPhone, services, and total gross margin matter more than unproven CXMT savings&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;P/E figures are displayed values from &lt;a class="link" href="https://www.google.com/finance/quote/005930:KRX" target="_blank" rel="noopener"
 &gt;Google Finance for Samsung&lt;/a&gt;, &lt;a class="link" href="https://www.google.com/finance/quote/000660:KRX" target="_blank" rel="noopener"
 &gt;SK Hynix&lt;/a&gt;, and &lt;a class="link" href="https://www.google.com/finance/quote/AAPL:NASDAQ" target="_blank" rel="noopener"
 &gt;Apple&lt;/a&gt;. Trailing P/E can overstate cheapness for cyclical memory companies near a profit peak.&lt;/p&gt;
&lt;h3 id="sk-hynix"&gt;SK Hynix
&lt;/h3&gt;&lt;p&gt;Confirmation gates: HBM4 qualification and production timing remain intact, HBM yield and ASP estimates do not fall, and CXMT-related share-price weakness occurs without an HBM earnings downgrade.&lt;/p&gt;
&lt;p&gt;Invalidation: HBM4 qualification delay, customer diversification that reduces share, or CXMT achieving commercial yield in advanced server DRAM or HBM.&lt;/p&gt;
&lt;h3 id="samsung-electronics"&gt;Samsung Electronics
&lt;/h3&gt;&lt;p&gt;Confirmation gates: primary-supplier status in the next Apple generation, CXMT use limited to China-market volume, and visible HBM yield and customer-qualification improvement.&lt;/p&gt;
&lt;p&gt;Invalidation: a material multi-year China-device allocation to CXMT, simultaneous loss of LPDDR share and pricing, or an HBM recovery too slow to offset mobile pressure.&lt;/p&gt;
&lt;h3 id="apple"&gt;Apple
&lt;/h3&gt;&lt;p&gt;Confirmation gates: gross-margin guidance holds despite memory inflation, regulation provides durable clarity, and iPhone and services estimates rise independently of supplier diversification.&lt;/p&gt;
&lt;p&gt;Invalidation: regulation removes the CXMT option, memory inflation drives both price hikes and demand weakness, or slower iPhone and services growth fails to support the valuation.&lt;/p&gt;
&lt;h2 id="11-evidence-that-would-change-the-thesis"&gt;11. Evidence That Would Change the Thesis
&lt;/h2&gt;&lt;p&gt;Upside evidence includes an official Apple supplier confirmation, teardown proof of CXMT DRAM, qualification beyond China-market products, a durable U.S. safe harbor, or high-volume yield for CXMT&amp;rsquo;s 10,667Mbps and next-generation products.&lt;/p&gt;
&lt;p&gt;Downside evidence includes failed Apple reliability testing, an Entity List action that interrupts equipment support, the qualification ending without a contract, dual-sourcing costs exceeding savings, or a DRAM surplus that removes Apple&amp;rsquo;s need for a fourth supplier.&lt;/p&gt;
&lt;h2 id="12-evidence-classification"&gt;12. Evidence Classification
&lt;/h2&gt;&lt;h3 id="fact"&gt;Fact
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Apple is reportedly testing CXMT DRAM for China-market devices.&lt;/li&gt;
&lt;li&gt;Apple reportedly sought U.S. regulatory clarity regarding CXMT.&lt;/li&gt;
&lt;li&gt;CXMT mass-produces 8,533Mbps and 9,600Mbps LPDDR5X and samples a 10,667Mbps product.&lt;/li&gt;
&lt;li&gt;Counterpoint estimated CXMT&amp;rsquo;s 1Q26 global DRAM revenue share at about 8%.&lt;/li&gt;
&lt;li&gt;CXMT appears on the June 2026 Section 1260H list.&lt;/li&gt;
&lt;li&gt;Section 1260H does not by itself automatically prohibit ordinary private purchases.&lt;/li&gt;
&lt;li&gt;CXMT had not been formally added to the Entity List as of mid-June 2026.&lt;/li&gt;
&lt;li&gt;The proposed FAR rule implements a December 23, 2027 federal procurement restriction for covered semiconductors.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="inference"&gt;Inference
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Initial commercial use is most likely limited to standard devices sold in China.&lt;/li&gt;
&lt;li&gt;Apple&amp;rsquo;s largest economic benefit may be incumbent negotiation leverage rather than direct chip discounts.&lt;/li&gt;
&lt;li&gt;CXMT&amp;rsquo;s near-term threat is concentrated in mobile and client DRAM rather than HBM.&lt;/li&gt;
&lt;li&gt;Separate China and global configurations are more plausible than a full global switch.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="speculation"&gt;Speculation
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Whether the test involves iPhone, iPad, or Mac&lt;/li&gt;
&lt;li&gt;Commercial launch timing&lt;/li&gt;
&lt;li&gt;Initial volume and contract duration&lt;/li&gt;
&lt;li&gt;Whether Washington will provide a durable safe harbor&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="blocked"&gt;Blocked
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Apple&amp;rsquo;s internal qualification results&lt;/li&gt;
&lt;li&gt;Apple-specific pricing and the discount versus incumbents&lt;/li&gt;
&lt;li&gt;Apple-standard power, package thickness, defect rates, and reliability&lt;/li&gt;
&lt;li&gt;Apple-volume yield and monthly CXMT capacity&lt;/li&gt;
&lt;li&gt;Contract duration and minimum volume&lt;/li&gt;
&lt;li&gt;Exact product-level incumbent share and pricing at Apple&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="conclusion"&gt;Conclusion
&lt;/h2&gt;&lt;p&gt;Apple can use CXMT, but the only reported path today is testing for devices sold in China. Public specifications justify qualification, while Apple-grade quality, high-volume yield, and multi-year regulatory security remain unproven.&lt;/p&gt;
&lt;p&gt;The first market variable to watch is not CXMT&amp;rsquo;s Apple share. It is whether a commercially credible fourth supplier changes mobile-DRAM contract pricing for the incumbent trio. That pressure is more likely to appear in LPDDR and client DRAM before HBM.&lt;/p&gt;
&lt;p&gt;There is no reason to treat every Korean memory stock identically. Samsung must balance mobile pricing pressure against HBM recovery. For SK Hynix, HBM4 yield and the durability of AI investment remain more important than CXMT. The next decision should follow official supplier confirmation, product teardowns, an Entity List decision, and disclosed commercial volume.&lt;/p&gt;</description></item><item><title>What Remains from 2,893 Messages: Two Weeks of Memory Pricing, AI Infrastructure Bottlenecks, and Leveraged Flow</title><link>https://koreainvestinsights.com/post/telegram-two-week-market-deep-dive-memory-ai-infra-leverage-2026-07-14/</link><pubDate>Tue, 14 Jul 2026 18:15:55 +0900</pubDate><guid>https://koreainvestinsights.com/post/telegram-two-week-market-deep-dive-memory-ai-infra-leverage-2026-07-14/</guid><description>&lt;p&gt;Between July 1 and July 14, 2026, 2,893 messages were collected from 18 Telegram channels. Of these, 1,890 were classified as high-information-density, and 308 were original source documents after stripping forwards and reposts. The most frequently recurring topics were: semiconductors and AI infrastructure (352 mentions), interest rates, currencies, and macro (272), China, policy, and demand (77), biotech (74), and power, nuclear, and energy (69).&lt;/p&gt;
&lt;p&gt;These numbers should not be read directly as a measure of market conviction. When the same article or brokerage memo circulates across multiple channels, the message count rises without adding a single new fact. Conversely, a single official disclosure may not be widely shared yet can carry far greater impact on company valuations. Telegram is a &lt;strong&gt;map of attention&lt;/strong&gt; — it shows what the market is watching. It is not a ledger of confirmed facts.&lt;/p&gt;
&lt;p&gt;This post re-selects from those 308 source documents the signals most likely to alter market structure. Nanya Technology&amp;rsquo;s quarterly results, monthly revenue from TSMC, Wiwynn, Delta, CCL makers, and copper foil producers, Meta&amp;rsquo;s 5 GW data center plan, Brookfield and Bloom Energy&amp;rsquo;s $25 billion power financing, Fermi&amp;rsquo;s convertible notes, Micron&amp;rsquo;s U.S. capacity expansion, Korea&amp;rsquo;s single-stock leveraged products, and SK Hynix ADR discussions were read in cross-reference against one another.&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Connected context: This post is a follow-up that re-links &lt;a class="link" href="https://koreainvestinsights.com/ko/post/h1-2026-ai-infra-bottleneck-korea-narrow-market-postmortem-2026-06-30/" &gt;the H1 2026 AI infrastructure and narrow market postmortem&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/ko/post/micron-fy3q26-ai-memory-sca-fcf-hold-2026-06-25/" &gt;the Micron FY3Q26 earnings analysis&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/ko/post/kospi-9pct-selloff-ai-hardware-derating-korea-leverage-2026-07-13/" &gt;the KOSPI sell-off and leverage amplification analysis&lt;/a&gt;, and &lt;a class="link" href="https://koreainvestinsights.com/ko/post/ai-token-value-memory-value-added-2026-07-09/" &gt;the AI token and memory value-added analysis&lt;/a&gt; through the flow of source documents from the first half of July.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;ol&gt;
&lt;li&gt;Memory earnings are still being driven by price, not volume. Nanya Technology&amp;rsquo;s Q2 DRAM average selling price rose more than 60% quarter-on-quarter while shipment volume barely changed. Current excess profits are powerful, but earnings sensitivity to any price reversal is equally high.&lt;/li&gt;
&lt;li&gt;AI investment has spread beyond GPUs and HBM to server ODMs, power supply units, liquid cooling, advanced copper foil, and CCL. Taiwan&amp;rsquo;s supply chain June revenue confirms this shift.&lt;/li&gt;
&lt;li&gt;The next bottleneck is power, and after power comes financing. Meta&amp;rsquo;s 5 GW plan and Brookfield&amp;rsquo;s $25 billion financing commitment show the scale of demand. Fermi&amp;rsquo;s $375 million convertible notes show that the same investments carry dilution and refinancing risk.&lt;/li&gt;
&lt;li&gt;Supply shortage is sowing the seeds of future supply. From 2028, when Nanya&amp;rsquo;s new fab, Micron&amp;rsquo;s U.S. investments, and the three major memory makers&amp;rsquo; capacity expansions converge, demand qualification cycles and depreciation will matter more than demand growth alone.&lt;/li&gt;
&lt;li&gt;In the Korean market, fundamentals and trading structure must be separated. Single-stock leveraged products can amplify both gains and losses, and ADR premiums may be the price of arbitrage friction rather than an increase in enterprise value.&lt;/li&gt;
&lt;li&gt;The claim that more agents means infinite memory demand remains a narrative. Total inference volume and context storage are growing, but compression, cache reuse, tiered storage, and hardware-software co-design are lowering cost per byte.&lt;/li&gt;
&lt;li&gt;The most important final metric is not AI capex. It is &lt;strong&gt;how much the end customer pays per unit of useful work completed, and how that money is divided among power, server, networking, and memory suppliers&lt;/strong&gt;.&lt;/li&gt;
&lt;/ol&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;
 The messages from the first half of July read less as a signal that AI investment is over and more as a signal that &lt;strong&gt;the bottleneck capturing returns is shifting from chips to memory, networking, power, and financing&lt;/strong&gt;. That said, price-driven earnings, large-scale borrowing, and leveraged positioning have all grown simultaneously, widening the gap between a good industry and a good entry price.
 &lt;/div&gt;
&lt;/div&gt;
&lt;p&gt;&lt;img alt="Telegram market signal map for the first half of July 2026" loading="lazy" sizes="(max-width: 767px) calc(100vw - 30px), (max-width: 1023px) 700px, (max-width: 1279px) 950px, 1232px" src="https://koreainvestinsights.com/images/posts/telegram-two-week-market-grid-2026-07-14.png"&gt;
&lt;/p&gt;
&lt;h2 id="1-how-to-read-two-weeks-evidence-quality-over-message-count"&gt;1. How to Read Two Weeks: Evidence Quality Over Message Count
&lt;/h2&gt;&lt;h3 id="11-2893-messages-are-not-2893-events"&gt;1.1 2,893 Messages Are Not 2,893 Events
&lt;/h3&gt;&lt;p&gt;The most frequently captured topic over the two weeks was semiconductors and AI infrastructure. But the number 352 does not mean 352 independent positive catalysts. When a single brokerage report passes through an original channel, a repost channel, and a summary channel, it gets counted three or more times. The same article with only a changed headline also repeats.&lt;/p&gt;
&lt;p&gt;This analysis therefore divided messages into four tiers.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Tier&lt;/th&gt;
 &lt;th&gt;Basis&lt;/th&gt;
 &lt;th&gt;How to Use&lt;/th&gt;
 &lt;th&gt;Examples&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;A&lt;/td&gt;
 &lt;td&gt;Company IR, regulatory filings, official announcements&lt;/td&gt;
 &lt;td&gt;Factual reference point&lt;/td&gt;
 &lt;td&gt;Nanya results, TSMC monthly revenue, Meta capex, SEC filings&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;B&lt;/td&gt;
 &lt;td&gt;Brokerage and industry data with figures and sources&lt;/td&gt;
 &lt;td&gt;Cross-checked against primary materials&lt;/td&gt;
 &lt;td&gt;Taiwan component IR summaries, industry supply/demand forecasts&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;C&lt;/td&gt;
 &lt;td&gt;Anonymous sourcing, Telegram interpretation, market memos&lt;/td&gt;
 &lt;td&gt;Signal of possibility and sentiment&lt;/td&gt;
 &lt;td&gt;Meta surplus compute sales, Samsung HBM yield, ADR premium&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;D&lt;/td&gt;
 &lt;td&gt;Long-term narratives and extreme forecasts&lt;/td&gt;
 &lt;td&gt;Material for forming questions&lt;/td&gt;
 &lt;td&gt;10 billion agents, infinite memory demand growth&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Even if the same claim appears across multiple channels, if the original source is one, the evidence is one. Conversely, a single Tier A document can matter more than ten Tier C posts. Collection on July 14 may have partial gaps due to Telegram web domain connection errors, so that final day&amp;rsquo;s frequency is best excluded from comparisons.&lt;/p&gt;
&lt;h3 id="12-repetition-reveals-positioning-not-truth"&gt;1.2 Repetition Reveals Positioning, Not Truth
&lt;/h3&gt;&lt;p&gt;Repetition itself has its uses. What it reveals, however, is market participants&amp;rsquo; attention and positioning, not the probability that something is true.&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;When multiple channels share the same HBM price increase, it means the memory bull thesis is already widely held.&lt;/li&gt;
&lt;li&gt;When posts about leveraged ETF side effects repeat, it means supply/demand anxiety has become investors&amp;rsquo; primary explanatory frame, independent of actual product scale.&lt;/li&gt;
&lt;li&gt;When power and cooling articles multiply, it means the market is searching for bottlenecks beyond GPUs.&lt;/li&gt;
&lt;li&gt;When ADR premium posts repeat, it means U.S.-listing structure has become a short-term price catalyst, overshadowing domestic earnings fundamentals.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The correct sequence for using Telegram in investment is therefore: &lt;code&gt;detect repetition → verify source document → cross-check against official data → connect to earnings and cash flow → check price reflection&lt;/code&gt;. Jumping from repetition detection directly to buying means purchasing the most crowded narrative at the latest possible entry.&lt;/p&gt;
&lt;h2 id="2-mapping-two-weeks-of-markets-onto-a-single-grid"&gt;2. Mapping Two Weeks of Markets onto a Single Grid
&lt;/h2&gt;&lt;p&gt;Placing the two weeks of source documents on four axes — price, volume, cost, and financing — makes the pattern clear.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Axis&lt;/th&gt;
 &lt;th&gt;Signal Confirmed in First Half of July&lt;/th&gt;
 &lt;th&gt;Investment Question&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Price (P)&lt;/td&gt;
 &lt;td&gt;Nanya DRAM ASP up more than 60% QoQ, memory supply shortage expected&lt;/td&gt;
 &lt;td&gt;How long does the price increase persist?&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Volume (Q)&lt;/td&gt;
 &lt;td&gt;Taiwan server, power, CCL, copper foil monthly revenue up; Meta 5 GW plan&lt;/td&gt;
 &lt;td&gt;Are actual shipments and utilization rates growing as fast as the narrative?&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cost (C)&lt;/td&gt;
 &lt;td&gt;Grid power, cooling, new fabs, depreciation, U.S. production costs&lt;/td&gt;
 &lt;td&gt;Does revenue growth translate into margin and free cash flow?&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Financing (F)&lt;/td&gt;
 &lt;td&gt;$25B power financing, $375M convertible notes, long-term contracts and prepayments&lt;/td&gt;
 &lt;td&gt;Who provides the capital, and who bears dilution and refinancing risk?&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Market narratives typically emphasize P and Q — prices are rising and demand is growing. But the durability of stock performance is determined by C and F. If the cost of building fabs and securing power surges, or if debt maturities arrive quickly, strong demand may not translate into strong shareholder returns.&lt;/p&gt;
&lt;p&gt;Within this grid, the key themes of the first half of July are:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;P is very strong. Memory prices are explosively expanding producer earnings.&lt;/li&gt;
&lt;li&gt;Q is also diffusing beyond chips. Server ODMs, power supply, cooling, and substrate raw materials all grew together.&lt;/li&gt;
&lt;li&gt;C is rising. Power, fab construction, cooling, and U.S. production costs are the next constraints on investment.&lt;/li&gt;
&lt;li&gt;The gap in F has widened. The divergence is growing between companies that can secure long-term capital cheaply and those that depend on dilutive funding.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id="3-what-nanya-technology-revealed-about-the-nature-of-memory-earnings"&gt;3. What Nanya Technology Revealed About the Nature of Memory Earnings
&lt;/h2&gt;&lt;p&gt;If one had to pick the single most important source document from the two weeks, it would be the &lt;a class="link" href="https://t.me/Yeouido_Lab/36066" target="_blank" rel="noopener"
 &gt;Nanya Technology FY2Q26 earnings summary&lt;/a&gt;. But more important than the Telegram post is the separation of price and volume in &lt;a class="link" href="https://www.nanya.com/tw/IR/14/%E6%B4%BB%E5%8B%95%E8%A1%8C%E4%BA%8B%E6%9B%86?IRId=4295" target="_blank" rel="noopener"
 &gt;Nanya Technology&amp;rsquo;s official earnings release&lt;/a&gt;.&lt;/p&gt;
&lt;h3 id="31-the-numbers"&gt;3.1 The Numbers
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th style="text-align: right"&gt;FY2Q26&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;/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;NT$82.549B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+68.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+684.2%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Gross margin&lt;/td&gt;
 &lt;td style="text-align: right"&gt;79.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+11.6pp&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Large improvement&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Net income&lt;/td&gt;
 &lt;td style="text-align: right"&gt;NT$50.192B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Increase&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Expanded profit&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;EPS&lt;/td&gt;
 &lt;td style="text-align: right"&gt;NT$14.66&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Increase&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Increase&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;NT$55.013B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Increase&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Increase&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Net capex&lt;/td&gt;
 &lt;td style="text-align: right"&gt;NT$4.046B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Limited&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Limited&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;NT$50.967B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Surge&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Surge&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;A 79.5% gross margin looks like a software company, yet the business is DRAM. The reason is price, not volume. According to company disclosures, DRAM average selling prices rose more than 60% quarter-on-quarter while shipments were nearly flat.&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;Revenue change ≈ Price change × Volume change
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;Nanya FY2Q26 = Price surge × Flat volume
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;This structure simultaneously explains the advantage and the risk of the current memory bull cycle.&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Advantage: When prices rise without additional wafers, a large portion of the revenue increase flows directly to operating income and cash flow.&lt;/li&gt;
&lt;li&gt;Risk: When price appreciation slows, volume growth cannot fill the gap, and earnings sensitivity works in reverse.&lt;/li&gt;
&lt;li&gt;Optical illusion: Applying a high P/E multiple to a peak-cycle gross margin means being bullish on both price and multiple simultaneously.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="32-why-this-cycle-may-last-longer-than-previous-ones"&gt;3.2 Why This Cycle May Last Longer Than Previous Ones
&lt;/h3&gt;&lt;p&gt;Nanya reported low inventory, AI infrastructure and server revenue exceeding 20% of the mix, and a supply shortage expected to continue for multiple quarters. Long-term supply contracts are also increasing. As HBM and server DRAM absorb wafer capacity and leading-edge process nodes, commodity DRAM supply has tightened as well.&lt;/p&gt;
&lt;p&gt;Micron similarly emphasized long-term customer contracts, price floors and ceilings, prepayments, and supply commitments in FY3Q26. As confirmed in the &lt;a class="link" href="https://koreainvestinsights.com/ko/post/micron-fy3q26-ai-memory-sca-fcf-hold-2026-06-25/" &gt;Micron earnings analysis&lt;/a&gt;, long-term contracts can extend the duration of memory earnings.&lt;/p&gt;
&lt;p&gt;But long-term contracts do not eliminate the cycle.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Contracts transform price risk into repricing-at-renewal risk and customer credit risk.&lt;/li&gt;
&lt;li&gt;Price ceilings protect downside but can cap additional upside during rapid price appreciation.&lt;/li&gt;
&lt;li&gt;Non-HBM DRAM and NAND remain exposed to spot and contract pricing cycles.&lt;/li&gt;
&lt;li&gt;When customers absorb excess capacity, they renegotiate volumes and prices in the next contract cycle.&lt;/li&gt;
&lt;/ol&gt;
&lt;h3 id="33-supply-shortage-is-seeding-2028-supply-expansion"&gt;3.3 Supply Shortage Is Seeding 2028 Supply Expansion
&lt;/h3&gt;&lt;p&gt;Nanya has set a FY2026 capex cap of NT$52B, and phase one of its new fab targets monthly output of 30,000 wafers by 2028, with estimated total investment of $16 billion for full 45,000 wafer-per-month capacity. Micron also announced plans to invest more than $250 billion in the United States through 2035, targeting 40% of global DRAM production on U.S. soil, per &lt;a class="link" href="https://investors.micron.com/news-releases/news-release-details/micron-accelerates-us-investments-pours-first-concrete-new-york" target="_blank" rel="noopener"
 &gt;Micron&amp;rsquo;s official announcement&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;The current supply shortage is real. At the same time, that shortage is building new supply and creating depreciation loads for 2028 and beyond. Investors must therefore hold two sentences simultaneously:&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Memory prices and earnings in 2026–2027 may be strong. Beginning in 2028, the pace at which incremental capacity passes customer qualification and the weight of accumulated depreciation must be recalculated.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="4-samsung-electronics-and-hbm4e-customer-shipments-matter-more-than-yield-numbers"&gt;4. Samsung Electronics and HBM4E: Customer Shipments Matter More Than Yield Numbers
&lt;/h2&gt;&lt;p&gt;In early July, a widely circulated Telegram post reported that Samsung Electronics&amp;rsquo; HBM4E trial yield had exceeded 70%. The &lt;a class="link" href="https://t.me/corevalue/45816" target="_blank" rel="noopener"
 &gt;source post&lt;/a&gt; is a secondary document relaying news coverage and internal commentary. It is not a number confirmed by Samsung&amp;rsquo;s official IR or customer qualification results.&lt;/p&gt;
&lt;p&gt;Even if the 70% yield figure is accurate, three additional steps are required before it informs investment judgment.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;What process, stack configuration, and test stage does the yield refer to?&lt;/li&gt;
&lt;li&gt;Has it passed customer reliability testing and system-level qualification?&lt;/li&gt;
&lt;li&gt;Is commercial volume being shipped repeatedly and recognized as revenue?&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;In HBM, test yield, mass production yield, and customer qualification yield are not the same number. The most common error the market makes is mapping a strong internal test result directly to customer revenue.&lt;/p&gt;
&lt;p&gt;Samsung Electronics earnings forecasts circulating over the same period also require careful categorization. The &lt;a class="link" href="https://t.me/corevalue/45934" target="_blank" rel="noopener"
 &gt;Samsung memory-related post&lt;/a&gt; relays brokerage estimates and industry interpretation. Q2 operating income, bonus accruals, NAND pricing, enterprise SSD and KV-cache demand, and foundry customer discussions are all mixed into a single post. Items that are not official earnings releases or contract disclosures should remain as estimates.&lt;/p&gt;
&lt;p&gt;Nonetheless, some directional conclusions are valid.&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;HBM shortages are reshuffling DRAM wafer allocation.&lt;/li&gt;
&lt;li&gt;As inference workloads grow, enterprise SSDs and NAND may become more critical at the lower tiers of the memory hierarchy.&lt;/li&gt;
&lt;li&gt;The investment thesis for Samsung Electronics is not solely about HBM share recovery, but about the optionality of combining DRAM, NAND, foundry, and packaging within a single company.&lt;/li&gt;
&lt;li&gt;The value of that optionality rises when customer qualification and recurring revenue are confirmed.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="5-taiwans-june-revenue-data-shows-ai-investment-diffusion"&gt;5. Taiwan&amp;rsquo;s June Revenue Data Shows AI Investment Diffusion
&lt;/h2&gt;&lt;p&gt;The strongest cross-verification in these two weeks came from the monthly revenue of Taiwan&amp;rsquo;s supply chain. It was not a single company&amp;rsquo;s guidance — rather, revenue at different nodes of the value chain moved in the same direction.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Company / Segment&lt;/th&gt;
 &lt;th style="text-align: right"&gt;June 2026 Revenue&lt;/th&gt;
 &lt;th style="text-align: right"&gt;MoM&lt;/th&gt;
 &lt;th style="text-align: right"&gt;YoY&lt;/th&gt;
 &lt;th&gt;What to Note&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;TSMC / Foundry&lt;/td&gt;
 &lt;td style="text-align: right"&gt;NT$442.680B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+6.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+67.9%&lt;/td&gt;
 &lt;td&gt;Sustained demand for AI compute chips and leading-edge processes&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Wiwynn / Server ODM&lt;/td&gt;
 &lt;td style="text-align: right"&gt;NT$111.371B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+32.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+29.8%&lt;/td&gt;
 &lt;td&gt;AI server and general server shipment expansion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Delta / Power &amp;amp; Cooling&lt;/td&gt;
 &lt;td style="text-align: right"&gt;NT$65.603B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+11.3%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+55.4%&lt;/td&gt;
 &lt;td&gt;Power supply unit and liquid cooling demand diffusing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;EMC / Advanced CCL&lt;/td&gt;
 &lt;td style="text-align: right"&gt;NT$17.733B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Increase per data&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+120.7%&lt;/td&gt;
 &lt;td&gt;Growth in advanced materials for AI servers and 800G switches&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;TUC / Advanced CCL&lt;/td&gt;
 &lt;td style="text-align: right"&gt;NT$4.895B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+2.1%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+106.3%&lt;/td&gt;
 &lt;td&gt;Higher share of M7/M8 high-value products&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Co-Tech / HVLP Copper Foil&lt;/td&gt;
 &lt;td style="text-align: right"&gt;NT$0.962B&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+2.2%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+43.4%&lt;/td&gt;
 &lt;td&gt;Growing bottleneck in low-roughness copper foil for high-speed signals&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;TSMC figures can be verified at the &lt;a class="link" href="https://investor.tsmc.com/english/monthly-revenue/2026" target="_blank" rel="noopener"
 &gt;official monthly revenue page&lt;/a&gt;. Wiwynn, Delta, EMC, TUC, and Co-Tech figures were cross-referenced against Telegram source posts relaying each company&amp;rsquo;s IR. The &lt;a class="link" href="https://t.me/merITz_tech/17574" target="_blank" rel="noopener"
 &gt;Wiwynn&lt;/a&gt;, &lt;a class="link" href="https://t.me/merITz_tech/17602" target="_blank" rel="noopener"
 &gt;Delta&lt;/a&gt;, and &lt;a class="link" href="https://t.me/merITz_tech/17598" target="_blank" rel="noopener"
 &gt;Co-Tech&lt;/a&gt; posts are the primary references.&lt;/p&gt;
&lt;h3 id="51-the-physical-bill-beyond-the-gpu"&gt;5.1 The Physical Bill Beyond the GPU
&lt;/h3&gt;&lt;p&gt;Adding one AI server requires more than a GPU.&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;Compute chip
&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; → Server assembly and rack
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; → Power supply unit and power conversion
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; → Liquid cooling
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; → High-speed PCB, CCL, HVLP copper foil
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; → Optical modules and switches
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; → Transformers, generation, grid interconnection
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;In 2024–2025, compute chips were the scarcest item. June&amp;rsquo;s monthly revenue data shows the bottleneck migrating to the periphery. The growth in CCL and HVLP copper foil in particular illustrates that materials reducing signal loss at higher data transfer rates gain value as speeds increase.&lt;/p&gt;
&lt;p&gt;Year-over-year growth rates alone should not be used to confirm the duration of supply shortages.&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;The prior-year base may be low.&lt;/li&gt;
&lt;li&gt;Price increases and volume growth must be separated.&lt;/li&gt;
&lt;li&gt;If new line ramp-ups drove revenue, the comparison base rises in subsequent quarters.&lt;/li&gt;
&lt;li&gt;Customer inventory building may be running ahead of end demand.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The correct verification sequence is therefore: &lt;code&gt;revenue → shipment volume → average selling price → inventory → operating cash flow&lt;/code&gt;. If revenue rises while inventory and accounts receivable grow faster, orders may be running ahead of final consumption.&lt;/p&gt;
&lt;h2 id="6-optical-networking-and-cpo-direction-is-right-numbers-need-more-verification"&gt;6. Optical Networking and CPO: Direction Is Right, Numbers Need More Verification
&lt;/h2&gt;&lt;p&gt;A &lt;a class="link" href="https://t.me/Yeouido_Lab/36085" target="_blank" rel="noopener"
 &gt;post on optical networking and CPO&lt;/a&gt; presented a brokerage forecast that the global optical networking market could grow from $15 billion in 2026 to $154 billion in 2028, with CPO accounting for $91 billion. The same source projected that network value and optical module counts per next-generation rack would increase substantially, and that silicon photonics adoption rates would surge.&lt;/p&gt;
&lt;p&gt;The directional case is persuasive. As GPU counts rise and data movement within and between racks grows, bandwidth and power constraints become increasingly difficult to resolve with electrical interconnects alone. Moving from 400G to 800G, 1.6T, and 3.2T makes light sources, lasers, silicon photonics, packaging, and precision alignment increasingly critical.&lt;/p&gt;
&lt;p&gt;However, the market size figures could not be independently verified against the original report and methodology. It is therefore safer to read them as follows:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Assessment&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;Bandwidth becomes a bottleneck in AI infrastructure&lt;/td&gt;
 &lt;td&gt;High confidence&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Optical module and silicon photonics share increases&lt;/td&gt;
 &lt;td&gt;Medium-high confidence&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CPO market reaches $91B by 2028&lt;/td&gt;
 &lt;td&gt;Secondary source; further verification needed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Specific Korean companies immediately capture large revenue&lt;/td&gt;
 &lt;td&gt;Unconfirmed until customer, component, and mass production disclosures&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;In Korean equities, as covered in the &lt;a class="link" href="https://koreainvestinsights.com/ko/post/korea-optical-cpo-value-chain-seven-companies-2026-05-09/" &gt;optical networking and CPO value chain analysis&lt;/a&gt;, actual products and customer qualifications matter more than thematic labels. A company whose name includes &amp;ldquo;optical&amp;rdquo; is not the same company as one supplying components into 1.6T optical links in an AI cluster.&lt;/p&gt;
&lt;h2 id="7-the-power-bottleneck-is-real-metas-5-gw-and-brookfields-25-billion"&gt;7. The Power Bottleneck Is Real: Meta&amp;rsquo;s 5 GW and Brookfield&amp;rsquo;s $25 Billion
&lt;/h2&gt;&lt;p&gt;Even with strong memory and server revenues, a data center generates no revenue if it cannot connect power. The clearest official data on this point from the first half of July came from Meta and Brookfield.&lt;/p&gt;
&lt;h3 id="71-meta-hyperion"&gt;7.1 Meta Hyperion
&lt;/h3&gt;&lt;p&gt;Meta&amp;rsquo;s Louisiana data center expansion plan outlined &lt;a class="link" href="https://about.fb.com/news/2026/07/teachers-local-businesses-win-as-meta-expands-louisiana-data-center/" target="_blank" rel="noopener"
 &gt;up to 5 GW of compute capacity and more than $50 billion in investment&lt;/a&gt;, including more than $1 billion in local infrastructure, seven new gas power plants, three grid battery storage systems, nuclear output uprates, and power purchase agreements — with Meta bearing the costs of power, water, and infrastructure.&lt;/p&gt;
&lt;p&gt;The importance of this announcement is not merely the scale of the numbers. It reflects the fact that AI data centers are evolving from semiconductor procurement projects into &lt;strong&gt;regional power system redesign projects&lt;/strong&gt;.&lt;/p&gt;
&lt;p&gt;Meta also announced in 2025 a joint venture with Blue Owl for Hyperion development and a structure to raise approximately $27 billion in development financing. The &lt;a class="link" href="https://about.fb.com/news/2025/10/meta-blue-owl-capital-develop-hyperion-data-center/" target="_blank" rel="noopener"
 &gt;official announcement&lt;/a&gt; shows that even big tech prefers to share capital and risk with external investors rather than putting all facilities directly on its own balance sheet.&lt;/p&gt;
&lt;h3 id="72-brookfield-and-bloom-energy"&gt;7.2 Brookfield and Bloom Energy
&lt;/h3&gt;&lt;p&gt;Brookfield and Bloom Energy expanded their AI infrastructure power project financing framework from an existing $5 billion to &lt;a class="link" href="https://investor.bloomenergy.com/press-releases/press-release-details/2026/Brookfield-and-Bloom-Energy-Expand-AI-Infrastructure-Partnership-to-25-Billion-Fivefold-Increase-to-Build-and-Finance-Rapid-Power-for-AI-Infrastructure/default.aspx" target="_blank" rel="noopener"
 &gt;$25 billion&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;This $25 billion should not be read as confirmed Bloom Energy revenue. A financing framework shows the ceiling and commitment of capital that can be deployed into projects. Actual revenue requires land, customers, power purchase agreements, fuel supply, equipment orders, and groundbreaking.&lt;/p&gt;
&lt;p&gt;Nevertheless, the signal quality is high.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Data centers find it difficult to wait for grid interconnection and are seeking on-site power solutions.&lt;/li&gt;
&lt;li&gt;Equipment vendor orders alone are insufficient — long-term financing must accompany them.&lt;/li&gt;
&lt;li&gt;Capital cost and contract structure may determine project cadence more than equipment availability.&lt;/li&gt;
&lt;li&gt;Profit sharing among big tech, infrastructure funds, and power operators becomes increasingly important.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id="8-same-power-theme-different-shareholder-outcomes-fermis-convertible-notes"&gt;8. Same Power Theme, Different Shareholder Outcomes: Fermi&amp;rsquo;s Convertible Notes
&lt;/h2&gt;&lt;p&gt;Strong power and data center demand does not mean all related equities perform well. The &lt;a class="link" href="https://t.me/Yeouido_Lab/36081" target="_blank" rel="noopener"
 &gt;Fermi-related Telegram post&lt;/a&gt; illustrated this counterexample clearly.&lt;/p&gt;
&lt;p&gt;Fermi priced $375 million of 5% convertible senior notes due 2031, with a potential over-allotment bringing the total to $431.25 million. Reading the &lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/2071778/000121390026076788/ea0297535-8k_fermi.htm" target="_blank" rel="noopener"
 &gt;SEC filing&lt;/a&gt; and &lt;a class="link" href="https://www.nasdaq.com/press-release/fermi-inc-prices-upsized-offering-375-million-convertible-senior-notes-dilution" target="_blank" rel="noopener"
 &gt;final pricing release&lt;/a&gt; together, it is clear that the company is at the stage of funding large-scale data center ambitions through capital markets before those ambitions generate revenue.&lt;/p&gt;
&lt;p&gt;When evaluating power themes, announced GW targets and confirmed cash flows must be distinguished.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Stage&lt;/th&gt;
 &lt;th&gt;Evidence to Verify&lt;/th&gt;
 &lt;th&gt;Meaning for Shareholders&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Concept&lt;/td&gt;
 &lt;td&gt;Site, target GW, development plan&lt;/td&gt;
 &lt;td&gt;Optionality exists but no revenue&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Contract&lt;/td&gt;
 &lt;td&gt;Customer, PPA, minimum usage commitments&lt;/td&gt;
 &lt;td&gt;Revenue visibility begins&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Financing&lt;/td&gt;
 &lt;td&gt;Interest rate, collateral, conversion price, maturity&lt;/td&gt;
 &lt;td&gt;Dilution and refinancing risk determined&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Construction&lt;/td&gt;
 &lt;td&gt;EPC, equipment orders, permits&lt;/td&gt;
 &lt;td&gt;Cost and schedule risk materializes&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Operations&lt;/td&gt;
 &lt;td&gt;Utilization, power cost, load factor&lt;/td&gt;
 &lt;td&gt;Operating cash flow confirmed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Brookfield&amp;rsquo;s $25 billion and Fermi&amp;rsquo;s $375 million are both AI power investments. The former shows the strength of long-term infrastructure financing; the latter shows the dilution cost borne by shareholders of a development-stage company with no revenue yet. Even within the same theme, the quality of capital determines shareholder returns.&lt;/p&gt;
&lt;h2 id="9-metas-excess-compute-sales-evidence-of-oversupply-or-an-option"&gt;9. Meta&amp;rsquo;s Excess Compute Sales: Evidence of Oversupply or an Option?
&lt;/h2&gt;&lt;p&gt;A &lt;a class="link" href="https://t.me/eqmirae/3050" target="_blank" rel="noopener"
 &gt;Meta cloud-related post&lt;/a&gt; covered reports that Meta is exploring selling AI computing capacity beyond its own internal use to external customers. This plan is based on a &lt;a class="link" href="https://news.bloomberglaw.com/tech-and-telecom-law/meta-is-building-a-cloud-business-to-sell-excess-ai-compute-1" target="_blank" rel="noopener"
 &gt;Bloomberg report&lt;/a&gt; citing anonymous sources at the planning stage. No confirmed product, pricing, or contract scale has been disclosed.&lt;/p&gt;
&lt;p&gt;This news can be read two ways.&lt;/p&gt;
&lt;h3 id="bearish-reading"&gt;Bearish Reading
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Meta has secured more capacity than it needs.&lt;/li&gt;
&lt;li&gt;The pace of monetization from internal AI services has not kept up with capacity expansion.&lt;/li&gt;
&lt;li&gt;Selling surplus compute signals a coming slowdown in capital expenditure.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="bullish-reading"&gt;Bullish Reading
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Training and inference demand varies by time-of-day and project.&lt;/li&gt;
&lt;li&gt;Selling unused capacity to external customers improves asset turnover and return on investment.&lt;/li&gt;
&lt;li&gt;It creates an incremental revenue stream beyond advertising.&lt;/li&gt;
&lt;li&gt;It lowers the downside of large-scale preemptive investment, creating an embedded option.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Current data alone cannot determine which reading is correct. The necessary indicators are:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;External sales volume and pricing&lt;/li&gt;
&lt;li&gt;GPU utilization rate for internal workloads&lt;/li&gt;
&lt;li&gt;Contribution margin inclusive of depreciation and power costs&lt;/li&gt;
&lt;li&gt;Additional capex guidance&lt;/li&gt;
&lt;li&gt;Minimum usage commitments from customers and contract duration&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;The ability to sell surplus capacity reveals the possibility of overcapacity but is not itself evidence of overcapacity. In the best case, it is monetization of underutilized assets; in the worst case, it is a post-hoc response to overestimated internal demand.&lt;/p&gt;
&lt;h2 id="10-the-agent-era-and-memory-demand-translating-narrative-into-physical-quantities"&gt;10. The Agent Era and Memory Demand: Translating Narrative into Physical Quantities
&lt;/h2&gt;&lt;p&gt;A &lt;a class="link" href="https://t.me/Yeouido_Lab/35634" target="_blank" rel="noopener"
 &gt;long-term outlook post on agents and memory&lt;/a&gt; presents a vision where AI advances through infrastructure construction, operations, an agent economy, and inter-agent transactions, making memory the foundation of civilization. Numbers like 10 billion agents are not validated forecasts — they are long-term narrative.&lt;/p&gt;
&lt;p&gt;But the question the post raises is valid. Agents use longer state and more tool calls than a chatbot that answers and exits. Breaking work into multiple steps, retaining prior results, and exchanging information with other agents can increase total inference volume and stored state.&lt;/p&gt;
&lt;p&gt;To translate this into memory demand, five variables must be tracked separately.&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;Total memory demand
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;≈ Number of useful tasks
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;× Inference steps per task
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;× Active context per step
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;× Fraction of context residing in memory
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;÷ Compression, cache, quantization, and routing efficiency
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;The bull case emphasizes the first four variables. The counter-argument is that the efficiency denominator is improving rapidly.&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;KV cache compression and reuse&lt;/li&gt;
&lt;li&gt;Speculative decoding&lt;/li&gt;
&lt;li&gt;Routing between small and large models&lt;/li&gt;
&lt;li&gt;Mixture-of-experts selective activation&lt;/li&gt;
&lt;li&gt;Tiered storage across HBM, DRAM, CXL, and SSD&lt;/li&gt;
&lt;li&gt;Prompt and state summarization&lt;/li&gt;
&lt;li&gt;Hardware-software co-design&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Therefore, &lt;code&gt;agents growing = HBM demand growing at the same rate&lt;/code&gt; is not a law of physics. A better benchmark is &lt;strong&gt;memory bytes-seconds per useful task completed, multiplied by total task volume&lt;/strong&gt;. If efficiency improves tenfold but total task volume grows a hundredfold, demand still increases. Conversely, if agent counts rise but most run on short, low-cost models, HBM demand may fall well short of the narrative.&lt;/p&gt;
&lt;h2 id="11-who-is-the-ultimate-payer-ai-software-profitability"&gt;11. Who Is the Ultimate Payer: AI Software Profitability
&lt;/h2&gt;&lt;p&gt;The infrastructure investment cycle can be sustained only if end customers pay the cost. In early July, a &lt;a class="link" href="https://t.me/gaoshoukorea/62002" target="_blank" rel="noopener"
 &gt;Telegram post&lt;/a&gt; shared that Anthropic is profitable in Q2 on a stock-based compensation-excluded basis, and that token-level gross margins on some models exceed 80%.&lt;/p&gt;
&lt;p&gt;The sources are a &lt;a class="link" href="https://sequoiacap.com/podcast/dylan-patel-of-semianalysis-why-hardware-software-co-design-is-ais-real-100x/" target="_blank" rel="noopener"
 &gt;Sequoia interview with Dylan Patel&lt;/a&gt; and &lt;a class="link" href="https://newsletter.semianalysis.com/p/anthropic-growth-and-bedrock-mix" target="_blank" rel="noopener"
 &gt;SemiAnalysis estimates&lt;/a&gt;. These are not Anthropic&amp;rsquo;s audited financial statements. They should be read as interview statements and model-based estimates.&lt;/p&gt;
&lt;p&gt;In particular, API list-price token margins and company-wide profitability are different things.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Item&lt;/th&gt;
 &lt;th&gt;Included in Token-Level Margin&lt;/th&gt;
 &lt;th&gt;What Is Additionally Required for Company-Wide Profit&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Inference revenue&lt;/td&gt;
 &lt;td&gt;Included&lt;/td&gt;
 &lt;td&gt;Reflect channel discounts and commitments&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Inference compute&lt;/td&gt;
 &lt;td&gt;Included&lt;/td&gt;
 &lt;td&gt;Reflect idle capacity and peak costs&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cloud distribution&lt;/td&gt;
 &lt;td&gt;Partial estimate&lt;/td&gt;
 &lt;td&gt;Bedrock/Vertex revenue sharing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;R&amp;amp;D&lt;/td&gt;
 &lt;td&gt;May be excluded&lt;/td&gt;
 &lt;td&gt;Large-scale training and headcount&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Stock-based compensation&lt;/td&gt;
 &lt;td&gt;May be excluded&lt;/td&gt;
 &lt;td&gt;Fully diluted cost&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Sales &amp;amp; marketing&lt;/td&gt;
 &lt;td&gt;May be excluded&lt;/td&gt;
 &lt;td&gt;Enterprise sales and customer support&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Even so, the relevance of the software profitability discussion is clear. If end services generate high contribution margins, the capex on GPUs, memory, networking, and power can be sustained. Conversely, if token price competition intensifies and revenue per task falls, it becomes harder for infrastructure providers to maintain high pricing for long.&lt;/p&gt;
&lt;p&gt;The true end of the AI supply chain is not HBM shipments.&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;Amount paid by users
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; → AI service gross margin
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; → Cloud and model provider capex capacity
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; → GPU, memory, networking, and power orders
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;If cash runs short at any link in this chain, the backlog upstream gets revised later.&lt;/p&gt;
&lt;h2 id="12-koreas-separate-variable-single-stock-leveraged-products"&gt;12. Korea&amp;rsquo;s Separate Variable: Single-Stock Leveraged Products
&lt;/h2&gt;&lt;p&gt;Source documents on the Korean equity market repeatedly claimed that single-stock leveraged products tied to Samsung Electronics and SK Hynix have been absorbing positioning flows and amplifying volatility. A &lt;a class="link" href="https://t.me/corevalue/46025" target="_blank" rel="noopener"
 &gt;representative post&lt;/a&gt; cited cumulative retail buying of ₩14.2 trillion since product launch. This specific cumulative figure and its measurement scope could not be independently verified.&lt;/p&gt;
&lt;p&gt;Separately, media reporting noted that during the four trading days around the June sell-off, retail investors net-bought ₩14.3 trillion in KOSPI and deployed ₩10.7 trillion into Samsung Electronics and SK Hynix. The &lt;a class="link" href="https://v.daum.net/v/20260625060155479" target="_blank" rel="noopener"
 &gt;relevant report&lt;/a&gt; is the reference. This figure covers a different period and scope than the &amp;ldquo;₩14.2 trillion cumulative since product launch&amp;rdquo; figure. The two should not be treated as the same statistic.&lt;/p&gt;
&lt;p&gt;What is officially confirmed is the existence of the products and the regulatory risk warnings.&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;The Financial Supervisory Service warned on May 27, prior to launch, about the &lt;a class="link" href="https://dart.fss.or.kr/dsaa003/selectBodoMain.ax?seqno=28181" target="_blank" rel="noopener"
 &gt;loss amplification and negative compounding effects of single-stock leveraged and inverse products&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;KRX listed the &lt;a class="link" href="https://kind.krx.co.kr/external/2026/05/22/000241/20260522000586/68152.htm" target="_blank" rel="noopener"
 &gt;Samsung Electronics single-stock leveraged ETF&lt;/a&gt; and an &lt;a class="link" href="https://kind.krx.co.kr/external/2026/05/22/000553/20260522001367/68342.htm" target="_blank" rel="noopener"
 &gt;SK Hynix leveraged ETN&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;KRX issued an &lt;a class="link" href="https://kind.krx.co.kr/external/2026/06/08/000981/20260608001993/99355.htm" target="_blank" rel="noopener"
 &gt;investor caution&lt;/a&gt; regarding the premium/discount to NAV in an SK Hynix futures-leveraged ETF.&lt;/li&gt;
&lt;li&gt;Creation and redemption disclosures for related ETFs appeared in July, indicating that product flows can move in either direction rather than representing fixed buying demand.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="121-why-volatility-increases"&gt;12.1 Why Volatility Increases
&lt;/h3&gt;&lt;p&gt;Leveraged products must rebalance daily to maintain their target multiple. When the underlying rises, they may need to add exposure; when it falls sharply, they may need to reduce it. The exact timing of transactions depends on the mix of futures, swaps, and equity hedges, but the structural tendency is to amplify both upside and downside.&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;Rising market
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Leveraged product NAV increases
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Exposure expansion needed to maintain target multiple
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Potential additional demand for underlying
&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;Falling market
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ NAV decline and redemptions
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Hedge reduction and potential futures selling
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;→ Amplified underlying volatility
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;This structure does not turn good earnings into bad ones. But it can steepen the price path even on days when fundamentals have not changed. Analyzing Samsung Electronics and SK Hynix therefore requires monitoring ETF NAV, creation and redemption flows, futures basis, and program trading flows alongside earnings forecasts.&lt;/p&gt;
&lt;h2 id="13-sk-hynix-adr-the-premium-may-be-the-price-of-friction-not-of-enterprise-value"&gt;13. SK Hynix ADR: The Premium May Be the Price of Friction, Not of Enterprise Value
&lt;/h2&gt;&lt;p&gt;A &lt;a class="link" href="https://t.me/Yeouido_Lab/36093" target="_blank" rel="noopener"
 &gt;post on SK Hynix ADR&lt;/a&gt; argued that when real-time arbitrage between the U.S. ADR and the Korean ordinary shares is constrained by settlement lag, taxes, and regulations, an initial ADR premium can emerge. The post also cited a TSMC-like premium range for reference.&lt;/p&gt;
&lt;p&gt;This interpretation is plausible but not confirmed. ADR pricing can be framed as follows:&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;ADR price
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;≈ Ordinary share price × conversion ratio × FX rate
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;+ U.S. investor accessibility premium
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;+ borrow, settlement, tax, and conversion friction
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;- new supply and arbitrage pressure
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;Initially, if U.S. investor demand is fast and the supply of ordinaries converted to ADRs is slow, a premium can emerge. But that premium does not mean SK Hynix&amp;rsquo;s operating income has increased. It is the market structure&amp;rsquo;s price for friction — one that is likely to compress as conversion and arbitrage normalize.&lt;/p&gt;
&lt;p&gt;For an ADR to create long-term value, at least one of the following is needed:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;An expanded U.S. investor base lowers the required rate of return.&lt;/li&gt;
&lt;li&gt;Higher trading liquidity and analyst coverage add a monitoring premium.&lt;/li&gt;
&lt;li&gt;U.S. capital markets provide more favorable financing for capex.&lt;/li&gt;
&lt;li&gt;Index inclusion or the lifting of institutional investment restrictions creates real incremental demand.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;Conversely, looking only at the initial premium and double-counting enterprise value across ordinary shares and the ADR would be an error. As discussed in the &lt;a class="link" href="https://koreainvestinsights.com/ko/post/sk-hynix-adr-000660-price-path-leverage-etf-plumbing-2026-07-04/" &gt;SK Hynix ADR and leverage structure analysis&lt;/a&gt;, the ADR is plumbing layered on top of fundamentals. Plumbing changes price discovery but does not automatically generate cash flow.&lt;/p&gt;
&lt;h2 id="14-when-four-signals-collide"&gt;14. When Four Signals Collide
&lt;/h2&gt;&lt;p&gt;The source documents from the first half of July do not all point in the same direction. The four key conflicts are what matter most.&lt;/p&gt;
&lt;h3 id="conflict-1-memory-prices-are-strong-but-volume-is-still-weak"&gt;Conflict 1: Memory Prices Are Strong But Volume Is Still Weak
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Nanya&amp;rsquo;s prices rose more than 60%.&lt;/li&gt;
&lt;li&gt;Shipment volume barely increased.&lt;/li&gt;
&lt;li&gt;Earnings exploded but peak-price risk also increased.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Judgment: In the next quarter, watch bit shipments and customer inventory more than the rate of price appreciation.&lt;/p&gt;
&lt;h3 id="conflict-2-ai-component-revenue-is-diffusing-but-power-and-financing-costs-are-rising-too"&gt;Conflict 2: AI Component Revenue Is Diffusing, But Power and Financing Costs Are Rising Too
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Server ODM, power supply, CCL, and copper foil revenues are strong.&lt;/li&gt;
&lt;li&gt;Meta plans 5 GW and more than $50 billion in investment.&lt;/li&gt;
&lt;li&gt;Power financing has expanded to $25 billion.&lt;/li&gt;
&lt;li&gt;Development-stage companies rely on convertible notes and dilutive capital.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Judgment: Watch contract counterparties, cost of capital, and free cash flow more than revenue growth rates.&lt;/p&gt;
&lt;h3 id="conflict-3-agent-demand-is-growing-but-per-unit-memory-cost-is-falling"&gt;Conflict 3: Agent Demand Is Growing, But Per-Unit Memory Cost Is Falling
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Long-horizon state, tool use, and inter-agent communication increase total inference volume.&lt;/li&gt;
&lt;li&gt;Compression, caching, tiered storage, and small-model routing lower cost per byte.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Judgment: Watch cost per useful task and total task volume together, not just agent count.&lt;/p&gt;
&lt;h3 id="conflict-4-korean-memory-companies-earnings-are-strong-but-the-price-plumbing-is-unstable"&gt;Conflict 4: Korean Memory Companies&amp;rsquo; Earnings Are Strong, But the Price Plumbing Is Unstable
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;Memory prices and cash flows are strong.&lt;/li&gt;
&lt;li&gt;Single-stock leverage products and ADR expectations amplify the price path.&lt;/li&gt;
&lt;li&gt;Even strong companies can experience sharp drawdowns from redemptions, hedge unwinds, and premium normalization.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Judgment: Earnings estimates and position sizing must be kept separate. Even with high conviction in the business, an unstable positioning structure requires a different approach to entry and sizing.&lt;/p&gt;
&lt;h2 id="15-scenarios-from-h2-2026-through-2028"&gt;15. Scenarios from H2 2026 Through 2028
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Scenario&lt;/th&gt;
 &lt;th&gt;H2 2026&lt;/th&gt;
 &lt;th&gt;2027&lt;/th&gt;
 &lt;th&gt;2028&lt;/th&gt;
 &lt;th&gt;Beneficiary&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;A. Orderly excess demand&lt;/td&gt;
 &lt;td&gt;Memory prices and Taiwan revenue stay strong&lt;/td&gt;
 &lt;td&gt;Growth continues via long-term contracts and power expansion&lt;/td&gt;
 &lt;td&gt;New supply absorbed gradually&lt;/td&gt;
 &lt;td&gt;HBM, eSSD, power supply, cooling, advanced materials&lt;/td&gt;
 &lt;td&gt;Overvalued option-like developers&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;B. Strong prices, flat volume&lt;/td&gt;
 &lt;td&gt;ASP drives earnings&lt;/td&gt;
 &lt;td&gt;Customer inventory and negotiating leverage weigh&lt;/td&gt;
 &lt;td&gt;Price normalization causes earnings plunge&lt;/td&gt;
 &lt;td&gt;Low-cost producers, cash-rich companies&lt;/td&gt;
 &lt;td&gt;Producers with high price sensitivity&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;C. Power and financing delays&lt;/td&gt;
 &lt;td&gt;Chip orders hold but site delays accumulate&lt;/td&gt;
 &lt;td&gt;Some projects cancelled or renegotiated&lt;/td&gt;
 &lt;td&gt;Demand reconcentrates on large operators&lt;/td&gt;
 &lt;td&gt;Companies with long-term PPAs and strong capital&lt;/td&gt;
 &lt;td&gt;Developers without pre-contracted revenue&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;D. Simultaneous supply and efficiency improvement&lt;/td&gt;
 &lt;td&gt;Bull narrative persists&lt;/td&gt;
 &lt;td&gt;Efficiency gains partially offset demand growth&lt;/td&gt;
 &lt;td&gt;New fabs and efficiency gains overlap&lt;/td&gt;
 &lt;td&gt;Low-cost platforms, software&lt;/td&gt;
 &lt;td&gt;Commodity memory and high-fixed-cost new fabs&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;E. Leverage unwind&lt;/td&gt;
 &lt;td&gt;Repeated sharp declines disconnected from earnings&lt;/td&gt;
 &lt;td&gt;Possible product structure adjustment&lt;/td&gt;
 &lt;td&gt;Prices re-converge to fundamentals&lt;/td&gt;
 &lt;td&gt;Companies with FCF and buyback capacity&lt;/td&gt;
 &lt;td&gt;High-leverage chase positions&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The most probable path at present is a mix of A and B. Memory prices and AI component revenues are likely to remain strong in the near term, but whether volume growth and final-service monetization keep pace with price appreciation must be confirmed. 2028 is the first major verification window, when new fabs, power project startups, AI efficiency improvements, and long-term contract renewals all converge.&lt;/p&gt;
&lt;h2 id="16-indicators-to-watch-a-checklist-for-converting-narrative-into-numbers"&gt;16. Indicators to Watch: A Checklist for Converting Narrative into Numbers
&lt;/h2&gt;&lt;h3 id="memory"&gt;Memory
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Indicator&lt;/th&gt;
 &lt;th&gt;Bull Confirmation&lt;/th&gt;
 &lt;th&gt;Caution Signal&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;DRAM ASP and shipment volume&lt;/td&gt;
 &lt;td&gt;Price and bit shipments both rising&lt;/td&gt;
 &lt;td&gt;Price rising alone, shipments stagnant&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Customer inventory&lt;/td&gt;
 &lt;td&gt;Low inventory and expanding long-term contracts&lt;/td&gt;
 &lt;td&gt;Front-loading orders followed by inventory surge&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM qualification&lt;/td&gt;
 &lt;td&gt;Repeated commercial shipments and expanding customer base&lt;/td&gt;
 &lt;td&gt;Only test yields disclosed, revenue delayed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;NAND and eSSD&lt;/td&gt;
 &lt;td&gt;Price, volume, and margin improving together&lt;/td&gt;
 &lt;td&gt;Only HBM narrative strong while NAND inventory builds&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Capex&lt;/td&gt;
 &lt;td&gt;Phased expansion matched to demand&lt;/td&gt;
 &lt;td&gt;Depreciation growing faster than revenue&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h3 id="ai-hardware-supply-chain"&gt;AI Hardware Supply Chain
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Indicator&lt;/th&gt;
 &lt;th&gt;Bull Confirmation&lt;/th&gt;
 &lt;th&gt;Caution Signal&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Taiwan monthly revenue&lt;/td&gt;
 &lt;td&gt;Multiple nodes growing together for 3+ months&lt;/td&gt;
 &lt;td&gt;Only one or two companies growing, others slowing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Advanced materials&lt;/td&gt;
 &lt;td&gt;Volume and high-value mix both rising&lt;/td&gt;
 &lt;td&gt;Revenue growth driven only by price increases&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Inventory and receivables&lt;/td&gt;
 &lt;td&gt;Growing at roughly the same pace as revenue&lt;/td&gt;
 &lt;td&gt;Growing faster than revenue&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Optical networking&lt;/td&gt;
 &lt;td&gt;800G and 1.6T volume production confirmed by customers&lt;/td&gt;
 &lt;td&gt;Only market size forecasts expanding&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h3 id="power-and-financing"&gt;Power and Financing
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Indicator&lt;/th&gt;
 &lt;th&gt;Bull Confirmation&lt;/th&gt;
 &lt;th&gt;Caution Signal&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Data centers&lt;/td&gt;
 &lt;td&gt;PPA, named customers, minimum usage commitments&lt;/td&gt;
 &lt;td&gt;Only target GW announced, customers undisclosed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Funding&lt;/td&gt;
 &lt;td&gt;Long-term low-cost capital, risk sharing&lt;/td&gt;
 &lt;td&gt;High-yield debt, convertibles, short maturities&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Cash flow&lt;/td&gt;
 &lt;td&gt;FCF conversion after commissioning&lt;/td&gt;
 &lt;td&gt;Capex consistently overwhelming FCF&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Power costs&lt;/td&gt;
 &lt;td&gt;Locked in via long-term contracts, predictable&lt;/td&gt;
 &lt;td&gt;Spot fuel prices and grid costs surging&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h3 id="korean-market-structure"&gt;Korean Market Structure
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Indicator&lt;/th&gt;
 &lt;th&gt;Bull Confirmation&lt;/th&gt;
 &lt;th&gt;Caution Signal&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Leveraged products&lt;/td&gt;
 &lt;td&gt;NAV premium stable, creation/redemption gradual&lt;/td&gt;
 &lt;td&gt;Premium widening, sharp redemptions&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Futures and program flow&lt;/td&gt;
 &lt;td&gt;Balanced underlying and futures supply/demand&lt;/td&gt;
 &lt;td&gt;Concentrated end-of-day hedge trading&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;ADR&lt;/td&gt;
 &lt;td&gt;Smooth conversion, premium stable&lt;/td&gt;
 &lt;td&gt;Initial premium spikes then new supply floods in&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Retail positioning&lt;/td&gt;
 &lt;td&gt;Cash-based diversified buying&lt;/td&gt;
 &lt;td&gt;Concentrated margin and leverage use&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="17-what-to-believe-and-what-to-wait-for"&gt;17. What to Believe and What to Wait For
&lt;/h2&gt;&lt;h3 id="high-confidence-conclusions"&gt;High-Confidence Conclusions
&lt;/h3&gt;&lt;ol&gt;
&lt;li&gt;Memory price appreciation is materially lifting producer quarterly earnings.&lt;/li&gt;
&lt;li&gt;AI investment has diffused into servers, power supplies, cooling, and advanced PCB materials.&lt;/li&gt;
&lt;li&gt;Power and long-term capital have become the critical bottlenecks in data center construction.&lt;/li&gt;
&lt;li&gt;Single-stock leveraged products can amplify the daily volatility of the underlying asset.&lt;/li&gt;
&lt;li&gt;New memory supply coming online around 2028 is the direct consequence of today&amp;rsquo;s shortage.&lt;/li&gt;
&lt;/ol&gt;
&lt;h3 id="directionally-sound-but-requiring-further-verification"&gt;Directionally Sound But Requiring Further Verification
&lt;/h3&gt;&lt;ol&gt;
&lt;li&gt;Samsung Electronics HBM4E trial yield of 70% and per-customer qualification status&lt;/li&gt;
&lt;li&gt;Scale and profitability of Meta&amp;rsquo;s external AI compute sales&lt;/li&gt;
&lt;li&gt;CPO market size in 2028 and Korean companies&amp;rsquo; actual share&lt;/li&gt;
&lt;li&gt;Initial ADR premium range for SK Hynix and its expected duration&lt;/li&gt;
&lt;li&gt;How much the agent economy will independently grow HBM, DRAM, and SSD demand, respectively&lt;/li&gt;
&lt;/ol&gt;
&lt;h3 id="assertions-that-cannot-be-supported-by-current-evidence"&gt;Assertions That Cannot Be Supported by Current Evidence
&lt;/h3&gt;&lt;ol&gt;
&lt;li&gt;Inferring that something is true because it appeared frequently in Telegram&lt;/li&gt;
&lt;li&gt;Interpreting the $25 billion financing framework as $25 billion in confirmed revenue for a specific equipment maker&lt;/li&gt;
&lt;li&gt;Assuming all announced data center GW will be commissioned&lt;/li&gt;
&lt;li&gt;Interpreting ADR premiums as evidence of increased enterprise value&lt;/li&gt;
&lt;li&gt;Claiming that agent count and HBM demand grow at the same rate&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id="18-final-assessment"&gt;18. Final Assessment
&lt;/h2&gt;&lt;p&gt;Two weeks of Telegram data cannot be reduced to a single bull or bear thesis. Instead, the data reveals that the market is simultaneously pricing four different time horizons.&lt;/p&gt;
&lt;p&gt;First, this is the time of price. As Nanya&amp;rsquo;s results show, memory prices are driving earnings higher faster than volume can. In this window, producers&amp;rsquo; operating leverage is powerful.&lt;/p&gt;
&lt;p&gt;Second, next comes the time of volume diffusion. Monthly revenue from TSMC, Wiwynn, Delta, CCL makers, and HVLP copper foil producers shows that the AI investment invoice is spreading beyond GPUs. But inventory and cash flows must follow.&lt;/p&gt;
&lt;p&gt;Third, after that comes the time of power and financing. Meta&amp;rsquo;s 5 GW and Brookfield&amp;rsquo;s $25 billion show the scale. Fermi&amp;rsquo;s convertible notes ask how much of that scale comes back to shareholders as dilution cost.&lt;/p&gt;
&lt;p&gt;Fourth, 2028 is the time of supply and efficiency. The fabs being built today by high prices, the generation assets being built today by high power demand, and the efficiency gains being made today by high token costs will all enter the market simultaneously.&lt;/p&gt;
&lt;p&gt;What investors should do is not abandon AI infrastructure or pursue it unconditionally. It is to verify where each company sits in the bottleneck structure, whether its growth is driven by price or volume, and who bears the cost and financing burden.&lt;/p&gt;
&lt;p&gt;The best companies own the bottleneck, have low cost of capital, generate recurring revenue and cash flow, and can expand capacity in stages. The most dangerous companies announce enormous target capacities but have weak customer contracts and thin financing, and convert stock price appreciation back into dilutive capital.&lt;/p&gt;
&lt;p&gt;Telegram does not reveal this distinction in advance. Telegram shows where people are running. From that point, official disclosures, income statements, cash flow statements, and the terms of contracts and financing must be read directly.&lt;/p&gt;
&lt;div class="thesis-callout"&gt;
 &lt;div class="thesis-callout__label"&gt;Two-Week Deep Dive Conclusion&lt;/div&gt;
 &lt;div class="thesis-callout__body"&gt;
 Demand for AI infrastructure remains strong. But the excess returns flow not from the "AI" label itself, but to &lt;strong&gt;memory producers earning today's prices, component makers whose actual volumes are growing, and operators who provide both power and capital&lt;/strong&gt;. Watch the shortage of 2026–2027, but also watch simultaneously for the supply and depreciation that shortage is building toward 2028.
 &lt;/div&gt;
&lt;/div&gt;
&lt;h2 id="key-source-documents-and-official-references"&gt;Key Source Documents and Official References
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://t.me/Yeouido_Lab/36066" target="_blank" rel="noopener"
 &gt;Nanya Technology FY2Q26 Telegram source&lt;/a&gt;, &lt;a class="link" href="https://www.nanya.com/tw/IR/14/%E6%B4%BB%E5%8B%95%E8%A1%8C%E4%BA%8B%E6%9B%86?IRId=4295" target="_blank" rel="noopener"
 &gt;Nanya Technology official IR&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://investor.tsmc.com/english/monthly-revenue/2026" target="_blank" rel="noopener"
 &gt;TSMC 2026 monthly revenue&lt;/a&gt;, &lt;a class="link" href="https://investor.tsmc.com/english/financial-calendar" target="_blank" rel="noopener"
 &gt;TSMC financial calendar&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://t.me/merITz_tech/17574" target="_blank" rel="noopener"
 &gt;Wiwynn June revenue source&lt;/a&gt;, &lt;a class="link" href="https://t.me/merITz_tech/17602" target="_blank" rel="noopener"
 &gt;Delta June revenue source&lt;/a&gt;, &lt;a class="link" href="https://t.me/merITz_tech/17598" target="_blank" rel="noopener"
 &gt;Co-Tech June revenue source&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://about.fb.com/news/2026/07/teachers-local-businesses-win-as-meta-expands-louisiana-data-center/" target="_blank" rel="noopener"
 &gt;Meta Louisiana data center official announcement&lt;/a&gt;, &lt;a class="link" href="https://about.fb.com/news/2025/10/meta-blue-owl-capital-develop-hyperion-data-center/" target="_blank" rel="noopener"
 &gt;Meta · Blue Owl joint venture official announcement&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://investor.bloomenergy.com/press-releases/press-release-details/2026/Brookfield-and-Bloom-Energy-Expand-AI-Infrastructure-Partnership-to-25-Billion-Fivefold-Increase-to-Build-and-Finance-Rapid-Power-for-AI-Infrastructure/default.aspx" target="_blank" rel="noopener"
 &gt;Brookfield and Bloom Energy $25 billion financing framework&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/2071778/000121390026076788/ea0297535-8k_fermi.htm" target="_blank" rel="noopener"
 &gt;Fermi SEC filing&lt;/a&gt;, &lt;a class="link" href="https://www.nasdaq.com/press-release/fermi-inc-prices-upsized-offering-375-million-convertible-senior-notes-dilution" target="_blank" rel="noopener"
 &gt;Convertible notes final pricing&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://investors.micron.com/news-releases/news-release-details/micron-accelerates-us-investments-pours-first-concrete-new-york" target="_blank" rel="noopener"
 &gt;Micron U.S. investment official announcement&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://dart.fss.or.kr/dsaa003/selectBodoMain.ax?seqno=28181" target="_blank" rel="noopener"
 &gt;FSS single-stock leveraged and inverse product warning&lt;/a&gt;, &lt;a class="link" href="https://kind.krx.co.kr/external/2026/06/08/000981/20260608001993/99355.htm" target="_blank" rel="noopener"
 &gt;KRX NAV premium/discount investor caution&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://news.bloomberglaw.com/tech-and-telecom-law/meta-is-building-a-cloud-business-to-sell-excess-ai-compute-1" target="_blank" rel="noopener"
 &gt;Meta excess compute sales report&lt;/a&gt;, &lt;a class="link" href="https://t.me/eqmirae/3050" target="_blank" rel="noopener"
 &gt;Related Telegram interpretation&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://sequoiacap.com/podcast/dylan-patel-of-semianalysis-why-hardware-software-co-design-is-ais-real-100x/" target="_blank" rel="noopener"
 &gt;AI software profitability interview&lt;/a&gt;, &lt;a class="link" href="https://newsletter.semianalysis.com/p/anthropic-growth-and-bedrock-mix" target="_blank" rel="noopener"
 &gt;SemiAnalysis estimates&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Data as of July 1–14, 2026. Channel and message counts are based on collected public Telegram posts; July 14 may have partial gaps due to connection errors. Company earnings and investment plans prioritize official sources; brokerage estimates, anonymous reports, and Telegram interpretations are separately identified. This post is a market analysis for informational purposes and does not constitute a recommendation to buy or sell any specific security.&lt;/p&gt;

 &lt;/blockquote&gt;</description></item><item><title>2028 Matters More Than the 2027 Boom: Integrated Scenarios for Samsung Electronics &amp; SK Hynix</title><link>https://koreainvestinsights.com/post/samsung-sk-hynix-2027-2028-integrated-scenarios-risk-adjusted-valuation-2026-07-13/</link><pubDate>Mon, 13 Jul 2026 20:39:17 +0900</pubDate><guid>https://koreainvestinsights.com/post/samsung-sk-hynix-2027-2028-integrated-scenarios-risk-adjusted-valuation-2026-07-13/</guid><description>
 &lt;blockquote&gt;
 &lt;p&gt;Related context: This post is a follow-up to the &lt;a class="link" href="https://koreainvestinsights.com/post/hbm-2030-supply-demand-267eb-demand-model-crosscheck-2026-07-13/" &gt;HBM 2030 Supply-Demand Model Cross-Check&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/samsung-hynix-2028e-profit-valuation-cycle-scenarios-2026-07-09/" &gt;Samsung Electronics &amp;amp; SK Hynix Valuation Through 2028E Earnings&lt;/a&gt;, and the &lt;a class="link" href="https://koreainvestinsights.com/post/kospi-9pct-selloff-ai-hardware-derating-korea-leverage-2026-07-13/" &gt;July 13 AI Hardware Selloff Analysis&lt;/a&gt;. Accepting 2027&amp;rsquo;s boom as the baseline, it quantifies the probability that supply, efficiency, and financing conditions shift simultaneously beginning in 2028.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;HBM supply/demand is likely to remain tight through 2027. However, the projection of &lt;code&gt;2030 demand 26.7EB vs. supply 10.6EB, 2.52× deficit&lt;/code&gt; is not a base-case forecast — it requires multiple bullish assumptions to hold simultaneously.&lt;/li&gt;
&lt;li&gt;The most important year is not 2027 but &lt;strong&gt;2028&lt;/strong&gt;. Supply expansion from Samsung Electronics, SK Hynix, and Micron, inference efficiency gains, and refinancing verification across AI data centers all converge in the same window.&lt;/li&gt;
&lt;li&gt;HBM long-term supply contracts extend earnings duration but do not eliminate the cycle. They convert price risk into contract renewal risk, customer credit risk, product-mix risk, and repricing risk.&lt;/li&gt;
&lt;li&gt;Conditional probabilities: P1 sustained excess demand 35%, P2 localized credit stress with hyperscaler re-concentration 40%, P3 efficiency gains and supply normalization 15%, P4 systemic credit seizure 10%.&lt;/li&gt;
&lt;li&gt;Probability-weighted EPS for 2027: Samsung Electronics 60,750원, SK Hynix 393,000원. For 2028: Samsung 49,900원, SK Hynix 316,000원.&lt;/li&gt;
&lt;li&gt;Present value of 2028 terminal values discounted at 11% annually: Samsung Electronics 317,246원, SK Hynix 1,956,316원 — implying 24.7% and 6.0% upside respectively to the July 13 KRX closing prices. Once 2028 normalization risk is incorporated, Samsung&amp;rsquo;s margin of safety is wider; SK Hynix is more sensitive to whether P1 holds.&lt;/li&gt;
&lt;/ul&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;
 The 2027 boom is relatively visible. The harder question is &lt;strong&gt;whether high HBM pricing and earnings persist through 2028 and beyond&lt;/strong&gt;. Current prices discount the duration of earnings well beyond the 2027 peak.
 &lt;/div&gt;
&lt;/div&gt;
&lt;p&gt;&lt;img alt="Integrated 2027–2028 Scenario Map for Samsung Electronics and SK Hynix" loading="lazy" sizes="(max-width: 767px) calc(100vw - 30px), (max-width: 1023px) 700px, (max-width: 1279px) 950px, 1232px" src="https://koreainvestinsights.com/images/posts/samsung-hynix-2027-2028-scenario-map-2026-07-13.png"&gt;
&lt;/p&gt;
&lt;h2 id="1-analytical-questions-and-evidence-quality"&gt;1. Analytical Questions and Evidence Quality
&lt;/h2&gt;&lt;p&gt;This analysis ties three questions together in a single framework.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;How much confidence can we place in the view that HBM supply/demand will remain tight through 2027?&lt;/li&gt;
&lt;li&gt;If supply, efficiency, and financing conditions change simultaneously in 2028, how do Samsung Electronics&amp;rsquo; and SK Hynix&amp;rsquo;s earnings differ across scenarios?&lt;/li&gt;
&lt;li&gt;Given those changes, at what implied earnings level does the current share price trade?&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;The numbers in this report must be read by category.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Category&lt;/th&gt;
 &lt;th&gt;Definition&lt;/th&gt;
 &lt;th&gt;Example in This Report&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;Verified from public filings or market data&lt;/td&gt;
 &lt;td&gt;July 13 KRX closing prices, Oracle FCF, CoreWeave borrowings&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Consensus&lt;/td&gt;
 &lt;td&gt;Aggregated market estimates&lt;/td&gt;
 &lt;td&gt;2027–2028 EPS and net income&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Broker estimate&lt;/td&gt;
 &lt;td&gt;Individual broker forecast&lt;/td&gt;
 &lt;td&gt;Korea Investment &amp;amp; Securities and BNK Investment &amp;amp; Securities SK Hynix EPS&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Model output&lt;/td&gt;
 &lt;td&gt;Derived from disclosed formulas and assumptions&lt;/td&gt;
 &lt;td&gt;P1–P4 probability-weighted EPS and present values&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Analyst judgment&lt;/td&gt;
 &lt;td&gt;Non-statistical analytical judgment&lt;/td&gt;
 &lt;td&gt;Scenario probabilities, fair PER, 11% required return&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Blocked&lt;/td&gt;
 &lt;td&gt;Cannot be determined from public sources alone&lt;/td&gt;
 &lt;td&gt;HBM contract-level pricing, customer-by-customer volumes, actual production yield&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Core formulas are as follows.&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;Scenario Terminal Value = Scenario EPS × Scenario Fair PER
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;Probability-Weighted EPS = Σ(Scenario Probability × Scenario EPS)
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;Probability-Weighted Terminal Value = Σ(Scenario Probability × Scenario Terminal Value)
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;Present Value = Terminal Value ÷ (1 + 11%)^Remaining Years
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;HBM Need/Fleet = Installed and active HBM demand stock ÷ supply capacity that can be served by installed fleet
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;Scenario probabilities are not derived from historical frequency distributions. They are conditional judgments reflecting currently observable demand, supply, efficiency, and financing conditions. Fair PER values are likewise not mechanically observed from the market but are analytical estimates incorporating earnings durability, business concentration, financial structure, and long-term risk.&lt;/p&gt;
&lt;h2 id="2-current-price-and-market-consensus"&gt;2. Current Price and Market Consensus
&lt;/h2&gt;&lt;p&gt;July 13 KRX closing prices and consensus estimates aggregated on the same date are used throughout.&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;2026-07-13 KRX Closing Price&lt;/td&gt;
 &lt;td style="text-align: right"&gt;254,500원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1,845,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2027 Consensus EPS&lt;/td&gt;
 &lt;td style="text-align: right"&gt;65,802원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;444,359원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2028 Consensus EPS&lt;/td&gt;
 &lt;td style="text-align: right"&gt;65,907원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;433,625원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2027 Consensus Net Income&lt;/td&gt;
 &lt;td style="text-align: right"&gt;44.3조원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;32.4조원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2028 Consensus Net Income&lt;/td&gt;
 &lt;td style="text-align: right"&gt;44.1조원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;31.9조원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Current Price / 2027 Consensus EPS&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3.87×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4.15×&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Current Price / 2028 Consensus EPS&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3.86×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4.25×&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;On the surface, both companies look remarkably cheap. But these low PER multiples do not mean the market is unaware of 2027 earnings — they are more likely a signal that the market doubts the durability of earnings beyond 2028.&lt;/p&gt;
&lt;p&gt;The dispersion among SK Hynix estimates illustrates this clearly.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Source&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2027 EPS&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2028 EPS&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;Market Consensus&lt;/td&gt;
 &lt;td style="text-align: right"&gt;444,359원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;433,625원&lt;/td&gt;
 &lt;td&gt;Moderate 2028 normalization&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Korea Investment &amp;amp; Securities&lt;/td&gt;
 &lt;td style="text-align: right"&gt;415,254원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;495,696원&lt;/td&gt;
 &lt;td&gt;Earnings growth continues through 2028&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;BNK Investment &amp;amp; Securities&lt;/td&gt;
 &lt;td style="text-align: right"&gt;214,642원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;66,989원&lt;/td&gt;
 &lt;td&gt;Rapid peak passage after 2027&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;For the same company, 2028 EPS estimates range from 66,989원 to 495,696원. The divergence stems not from 2026 demand forecasts but from differing assumptions on &lt;strong&gt;2028 HBM average selling prices, the pace of supply expansion, and the duration of AI capex&lt;/strong&gt;.&lt;/p&gt;
&lt;p&gt;Recalculating current PER using this report&amp;rsquo;s probability-weighted EPS:&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;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;2027&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4.19×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4.69×&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2028&lt;/td&gt;
 &lt;td style="text-align: right"&gt;5.10×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;5.84×&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Applying the probability-weighted fair PER in reverse to the current share price, the implied EPS the market is pricing in is approximately 31,700원 for Samsung and approximately 245,000원 for SK Hynix — roughly 52% and 45% below the respective 2027 consensus. The market is discounting not 2027 earnings per se, but how long those earnings will last.&lt;/p&gt;
&lt;h2 id="3-what-has-changed-in-the-memory-cycle--and-what-hasnt"&gt;3. What Has Changed in the Memory Cycle — and What Hasn&amp;rsquo;t
&lt;/h2&gt;&lt;h3 id="what-has-changed"&gt;What Has Changed
&lt;/h3&gt;&lt;ol&gt;
&lt;li&gt;AI servers, HBM, and enterprise SSDs have raised the share of enterprise and data-center demand in the total mix.&lt;/li&gt;
&lt;li&gt;HBM requires customer qualification, advanced packaging, yield ramps, and long-term contracts, making short-run supply response slower than for commodity DRAM.&lt;/li&gt;
&lt;li&gt;As HBM absorbs a growing share of wafer starts, commodity DRAM output from the same fabs declines, which can support commodity DRAM pricing.&lt;/li&gt;
&lt;li&gt;Financing by large cloud providers and AI data center operators front-loads actual memory orders, acting as an accelerant.&lt;/li&gt;
&lt;li&gt;Conversely, when credit problems emerge, demand does not fade gradually — it can step down abruptly through contract renegotiations and investment deferrals.&lt;/li&gt;
&lt;/ol&gt;
&lt;h3 id="what-has-not-changed"&gt;What Has Not Changed
&lt;/h3&gt;&lt;ol&gt;
&lt;li&gt;High prices ultimately invite new fabs, yield improvement, process-node transitions, and competitor capacity additions.&lt;/li&gt;
&lt;li&gt;High HBM pricing and margins increase the incentive for a second-source supplier to enter and for customers to apply cost-reduction pressure.&lt;/li&gt;
&lt;li&gt;Long-term contracts do not eliminate price risk — they convert it into renewal pricing and customer credit risk.&lt;/li&gt;
&lt;li&gt;Simultaneously applying peak EPS to a high PER is a double-optimism error that counts both earnings and multiples at their most favorable.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;The claim that HBM has made memory a better business than legacy DRAM, and the claim that the memory cycle has been abolished, are not the same assertion.&lt;/p&gt;
&lt;h2 id="4-hbm-supply-demand-model-and-the-2028-inflection"&gt;4. HBM Supply-Demand Model and the 2028 Inflection
&lt;/h2&gt;&lt;p&gt;In the &lt;a class="link" href="https://koreainvestinsights.com/post/hbm-2030-supply-demand-267eb-demand-model-crosscheck-2026-07-13/" &gt;HBM 2030 Supply-Demand Model Cross-Check&lt;/a&gt;, we reproduced the original model&amp;rsquo;s &lt;code&gt;demand 26.7EB, supply 10.6EB, 2.52× deficit&lt;/code&gt; in consistent units. The formula itself does not incorrectly compare a demand flow against a supply stock. The issue is that long-horizon assumptions multiply together in series.&lt;/p&gt;
&lt;p&gt;These include:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Monthly token consumption growth&lt;/li&gt;
&lt;li&gt;Model size, context length, and agent state expansion&lt;/li&gt;
&lt;li&gt;KV cache and memory residency ratios&lt;/li&gt;
&lt;li&gt;Throughput, quantization, and routing efficiency&lt;/li&gt;
&lt;li&gt;HBM service life and migration of inference workloads to other memory types&lt;/li&gt;
&lt;li&gt;Achievable production volumes and packaging yield across the three major memory makers&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;For this reason, the 2030 2.52× deficit is best treated as a bullish stress scenario rather than a base case. Re-deriving central assumptions aligned to each of P1–P4 gives the following:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Path&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Token Multiplier&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Model Scale&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Throughput Efficiency&lt;/th&gt;
 &lt;th style="text-align: right"&gt;KV Efficiency&lt;/th&gt;
 &lt;th style="text-align: right"&gt;HBM Residency&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Demand&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Need/Fleet&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;P1 Orderly Excess Demand&lt;/td&gt;
 &lt;td style="text-align: right"&gt;20×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;12×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;5.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;65%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;16.1EB&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1.52×&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;P2 Localized Credit Stress &amp;amp; Re-concentration&lt;/td&gt;
 &lt;td style="text-align: right"&gt;18×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3.5×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;12×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;5.5×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;60%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;12.8EB&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1.21×&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;P3 Efficiency Gains &amp;amp; Supply Expansion&lt;/td&gt;
 &lt;td style="text-align: right"&gt;12×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;14×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;6.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;50%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;6.5EB&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0.62×&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;P4 Systemic Credit Seizure&lt;/td&gt;
 &lt;td style="text-align: right"&gt;8×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;14×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;7.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;45%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2.6EB&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0.25×&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;For comparability, supply capacity is held fixed at 10.6EB. Actual HBM wafer starts, product mix, good-die yield, and packaging capacity are non-public, so absolute values cannot be confirmed.&lt;/p&gt;
&lt;p&gt;Looking along the time axis, the judgment simplifies considerably.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Period&lt;/th&gt;
 &lt;th&gt;Assessment&lt;/th&gt;
 &lt;th&gt;Confidence&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;2026–2027&lt;/td&gt;
 &lt;td&gt;HBM4 transition and AI demand outpace new-fab supply additions&lt;/td&gt;
 &lt;td&gt;High&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2028&lt;/td&gt;
 &lt;td&gt;SK Hynix Yongin Line 1, Micron ID1/Tongluo, and Samsung HBM expansion begin easing supply&lt;/td&gt;
 &lt;td&gt;Medium-High&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2029–2030&lt;/td&gt;
 &lt;td&gt;Remaining deficit is more likely to narrow than to deepen&lt;/td&gt;
 &lt;td&gt;Medium&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2030 2.5× deficit&lt;/td&gt;
 &lt;td&gt;Requires multiple bullish assumptions to hold simultaneously&lt;/td&gt;
 &lt;td&gt;Low&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The investment decision does not therefore hinge on whether the 2027 boom occurs. The pivot is whether the additional supply coming online in 2028 clears customer qualifications, and whether that timing coincides with inference efficiency gains and tighter AI data center financing.&lt;/p&gt;
&lt;h2 id="5-sk-hynix-long-term-contracts-and-the-earnings-curve"&gt;5. SK Hynix Long-Term Contracts and the Earnings Curve
&lt;/h2&gt;&lt;p&gt;Korea Investment &amp;amp; Securities&amp;rsquo; July 13 SK Hynix report attributes the Q2 consensus miss not to demand collapse but to a timing lag in how HBM product-mix shifts and long-term supply contract pricing flow through to reported results.&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;th style="text-align: right"&gt;2028F&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;32.83조원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;48.77조원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;58.53조원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Operating Profit&lt;/td&gt;
 &lt;td style="text-align: right"&gt;24.51조원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;37.45조원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;44.70조원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;EPS&lt;/td&gt;
 &lt;td style="text-align: right"&gt;292,068원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;415,254원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;495,696원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;ROE&lt;/td&gt;
 &lt;td style="text-align: right"&gt;94.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;65.4%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;46.5%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;HBM assumptions are also aggressive.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;HBM Metric&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;Bit Growth&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+23.0%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+31.2%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;ASP&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$1.60/Gb&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$2.82/Gb&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;ASP Growth&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-2.5%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+76.3%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Revenue&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3.71조원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;8.30조원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Revenue Growth&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+28.0%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+123.6%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Key estimate adjustments within the report include:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;HBM4 sales begin in Q3 2026, but the largest earnings inflection is modeled in Q4 when HBM ASP rises &lt;code&gt;+52% QoQ&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;Compared to the May estimates, non-HBM DRAM ASP for 2026–2027 was trimmed by approximately 16%.&lt;/li&gt;
&lt;li&gt;2026 HBM revenue was raised by 17%.&lt;/li&gt;
&lt;li&gt;2027 HBM shipment volume was held flat while ASP was cut by 5%.&lt;/li&gt;
&lt;li&gt;NAND shipment volumes and pricing were raised.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;This report is not a retraction of the long-term HBM growth thesis. It is a revision that lowers short-run commodity DRAM price elasticity and pushes back the point at which HBM4 economics enter reported results.&lt;/p&gt;
&lt;h3 id="why-long-term-contracts-should-not-be-taken-at-face-value"&gt;Why Long-Term Contracts Should Not Be Taken at Face Value
&lt;/h3&gt;&lt;ol&gt;
&lt;li&gt;Contract-level price floors, ceilings, minimum purchase obligations, and prepayment terms are not publicly disclosed.&lt;/li&gt;
&lt;li&gt;Reports that contracts carry &amp;ldquo;no price ceiling&amp;rdquo; cannot by themselves confirm that prices rise every year — one must know whether pricing is linked to spot rates and how often it is reset.&lt;/li&gt;
&lt;li&gt;Long-term contracts lock in volumes and customer relationships but do not eliminate the risk of customer renegotiation, default, or credit deterioration.&lt;/li&gt;
&lt;li&gt;Non-HBM DRAM and NAND outside those contracts remain fully exposed to the commodity price cycle.&lt;/li&gt;
&lt;li&gt;If HBM4 yield is low, SK Hynix&amp;rsquo;s shipments and incremental earnings may still fall short even if the market remains undersupplied.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;The largest benefit of long-term contracts is not cycle elimination but &lt;strong&gt;extending earnings duration and converting the form of risk&lt;/strong&gt;.&lt;/p&gt;
&lt;h2 id="6-how-ai-infrastructure-financing-flows-into-memory-orders"&gt;6. How AI Infrastructure Financing Flows Into Memory Orders
&lt;/h2&gt;&lt;p&gt;AI memory demand is not driven purely by technology pull. Data center operators&amp;rsquo; financing activity creates front-loaded orders, and when financial conditions tighten, order gaps can emerge sharply.&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;Rising AI Expectations
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; Cloud/AI data center investment and long-term contract expansion
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; Front-loaded capex in GPU, HBM, power, and data centers
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; Asset, revenue, and enterprise value appreciation
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; Additional debt and equity financing
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; Larger capex programs
&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;Reverse direction
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;Monetization delay or higher refinancing costs
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; Contract renegotiation / expansion delays
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; Memory order gaps
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; ASP and share price decline
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; Collateral value and financing conditions deteriorate
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; Further capex reduction
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;The numbers confirmed from public disclosures are substantial.&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;Confirmed Value&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;Oracle FY2026 Free Cash Flow&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-$23.7B&lt;/td&gt;
 &lt;td&gt;Large-scale AI capex exceeds cash generation&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Oracle Debt Financing&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$43B&lt;/td&gt;
 &lt;td&gt;Heavy external funding for AI investment&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Oracle Equity Financing&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$5B&lt;/td&gt;
 &lt;td&gt;Debt alone is insufficient to cover investment&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Oracle RPO&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$638B&lt;/td&gt;
 &lt;td&gt;Long-term contracts provide high revenue visibility&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Oracle Customer Prepayments &amp;amp; Direct GPU Provision&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$75B&lt;/td&gt;
 &lt;td&gt;End customers share investment risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CoreWeave Total Debt (End 2025)&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$21.6B&lt;/td&gt;
 &lt;td&gt;High financial leverage in AI data centers&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CoreWeave New DDTL 4.0&lt;/td&gt;
 &lt;td style="text-align: right"&gt;$8.5B&lt;/td&gt;
 &lt;td&gt;Additional financing to continue expansion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CoreWeave DSCR Threshold&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Minimum 1.15×&lt;/td&gt;
 &lt;td&gt;Full refinancing verification begins no later than after June 30, 2027&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;CoreWeave&amp;rsquo;s end-2025 detailed borrowings comprised DDTL 2.0 approximately $5.04B, DDTL 2.1 approximately $2.74B, DDTL 1.0 approximately $1.55B, DDTL 3.0 approximately $0.34B, and equipment/software finance leases approximately $4.17B. DDTL 4.0, signed in March 2026, totals $8.5B.&lt;/p&gt;
&lt;p&gt;These figures do not imply that AI demand is fictitious. The contracts, GPUs, power infrastructure, and data centers are real. However, if cash flow fails to keep pace with financing costs in 2027–2028, weaker operators are likely to sell assets to be absorbed by large cloud providers.&lt;/p&gt;
&lt;p&gt;This is why the probability assigned to &lt;strong&gt;P2 — localized stress followed by asset re-concentration among hyperscalers&lt;/strong&gt; — exceeds that of P4 systemic collapse. In P2, memory orders do not disappear but a gap of one or more quarters is plausible.&lt;/p&gt;
&lt;h2 id="7-p1p4-conditional-probability-tree"&gt;7. P1–P4 Conditional Probability Tree
&lt;/h2&gt;&lt;p&gt;Final probabilities are derived by sequencing &lt;code&gt;real demand → financial verification → supply/efficiency → asset absorption&lt;/code&gt;, not from historical memory cycle frequencies.&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;2028 AI Monetization and Refinancing Verification
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; Pass 50%
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; ├─ P1 Sustained excess demand 70% = 35%
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; └─ P3 Efficiency/supply normalization 30% = 15%
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; Stress 50%
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; ├─ P2 Localized stress/re-concentration 80% = 40%
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; └─ P4 Systemic credit seizure 20% = 10%
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;The 50/50 first branch is not a statistical prior. It is a neutral starting point reflecting the fact that real demand and contracts are strong but financial verification is not yet complete. Calculating probabilities to single percentage-point precision from current data would produce false precision — adjustments are made only in 5-percentage-point increments.&lt;/p&gt;
&lt;h3 id="evidence-supporting-verification-pass-vs-financial-stress"&gt;Evidence Supporting Verification Pass vs. Financial Stress
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Evidence Supporting Verification Pass&lt;/th&gt;
 &lt;th&gt;Evidence Supporting Financial Stress&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM mass-production ramp timing lag through 2027&lt;/td&gt;
 &lt;td&gt;Oracle FY2026 FCF -$23.7B, debt financing $43B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Oracle RPO $638B, customer prepayments and direct GPU provision $75B&lt;/td&gt;
 &lt;td&gt;CoreWeave total debt $21.6B, new DDTL $8.5B&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM long-term contracts and prepayments raise order visibility&lt;/td&gt;
 &lt;td&gt;CoreWeave DSCR 1.15× verification after mid-2027&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Large cloud providers&amp;rsquo; actual data center investment&lt;/td&gt;
 &lt;td&gt;Rising risk premium on long-dated AI bonds&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The 70/30 split between P1 and P3 on the verification-pass branch reflects the fact that through 2027, qualification, yield, and packaging bottlenecks remain, while memory-side MoE, MLA, KV compression, routing, and offloading can all scale simultaneously from 2028 onward.&lt;/p&gt;
&lt;p&gt;The 80/20 split between P2 and P4 on the stress branch is also clear. GPUs and data centers are not assets that disappear — hyperscalers with strong cash flow can acquire and redeploy them. Because simultaneous capex pullbacks, multiple covenant breaches, and HBM long-term contract impairments have not yet materialized together, P4 is treated as a tail risk at 10%.&lt;/p&gt;
&lt;h3 id="evidence-ledger"&gt;Evidence Ledger
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Evidence&lt;/th&gt;
 &lt;th style="text-align: right"&gt;P1&lt;/th&gt;
 &lt;th style="text-align: right"&gt;P2&lt;/th&gt;
 &lt;th style="text-align: right"&gt;P3&lt;/th&gt;
 &lt;th style="text-align: right"&gt;P4&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;HBM supply lag through 2027&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;-&lt;/td&gt;
 &lt;td style="text-align: right"&gt;&amp;ndash;&lt;/td&gt;
 &lt;td&gt;Shortage persists&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Top-3 memory supply expansion post-2028&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0&lt;/td&gt;
 &lt;td style="text-align: right"&gt;++&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0&lt;/td&gt;
 &lt;td&gt;P3 independent pathway&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CXMT client DDR5/LPDDR expansion&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;++&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0&lt;/td&gt;
 &lt;td&gt;Caps commodity price ceiling before HBM&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;YMTC high-layer NAND ramp and customer adoption&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;++&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0&lt;/td&gt;
 &lt;td&gt;Pressure on Samsung NAND and enterprise SSD mix first&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;China HBM/server memory customer qualification&lt;/td&gt;
 &lt;td style="text-align: right"&gt;&amp;ndash;&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;0&lt;/td&gt;
 &lt;td&gt;If confirmed, shifts from P1 to P3&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Oracle RPO &amp;amp; customer-provided GPU&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;0&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-&lt;/td&gt;
 &lt;td&gt;Demand is real; buffers contagion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Oracle negative FCF &amp;amp; large financing&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;0&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+&lt;/td&gt;
 &lt;td&gt;Localized credit stress risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CoreWeave debt &amp;amp; DSCR verification&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;0&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+&lt;/td&gt;
 &lt;td&gt;2027–2028 financial inflection&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Rising AI long-dated bond risk premium&lt;/td&gt;
 &lt;td style="text-align: right"&gt;&amp;ndash;&lt;/td&gt;
 &lt;td style="text-align: right"&gt;++&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+&lt;/td&gt;
 &lt;td&gt;Shifts 5pp from P1 to P2&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Inference efficiency gains &amp;amp; offloading&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0&lt;/td&gt;
 &lt;td style="text-align: right"&gt;++&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0&lt;/td&gt;
 &lt;td&gt;Still insufficient production-level evidence&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Hyperscalers&amp;rsquo; asset absorption capacity&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0&lt;/td&gt;
 &lt;td style="text-align: right"&gt;++&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0&lt;/td&gt;
 &lt;td style="text-align: right"&gt;&amp;ndash;&lt;/td&gt;
 &lt;td&gt;Why P2 outweighs P4&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;&lt;code&gt;++&lt;/code&gt; and &lt;code&gt;--&lt;/code&gt; denote direction only and are not statistical likelihood ratios.&lt;/p&gt;
&lt;h3 id="probability-revision-history"&gt;Probability Revision History
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Step&lt;/th&gt;
 &lt;th style="text-align: right"&gt;P1&lt;/th&gt;
 &lt;th style="text-align: right"&gt;P2&lt;/th&gt;
 &lt;th style="text-align: right"&gt;P3&lt;/th&gt;
 &lt;th style="text-align: right"&gt;P4&lt;/th&gt;
 &lt;th&gt;Rationale&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Initial&lt;/td&gt;
 &lt;td style="text-align: right"&gt;40%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;35%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;15%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;10%&lt;/td&gt;
 &lt;td&gt;Physical supply shortage dominance, localized financial risk, supply/efficiency path&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AI long-duration bond weakness reflected&lt;/td&gt;
 &lt;td style="text-align: right"&gt;35%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;40%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;15%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;10%&lt;/td&gt;
 &lt;td&gt;P1 to P2 shift of 5pp; no contagion evidence&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Hyperscaler capex review&lt;/td&gt;
 &lt;td style="text-align: right"&gt;35%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;40%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;15%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;10%&lt;/td&gt;
 &lt;td&gt;Demand execution and asset concentration effects offset&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AI long-term scenario review&lt;/td&gt;
 &lt;td style="text-align: right"&gt;35%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;40%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;15%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;10%&lt;/td&gt;
 &lt;td&gt;2027–2028 unchanged; only post-2030 P3 probability expands&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;China supply path review&lt;/td&gt;
 &lt;td style="text-align: right"&gt;35%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;40%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;15%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;10%&lt;/td&gt;
 &lt;td&gt;Commodity supply risk reflected; HBM customer qualification evidence lacking&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h3 id="probability-uncertainty-ranges-and-revision-rules"&gt;Probability Uncertainty Ranges and Revision Rules
&lt;/h3&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Path&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Central&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Uncertainty Range&lt;/th&gt;
 &lt;th&gt;Strong Signals That Would Shift the Central Estimate&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;P1&lt;/td&gt;
 &lt;td style="text-align: right"&gt;35%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;25–45%&lt;/td&gt;
 &lt;td&gt;Two or more hyperscalers raise capex guidance; financing costs stabilize; long-term contracts strengthened&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;P2&lt;/td&gt;
 &lt;td style="text-align: right"&gt;40%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;30–50%&lt;/td&gt;
 &lt;td&gt;Single weak-hand operator stress event with hyperscaler asset absorption&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;P3&lt;/td&gt;
 &lt;td style="text-align: right"&gt;15%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;10–25%&lt;/td&gt;
 &lt;td&gt;Top-3 or China supply expansion confirmed in volume; per-token HBM consumption declines&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;P4&lt;/td&gt;
 &lt;td style="text-align: right"&gt;10%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;5–15%&lt;/td&gt;
 &lt;td&gt;Multiple operator defaults, two or more capex cuts, and long-term contract impairments occur simultaneously&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;ul&gt;
&lt;li&gt;A single news article or anecdote warrants only a 0–2pp adjustment.&lt;/li&gt;
&lt;li&gt;A single strong signal from a company filing or earnings report warrants at most a 5pp adjustment.&lt;/li&gt;
&lt;li&gt;Two or more independent strong signals in the same direction warrant a 5–10pp shift.&lt;/li&gt;
&lt;li&gt;Localized credit stress moves probability from P1 to P2; contagion evidence moves it from P2 to P4; supply/efficiency evidence moves it from P1 to P3.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="8-final-scenarios-and-hbm-price-pressure"&gt;8. Final Scenarios and HBM Price Pressure
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Path&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Prob.&lt;/th&gt;
 &lt;th&gt;2028 State&lt;/th&gt;
 &lt;th style="text-align: right"&gt;2030 Need/Fleet&lt;/th&gt;
 &lt;th style="text-align: right"&gt;HBM Price Pressure&lt;/th&gt;
 &lt;th&gt;Key Outcome&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;P1 Orderly Excess Demand&lt;/td&gt;
 &lt;td style="text-align: right"&gt;35%&lt;/td&gt;
 &lt;td&gt;Tight or balanced&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1.3–1.8×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Flat to -10%&lt;/td&gt;
 &lt;td&gt;Contract price floor holds; HBM earnings plateau&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;P2 Localized Credit Stress &amp;amp; Re-concentration&lt;/td&gt;
 &lt;td style="text-align: right"&gt;40%&lt;/td&gt;
 &lt;td&gt;Order gap, modest supply slack&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0.9–1.3×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-15 to -30%&lt;/td&gt;
 &lt;td&gt;Weak operators exit; hyperscalers absorb assets&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;P3 Efficiency Gains &amp;amp; Supply Expansion&lt;/td&gt;
 &lt;td style="text-align: right"&gt;15%&lt;/td&gt;
 &lt;td&gt;Growing supply surplus&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0.6–0.9×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-20 to -35%&lt;/td&gt;
 &lt;td&gt;Token usage grows but HBM premium compresses&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;P4 Systemic Credit Seizure&lt;/td&gt;
 &lt;td style="text-align: right"&gt;10%&lt;/td&gt;
 &lt;td&gt;Capex cuts and inventory correction&lt;/td&gt;
 &lt;td style="text-align: right"&gt;0.3–0.6×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-40 to -55%&lt;/td&gt;
 &lt;td&gt;LT contract renegotiation, output cuts, capex cancellations, policy intervention&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Holding all other variables constant, shifting 10 percentage points out of P2 changes the 2028 terminal values as follows:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Probability Shift&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Samsung Terminal Value Change&lt;/th&gt;
 &lt;th style="text-align: right"&gt;SK Hynix Terminal Value Change&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;10pp from P2 to P1&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+21,800원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+227,000원&lt;/td&gt;
 &lt;td&gt;Higher probability of sustained undersupply&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;10pp from P2 to P3&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-7,500원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-70,000원&lt;/td&gt;
 &lt;td&gt;AI economy grows but HBM premium compresses&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;10pp from P2 to P4&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-16,800원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-132,000원&lt;/td&gt;
 &lt;td&gt;Localized stress contagion to systemic risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;SK Hynix is more sensitive to every probability shift. Even using the same central probabilities, SK Hynix&amp;rsquo;s valuation error bars and share price volatility are larger due to its higher HBM concentration.&lt;/p&gt;
&lt;h2 id="9-china-supply-expansion-is-a-variable-across-all-paths-not-a-separate-scenario"&gt;9. China Supply Expansion Is a Variable Across All Paths, Not a Separate Scenario
&lt;/h2&gt;&lt;p&gt;CXMT and YMTC capacity growth is not a fifth scenario independent of P1–P4. It is a supply variable that alters product-level pricing and market share within every path.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;China Progress Stage&lt;/th&gt;
 &lt;th&gt;Market Affected First&lt;/th&gt;
 &lt;th&gt;Samsung Electronics&lt;/th&gt;
 &lt;th&gt;SK Hynix&lt;/th&gt;
 &lt;th&gt;Scenario Mapping&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;CXMT domestic DDR4/DDR5/LPDDR expansion&lt;/td&gt;
 &lt;td&gt;Commodity DRAM&lt;/td&gt;
 &lt;td&gt;ASP and product mix pressure&lt;/td&gt;
 &lt;td&gt;Relatively limited direct impact&lt;/td&gt;
 &lt;td&gt;EPS trimmed across all paths; probabilities unchanged&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;CXMT global PC customer qualification&lt;/td&gt;
 &lt;td&gt;Global client DDR5/LPDDR&lt;/td&gt;
 &lt;td&gt;ASP ceiling and share pressure simultaneously&lt;/td&gt;
 &lt;td&gt;Client exposure at risk&lt;/td&gt;
 &lt;td&gt;P3 upside risk&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;China low-end server RDIMM penetration&lt;/td&gt;
 &lt;td&gt;Standard server DRAM&lt;/td&gt;
 &lt;td&gt;Partial server mix pressure&lt;/td&gt;
 &lt;td&gt;DRAM outside AI servers at risk&lt;/td&gt;
 &lt;td&gt;P3 upside review&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;YMTC high-layer NAND ramp and customer adoption&lt;/td&gt;
 &lt;td&gt;Consumer NAND/enterprise SSD&lt;/td&gt;
 &lt;td&gt;Largest NAND share and margin pressure&lt;/td&gt;
 &lt;td&gt;Solidigm and enterprise SSD at risk&lt;/td&gt;
 &lt;td&gt;P3 EPS and multiple reduction&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;China HBM mass production and AI customer qualification&lt;/td&gt;
 &lt;td&gt;HBM&lt;/td&gt;
 &lt;td&gt;Partial HBM catch-up offset&lt;/td&gt;
 &lt;td&gt;Large scarcity premium and share risk&lt;/td&gt;
 &lt;td&gt;5–10pp shift from P1 to P3&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;China ramp and global AI capex slowdown concurrent&lt;/td&gt;
 &lt;td&gt;All memory&lt;/td&gt;
 &lt;td&gt;Commodity/NAND/HBM combined pressure&lt;/td&gt;
 &lt;td&gt;HBM/DRAM combined pressure&lt;/td&gt;
 &lt;td&gt;Monitor for P3-to-P4 contagion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;In 2026–2027, CXMT is more likely to cap the client DDR5 and LPDDR price ceiling than to directly undermine HBM pricing. YMTC may similarly intensify price competition in consumer NAND and some enterprise SSDs first.&lt;/p&gt;
&lt;p&gt;For China to emerge as a direct HBM threat, DRAM die production alone is insufficient. TSV bonding, stacking, base-die integration, advanced packaging, and repeated mass-production qualifications by AI customers are all required.&lt;/p&gt;
&lt;p&gt;The following signals must be confirmed before Chinese capacity expansion is incorporated numerically in meaningful quantities:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;CXMT DDR5/LPDDR is adopted at volume by global PC makers at prices more than 20% below those of the top-3 memory suppliers.&lt;/li&gt;
&lt;li&gt;CXMT secures qualification beyond consumer PCs into enterprise PCs and server RDIMM.&lt;/li&gt;
&lt;li&gt;YMTC high-layer NAND is confirmed by actual shipment volumes and enterprise SSD customer wins — not just stated targets.&lt;/li&gt;
&lt;li&gt;China HBM moves beyond sample qualification to repeated mass-production qualification by AI accelerator customers.&lt;/li&gt;
&lt;li&gt;Supply growth and commodity DRAM/NAND price declines among the top-3 suppliers are observed together for two or more consecutive quarters.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id="10-samsung-electronics-eps-and-fair-value-by-scenario"&gt;10. Samsung Electronics: EPS and Fair Value by Scenario
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Year&lt;/th&gt;
 &lt;th&gt;Scenario&lt;/th&gt;
 &lt;th style="text-align: right"&gt;EPS&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Fair PER&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Terminal Value&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;2027&lt;/td&gt;
 &lt;td&gt;P1&lt;/td&gt;
 &lt;td style="text-align: right"&gt;65,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;8.5×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;552,500원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2027&lt;/td&gt;
 &lt;td&gt;P2&lt;/td&gt;
 &lt;td style="text-align: right"&gt;62,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;8.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;496,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2027&lt;/td&gt;
 &lt;td&gt;P3&lt;/td&gt;
 &lt;td style="text-align: right"&gt;58,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;7.5×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;435,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2027&lt;/td&gt;
 &lt;td&gt;P4&lt;/td&gt;
 &lt;td style="text-align: right"&gt;45,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;7.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;315,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2028&lt;/td&gt;
 &lt;td&gt;P1&lt;/td&gt;
 &lt;td style="text-align: right"&gt;68,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;8.5×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;578,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2028&lt;/td&gt;
 &lt;td&gt;P2&lt;/td&gt;
 &lt;td style="text-align: right"&gt;45,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;8.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;360,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2028&lt;/td&gt;
 &lt;td&gt;P3&lt;/td&gt;
 &lt;td style="text-align: right"&gt;38,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;7.5×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;285,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2028&lt;/td&gt;
 &lt;td&gt;P4&lt;/td&gt;
 &lt;td style="text-align: right"&gt;24,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;8.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;192,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;A base PER above 9× is not assigned to Samsung because foundry loss reduction has not yet been confirmed. An 8.5× P1 multiple becomes achievable only if HBM share recovery, commodity DRAM/NAND pricing, and foundry normalization all operate together.&lt;/p&gt;
&lt;p&gt;An 8.0× multiple in P4 is applied to avoid the double-discount error of multiplying a trough EPS by an also-depressed PER. In an actual recession scenario, normalized EPS, net cash, and book value must all be considered together.&lt;/p&gt;
&lt;h2 id="11-sk-hynix-eps-and-fair-value-by-scenario"&gt;11. SK Hynix: EPS and Fair Value by Scenario
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Year&lt;/th&gt;
 &lt;th&gt;Scenario&lt;/th&gt;
 &lt;th style="text-align: right"&gt;EPS&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Fair PER&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Terminal Value&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;2027&lt;/td&gt;
 &lt;td&gt;P1&lt;/td&gt;
 &lt;td style="text-align: right"&gt;430,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;9.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;3,870,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2027&lt;/td&gt;
 &lt;td&gt;P2&lt;/td&gt;
 &lt;td style="text-align: right"&gt;405,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;7.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2,835,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2027&lt;/td&gt;
 &lt;td&gt;P3&lt;/td&gt;
 &lt;td style="text-align: right"&gt;370,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;6.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2,220,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2027&lt;/td&gt;
 &lt;td&gt;P4&lt;/td&gt;
 &lt;td style="text-align: right"&gt;250,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;5.5×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1,375,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2028&lt;/td&gt;
 &lt;td&gt;P1&lt;/td&gt;
 &lt;td style="text-align: right"&gt;470,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;9.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;4,230,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2028&lt;/td&gt;
 &lt;td&gt;P2&lt;/td&gt;
 &lt;td style="text-align: right"&gt;280,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;7.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1,960,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2028&lt;/td&gt;
 &lt;td&gt;P3&lt;/td&gt;
 &lt;td style="text-align: right"&gt;210,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;6.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1,260,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2028&lt;/td&gt;
 &lt;td&gt;P4&lt;/td&gt;
 &lt;td style="text-align: right"&gt;80,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;8.0×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;640,000원&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The 9× P1 multiple presupposes HBM market share, long-term contract terms, high ROE, and a structural rerating of SK Hynix as a strategic AI asset. Maintaining this multiple through 2028 requires confirmed HBM4E share defense, long-term contract repricing, and shareholder return activity.&lt;/p&gt;
&lt;p&gt;Korea Investment &amp;amp; Securities&amp;rsquo; target price of 3,800,000원 applies a 6.0× PBR to 12-month forward BPS — above the historical band peak of 5.0× cited in the same report. A 3,800,000원 target therefore requires not just earnings delivery but a structural rerating of the memory business.&lt;/p&gt;
&lt;p&gt;In P2 and P3, HBM concentration shifts from an advantage to a vulnerability. When ASP falls, earnings and multiples can compress simultaneously.&lt;/p&gt;
&lt;h2 id="12-probability-weighted-results-and-risk-adjusted-comparison"&gt;12. Probability-Weighted Results and Risk-Adjusted Comparison
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Year&lt;/th&gt;
 &lt;th&gt;Company&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Prob.-Weighted EPS&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Blended Fair PER&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Prob.-Weighted Terminal Value&lt;/th&gt;
 &lt;th style="text-align: right"&gt;PV at 11% Annual Discount&lt;/th&gt;
 &lt;th style="text-align: right"&gt;vs. Current Price&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;2027&lt;/td&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td style="text-align: right"&gt;60,750원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;8.04×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;488,525원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;421,385원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+65.6%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2027&lt;/td&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td style="text-align: right"&gt;393,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;7.53×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2,959,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2,552,333원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+38.3%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2028&lt;/td&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td style="text-align: right"&gt;49,900원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;8.18×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;408,250원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;317,246원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+24.7%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;2028&lt;/td&gt;
 &lt;td&gt;SK Hynix&lt;/td&gt;
 &lt;td style="text-align: right"&gt;316,000원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;7.97×&lt;/td&gt;
 &lt;td style="text-align: right"&gt;2,517,500원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;1,956,316원&lt;/td&gt;
 &lt;td style="text-align: right"&gt;+6.0%&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Four observations follow from these results:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Viewed through 2027 alone, both companies show risk-adjusted values above current prices.&lt;/li&gt;
&lt;li&gt;Once 2028 normalization probability is discounted, Samsung&amp;rsquo;s margin of safety is wider.&lt;/li&gt;
&lt;li&gt;SK Hynix&amp;rsquo;s 2028 present value is only 6% above the current share price. A large portion of expected return depends on P1 holding.&lt;/li&gt;
&lt;li&gt;Samsung&amp;rsquo;s commodity memory, foundry, and net cash position provide a partial downside buffer in P2 and P3.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;What is already well known to the market is HBM undersupply and higher memory pricing. What may be underpriced in two directions:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;If Samsung&amp;rsquo;s HBM catch-up coincides with commodity DRAM/NAND price recovery, both earnings and multiples could improve together.&lt;/li&gt;
&lt;li&gt;If SK Hynix&amp;rsquo;s long-term contracts and HBM4E defend price floors and share even after 2028 supply additions, the current normalization discount may be excessive.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Conversely, the market may prove correct if supply, efficiency, and credit conditions all improve simultaneously in 2028, cutting peak earnings short.&lt;/p&gt;
&lt;h2 id="13-how-to-think-about-ai-long-term-demand-beyond-2030"&gt;13. How to Think About AI Long-Term Demand Beyond 2030
&lt;/h2&gt;&lt;p&gt;&lt;a class="link" href="https://ai-2040.com/?choices=plan-a-root" target="_blank" rel="noopener"
 &gt;AI 2040 Plan A&lt;/a&gt; presents a path from approximately 20 million H100e in 2026 to 60 billion H100e by 2034, with AI compute power consumption reaching 5TW by 2034. However, the document itself describes this as a policy scenario, not a best-estimate forecast, and notes that certain variables were adjusted alongside the narrative.&lt;/p&gt;
&lt;p&gt;Long-term scenarios are therefore not fed directly into 2027–2028 EPS estimates.&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;P1 Excess Demand&lt;/th&gt;
 &lt;th style="text-align: right"&gt;P2 Re-concentration&lt;/th&gt;
 &lt;th style="text-align: right"&gt;P3 Volume Growth / Price Normalization&lt;/th&gt;
 &lt;th style="text-align: right"&gt;P4 Failure&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;2026–2028&lt;/td&gt;
 &lt;td style="text-align: right"&gt;35%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;40%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;15%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;10%&lt;/td&gt;
 &lt;td&gt;Current analysis probabilities&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Post-2030 conditional baseline&lt;/td&gt;
 &lt;td style="text-align: right"&gt;25%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;35%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;30%&lt;/td&gt;
 &lt;td style="text-align: right"&gt;10%&lt;/td&gt;
 &lt;td&gt;Long-term judgment; not a statistical probability&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;If AI and robotics raise the efficiency of chip design, fab construction, and production, total memory bit demand may grow while the scarcity premium on HBM specifically narrows. &lt;strong&gt;HBM market size expansion and long-term PER expansion must therefore be evaluated separately.&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;As long-term memory demand grows, the opportunity set expands not only for HBM but also for commodity DRAM, NAND, enterprise SSDs, controllers, CXL, power, cooling, and interconnects. This is a relative advantage for Samsung, with its broader product portfolio.&lt;/p&gt;
&lt;h2 id="14-what-needs-to-materialize-for-this-analysis-to-hold"&gt;14. What Needs to Materialize for This Analysis to Hold
&lt;/h2&gt;&lt;h3 id="samsung-electronics"&gt;Samsung Electronics
&lt;/h3&gt;&lt;ol&gt;
&lt;li&gt;HBM4 and HBM4E customer qualifications must convert into actual mass-production volumes and revenue share.&lt;/li&gt;
&lt;li&gt;Commodity DRAM and NAND contract price recovery must offset HBM investment costs, bonuses, and provisions.&lt;/li&gt;
&lt;li&gt;Foundry losses must narrow and begin making a meaningful contribution to 2027–2028 earnings.&lt;/li&gt;
&lt;li&gt;Net cash and shareholder return capacity must be maintained despite heavy capex.&lt;/li&gt;
&lt;/ol&gt;
&lt;h3 id="sk-hynix"&gt;SK Hynix
&lt;/h3&gt;&lt;ol&gt;
&lt;li&gt;HBM4E share and long-term contract price floors must hold even as Samsung and Micron expand supply.&lt;/li&gt;
&lt;li&gt;HBM repricing beginning in Q4 2026 must track toward the 2027 average of $2.82/Gb.&lt;/li&gt;
&lt;li&gt;TSV and stacking yield weakness must prove temporary and attributable to product transition costs.&lt;/li&gt;
&lt;li&gt;High free cash flow must flow toward shareholder returns and capital policy, not just capacity expansion.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id="15-risk-signals-and-leading-indicators"&gt;15. Risk Signals and Leading Indicators
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Risk&lt;/th&gt;
 &lt;th&gt;Leading Indicator&lt;/th&gt;
 &lt;th&gt;Samsung Impact&lt;/th&gt;
 &lt;th&gt;SK Hynix Impact&lt;/th&gt;
 &lt;th&gt;Analysis Adjustment&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;HBM long-term contract renegotiation&lt;/td&gt;
 &lt;td&gt;Price floor, prepayment, and minimum purchase weakening&lt;/td&gt;
 &lt;td&gt;Moderate&lt;/td&gt;
 &lt;td&gt;Very large&lt;/td&gt;
 &lt;td&gt;P2/P4 up&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Hyperscaler capex guidance cut&lt;/td&gt;
 &lt;td&gt;Two or more simultaneous guidance cuts&lt;/td&gt;
 &lt;td&gt;Large&lt;/td&gt;
 &lt;td&gt;Very large&lt;/td&gt;
 &lt;td&gt;P2/P4 up&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AI data center refinancing event&lt;/td&gt;
 &lt;td&gt;CoreWeave covenant terms/financing cost, Oracle funding&lt;/td&gt;
 &lt;td&gt;Moderate&lt;/td&gt;
 &lt;td&gt;Large&lt;/td&gt;
 &lt;td&gt;P2 up; P4 up if contagion&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Accelerated supply expansion&lt;/td&gt;
 &lt;td&gt;Samsung/Micron volume-capable HBM output rises&lt;/td&gt;
 &lt;td&gt;Positive for HBM catch-up; negative for pricing&lt;/td&gt;
 &lt;td&gt;Very negative&lt;/td&gt;
 &lt;td&gt;P3 up&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Sharp inference efficiency improvement&lt;/td&gt;
 &lt;td&gt;Per-token HBM bits decline&lt;/td&gt;
 &lt;td&gt;Broad product portfolio provides partial buffer&lt;/td&gt;
 &lt;td&gt;Negative&lt;/td&gt;
 &lt;td&gt;HBM multiple lower&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;China commodity supply expansion&lt;/td&gt;
 &lt;td&gt;CXMT/YMTC pricing, share, global customer qualifications&lt;/td&gt;
 &lt;td&gt;More negative for commodity DRAM/NAND&lt;/td&gt;
 &lt;td&gt;Client DRAM and Solidigm under pressure&lt;/td&gt;
 &lt;td&gt;Product-level EPS reduction&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;China HBM challenge&lt;/td&gt;
 &lt;td&gt;TSV/stacking/base die/packaging/customer qualification&lt;/td&gt;
 &lt;td&gt;Partial offset of HBM catch-up value&lt;/td&gt;
 &lt;td&gt;Very negative for scarcity premium&lt;/td&gt;
 &lt;td&gt;P1 to P3 shift&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Foundry normalization failure&lt;/td&gt;
 &lt;td&gt;Continued losses, no yield improvement&lt;/td&gt;
 &lt;td&gt;Samsung-specific risk&lt;/td&gt;
 &lt;td&gt;Limited impact&lt;/td&gt;
 &lt;td&gt;Samsung P1 multiple reduction&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h3 id="upside-signals"&gt;Upside Signals
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;2028 HBM long-term contracts with higher volumes and prices and broader customer diversification&lt;/li&gt;
&lt;li&gt;Early stabilization of HBM4E yield and mass-production volumes&lt;/li&gt;
&lt;li&gt;Simultaneous confirmation of Samsung&amp;rsquo;s rising HBM share and narrowing foundry losses&lt;/li&gt;
&lt;li&gt;AI service revenue and cash flow growing faster than compute resource contracts&lt;/li&gt;
&lt;li&gt;HBM lead times and allocations hold even as supply expansion lags&lt;/li&gt;
&lt;li&gt;SK Hynix shareholder return plan becomes concrete&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id="downside-signals"&gt;Downside Signals
&lt;/h3&gt;&lt;ul&gt;
&lt;li&gt;HBM ASP decline and shipment volume drop occur simultaneously&lt;/li&gt;
&lt;li&gt;Two or more major hyperscalers lower capex growth rates&lt;/li&gt;
&lt;li&gt;Long-term contract renegotiation, breach, or prepayment impairment&lt;/li&gt;
&lt;li&gt;Memory inventory build and spot-to-contract price inversion&lt;/li&gt;
&lt;li&gt;Actual AI service per-token HBM consumption falls 20%+ for two consecutive quarters&lt;/li&gt;
&lt;li&gt;Samsung/Micron volume supply increase coincides with HBM premium decline&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;On a monthly basis, the indicators to track together are: HBM4/4E contract prices and prepayments, top-3 memory producer volumes and yields, advanced packaging lead times, hyperscaler capex and cash flow, Oracle and CoreWeave financing costs, Taiwan server ODM revenue, and DRAM/NAND contract prices and inventory.&lt;/p&gt;
&lt;h2 id="16-relative-investment-interpretation"&gt;16. Relative Investment Interpretation
&lt;/h2&gt;&lt;p&gt;The two names are not equivalent ways to buy the same memory upcycle.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Scenario&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;P1 Strengthening&lt;/td&gt;
 &lt;td&gt;HBM catch-up and commodity memory tailwind&lt;/td&gt;
 &lt;td&gt;Largest earnings leverage from HBM share and LT contracts&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;P2 Warning&lt;/td&gt;
 &lt;td&gt;Business diversification and net cash provide partial defense&lt;/td&gt;
 &lt;td&gt;More sensitive to order gaps and multiple compression&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;P3 Strengthening&lt;/td&gt;
 &lt;td&gt;Broad memory product range and foundry execution provide partial buffer&lt;/td&gt;
 &lt;td&gt;Direct exposure to HBM premium compression&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;P4 Warning&lt;/td&gt;
 &lt;td&gt;Absolute decline unavoidable but relative defense possible&lt;/td&gt;
 &lt;td&gt;Greatest risk of simultaneous earnings and multiple decline&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;From a risk-adjusted standpoint, Samsung offers a wider and more defensive choice. SK Hynix has greater earnings leverage in P1, but its valuation shifts faster when supply, efficiency, and credit conditions change.&lt;/p&gt;
&lt;p&gt;Equipment and materials names also should not be grouped together indiscriminately. Equipment companies exposed to order gaps must be separated from consumable and parts suppliers whose revenues track utilization rates. Equipment with confirmed purchase orders and components/materials with recurring revenue streams are a better fit for this scenario framework than a pure HBM theme basket.&lt;/p&gt;
&lt;h2 id="17-conclusion"&gt;17. Conclusion
&lt;/h2&gt;&lt;p&gt;The memory boom through 2027 can be maintained as the central path for earnings estimates. But the real test of current share prices is not 2027 EPS — it is the duration of earnings beyond 2028. The claim that the HBM supply deficit has lasted longer than past cycles, and the claim that memory companies deserve a permanently elevated scarcity multiple, are not the same argument.&lt;/p&gt;
&lt;p&gt;Samsung Electronics holds HBM catch-up potential, commodity DRAM/NAND, foundry operations, and net cash simultaneously, offering P1 upside and relative downside defense across P2–P4. SK Hynix&amp;rsquo;s HBM share and long-term contracts provide the highest earnings leverage in P1, but if supply, efficiency, and financing conditions normalize, earnings and multiples can compress together.&lt;/p&gt;
&lt;p&gt;The current probability-weighted results say: &lt;strong&gt;through 2027, both companies offer meaningful upside; looking through to 2028, Samsung&amp;rsquo;s risk-adjusted margin of safety is wider.&lt;/strong&gt; For SK Hynix to see further rerating, confirmation is needed that HBM4E yield, 2028 long-term contract terms, and AI infrastructure financing continue to support P1.&lt;/p&gt;
&lt;h2 id="items-that-cannot-yet-be-confirmed-from-public-sources"&gt;Items That Cannot Yet Be Confirmed from Public Sources
&lt;/h2&gt;&lt;ol&gt;
&lt;li&gt;Samsung Electronics&amp;rsquo; segment-level EPS for 2027–2028 and foundry normalization sensitivity&lt;/li&gt;
&lt;li&gt;SK Hynix long-term contract customer weighting, repricing formula, and prepayment/minimum purchase terms&lt;/li&gt;
&lt;li&gt;Top-3 memory makers&amp;rsquo; HBM-dedicated wafer input, good-die and stacking yield, and customer-specific producible volumes&lt;/li&gt;
&lt;li&gt;Composition of 2028 hyperscaler capex between internal cash flow, corporate bonds, and project finance&lt;/li&gt;
&lt;li&gt;CoreWeave and Oracle refinancing schedules and covenant headroom&lt;/li&gt;
&lt;li&gt;Per-generation HBM4E capacity, bandwidth, and memory residency rate changes per accelerator&lt;/li&gt;
&lt;li&gt;CXMT/YMTC actual shipment volumes, global customer qualifications, and product-level earnings sensitivity&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id="key-sources"&gt;Key Sources
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://vo.la/fd9udR9" target="_blank" rel="noopener"
 &gt;Korea Investment &amp;amp; Securities, SK Hynix 2Q26 Preview, 2026-07-13&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://vo.la/jBPy7zV" target="_blank" rel="noopener"
 &gt;Korea Investment &amp;amp; Securities, &amp;ldquo;The Start Begins Now,&amp;rdquo; 2026-05-20&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://koreainvestinsights.com/post/hbm-2030-supply-demand-267eb-demand-model-crosscheck-2026-07-13/" &gt;Korea Invest Insights, HBM 2030 Supply-Demand Model Cross-Check&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://ai-2040.com/?choices=plan-a-root" target="_blank" rel="noopener"
 &gt;AI 2040, Plan A&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://ai-2040.com/supplements/compute-supplement" target="_blank" rel="noopener"
 &gt;AI 2040, Compute Supplement&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://ai-2040.com/supplements/economics-of-plan-a" target="_blank" rel="noopener"
 &gt;AI 2040, Economics of Plan A&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.oracle.com/news/announcement/q4fy26-earnings-release-2026-06-10/" target="_blank" rel="noopener"
 &gt;Oracle FY2026 Earnings&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/1769628/000176962826000104/crwv-20251231.htm" target="_blank" rel="noopener"
 &gt;CoreWeave 2025 Form 10-K&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.sec.gov/Archives/edgar/data/1769628/000176962826000129/crwv-20260330.htm" target="_blank" rel="noopener"
 &gt;CoreWeave DDTL 4.0 Form 8-K&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The scenario probabilities and fair values in this report are analytical estimates based on publicly available information and do not constitute personalized investment advice. Actual investment decisions must separately account for price volatility, investment horizon, taxation, currency risk, and individual loss tolerance.&lt;/p&gt;</description></item><item><title>Inside KOSPI's 9.05% Selloff: AI Hardware Derating and Korea's Leverage Amplifier</title><link>https://koreainvestinsights.com/post/kospi-9pct-selloff-ai-hardware-derating-korea-leverage-2026-07-13/</link><pubDate>Mon, 13 Jul 2026 18:20:00 +0900</pubDate><guid>https://koreainvestinsights.com/post/kospi-9pct-selloff-ai-hardware-derating-korea-leverage-2026-07-13/</guid><description>
 &lt;blockquote&gt;
 &lt;p&gt;Context: this article follows our work on &lt;a class="link" href="https://koreainvestinsights.com/post/h1-2026-ai-infra-bottleneck-korea-narrow-market-postmortem-2026-06-30/" &gt;H1 2026 AI infrastructure bottlenecks and narrow leadership&lt;/a&gt;, &lt;a class="link" href="https://koreainvestinsights.com/post/samsung-earnings-selloff-nvidia-q4fy26-rebound-trigger-2026-07-08/" &gt;Samsung Electronics&amp;rsquo; post-earnings selloff and the NVIDIA precedent&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 and leveraged-ETF plumbing&lt;/a&gt;, and &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 scenarios for the memory thesis&lt;/a&gt;. The purpose is to separate long-term AI demand from short-term price mechanics.&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;h2 id="tldr"&gt;TL;DR
&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;On July 13, 2026, KOSPI closed at &lt;strong&gt;6,799.24, down 9.05%&lt;/strong&gt;. KOSDAQ fell 4.43% to 800.35. Samsung Electronics lost 9.21% to KRW 258,750 and SK hynix fell 13.35% to KRW 1,889,000.&lt;/li&gt;
&lt;li&gt;A Korea-only market-plumbing accident does not explain the full move. Japanese NAND, MLCC and FC-BGA names also sold off sharply. The best causal frame is that &lt;strong&gt;a global unwind in crowded AI and electronics hardware set the direction, while Korea&amp;rsquo;s leveraged ETFs, margin exposure and program trading magnified the decline&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;The evidence does not establish a collapse in physical AI demand. TSMC reported record June revenue, while early-July memory exports and long-term HBM supply evidence still point to strong demand. Strong demand, however, does not automatically protect an elevated earnings path or valuation multiple.&lt;/li&gt;
&lt;li&gt;The neutral reading is that deleveraging has started but completion is unconfirmed. A durable bottom requires fewer new lows, weaker program selling, a decline in leveraged-ETF and margin exposure, and improving relative strength in U.S. semiconductors over the next one to three trading days.&lt;/li&gt;
&lt;/ul&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;
 The selloff looks more like a simultaneous repricing of &lt;strong&gt;AI hardware profit duration and crowded positioning&lt;/strong&gt; than proof that AI demand has disappeared. Long-term business quality and short-term equity timing are different questions.
 &lt;/div&gt;
&lt;/div&gt;
&lt;p&gt;&lt;img alt="Korean market selloff and causal structure on July 13, 2026" loading="lazy" sizes="(max-width: 767px) calc(100vw - 30px), (max-width: 1023px) 700px, (max-width: 1279px) 950px, 1232px" src="https://koreainvestinsights.com/images/posts/kospi-9pct-selloff-ai-hardware-derating-2026-07-13.png"&gt;
&lt;/p&gt;
&lt;h2 id="1-the-closing-tape"&gt;1. The closing tape
&lt;/h2&gt;&lt;p&gt;Intraday screens showed KOSPI near 6,803, Samsung Electronics at KRW 254,500 and SK hynix at KRW 1,847,000. A market report should start with closing data rather than intraday lows.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Asset&lt;/th&gt;
 &lt;th style="text-align: right"&gt;July 13 close&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Daily return&lt;/th&gt;
 &lt;th&gt;Data type&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;6,799.24&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-9.05%&lt;/td&gt;
 &lt;td&gt;Naver Finance 15:30 close&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;KOSDAQ&lt;/td&gt;
 &lt;td style="text-align: right"&gt;800.35&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-4.43%&lt;/td&gt;
 &lt;td&gt;Naver Finance 15:30 close&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Samsung Electronics&lt;/td&gt;
 &lt;td style="text-align: right"&gt;KRW 258,750&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-9.21%&lt;/td&gt;
 &lt;td&gt;Kiwoom REST daily bar&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 1,889,000&lt;/td&gt;
 &lt;td style="text-align: right"&gt;-13.35%&lt;/td&gt;
 &lt;td&gt;Kiwoom REST daily bar&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The index closes can be checked on the &lt;a class="link" href="https://finance.naver.com/sise/sise_index.naver?code=KOSPI" target="_blank" rel="noopener"
 &gt;KOSPI&lt;/a&gt; and &lt;a class="link" href="https://finance.naver.com/sise/sise_index.naver?code=KOSDAQ" target="_blank" rel="noopener"
 &gt;KOSDAQ&lt;/a&gt; pages.&lt;/p&gt;
&lt;p&gt;Program trading played a direct role. The Kiwoom market proxy recorded approximately KRW 2.14 trillion of net program selling. Arbitrage flows were about KRW 172 billion net positive, but non-arbitrage flows were approximately KRW 2.31 trillion net negative. The dominant force was therefore broad basket reduction rather than a simple basis trade.&lt;/p&gt;
&lt;p&gt;These are Kiwoom REST market proxies, not final KRX official statistics. They are useful for direction and scale but should be superseded when official exchange data are available.&lt;/p&gt;
&lt;h2 id="2-a-multi-stage-causal-chain"&gt;2. A multi-stage causal chain
&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-text" data-lang="text"&gt;&lt;span style="display:flex;"&gt;&lt;span&gt;Middle East, oil, rates and AI-capex funding concerns
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; global reduction of crowded AI and electronics-hardware exposure
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; lower SK hynix 2026-2027 earnings estimates become Korea&amp;#39;s direct catalyst
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; selling spreads into Japanese NAND, MLCC and FC-BGA baskets
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; Korea&amp;#39;s index concentration, margin exposure, single-stock leveraged ETFs and program selling magnify losses
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; positive physical-demand evidence fails to lift prices, triggering further liquidation
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Factor&lt;/th&gt;
 &lt;th&gt;Role&lt;/th&gt;
 &lt;th style="text-align: right"&gt;Confidence&lt;/th&gt;
 &lt;th&gt;Neutral interpretation&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Global AI and electronics hardware unwind&lt;/td&gt;
 &lt;td&gt;Base direction&lt;/td&gt;
 &lt;td style="text-align: right"&gt;High&lt;/td&gt;
 &lt;td&gt;Japanese NAND, MLCC and FC-BGA names fell together, indicating a factor-level move beyond Korea.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Semiconductor leverage liquidation&lt;/td&gt;
 &lt;td&gt;Korean amplifier&lt;/td&gt;
 &lt;td style="text-align: right"&gt;High&lt;/td&gt;
 &lt;td&gt;Large-cap program selling and leveraged-ETF rebalancing fit the observed nonlinear decline.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SK hynix earnings revision&lt;/td&gt;
 &lt;td&gt;Direct catalyst&lt;/td&gt;
 &lt;td style="text-align: right"&gt;High&lt;/td&gt;
 &lt;td&gt;Long-term demand remained constructive, but near-term earnings and the 2026-2027 trajectory were reduced.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Oil, rates and geopolitical risk&lt;/td&gt;
 &lt;td&gt;Macro trigger&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Medium-high&lt;/td&gt;
 &lt;td&gt;These explain higher discount rates and risk aversion but not Korea&amp;rsquo;s full relative loss.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;AI capex funding cost&lt;/td&gt;
 &lt;td&gt;Multiple pressure&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Medium&lt;/td&gt;
 &lt;td&gt;Capital costs may separate cash-rich investors from more leveraged infrastructure spenders.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Physical AI-demand collapse&lt;/td&gt;
 &lt;td&gt;Structural bear case&lt;/td&gt;
 &lt;td style="text-align: right"&gt;Low&lt;/td&gt;
 &lt;td&gt;This conflicts with TSMC sales, memory exports, long-term contracts and upstream capacity planning.&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The distinction between cause and amplifier matters. Korea&amp;rsquo;s leverage structure alone cannot easily explain the synchronized fall in Japanese hardware names. Conversely, the regional unwind alone does not explain why Korea and its two largest memory stocks suffered such extreme losses.&lt;/p&gt;
&lt;h2 id="3-a-regional-shock-with-a-korean-multiplier"&gt;3. A regional shock with a Korean multiplier
&lt;/h2&gt;&lt;p&gt;Around early afternoon, KOSPI was down roughly 6.8% to 7.2%, while Taiwan was slightly positive, Japan was down around 1.8%, China about 1.5%, and Hong Kong only modestly lower. These were intraday observations, not final closes.&lt;/p&gt;
&lt;p&gt;Later, Japanese hardware losses deepened. Intraday screens showed Taiyo Yuden down 18.87%, Yaskawa Electric 14.34%, Kioxia 10.77%, Murata 8.06% and Ibiden 7.39%. The figures are directional observations only.&lt;/p&gt;
&lt;p&gt;The common move does not mean every company faced the same fundamental shock. Yaskawa had company-specific execution issues. Kioxia also had concentrated domestic margin exposure. Murata and Taiyo Yuden have significant smartphone and auto exposure, making their declines closer to high-beta electronics selling. Ibiden&amp;rsquo;s FC-BGA exposure gives it stronger evidentiary value for an AI-hardware derating.&lt;/p&gt;
&lt;p&gt;The most coherent decomposition is therefore &lt;strong&gt;a global hardware crowding unwind plus a Korean leverage amplifier&lt;/strong&gt;.&lt;/p&gt;
&lt;h2 id="4-why-korea-moved-more-than-taiwan"&gt;4. Why Korea moved more than Taiwan
&lt;/h2&gt;&lt;p&gt;Both Korea and Taiwan are semiconductor-heavy, but corporate concentration is not the full story.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Market&lt;/th&gt;
 &lt;th&gt;Main H1 driver&lt;/th&gt;
 &lt;th&gt;Nature of concentration&lt;/th&gt;
 &lt;th&gt;Vulnerability in a correction&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Korea&lt;/td&gt;
 &lt;td&gt;Memory earnings upgrades, domestic retail and financial-investment flows, margin and leveraged exposure&lt;/td&gt;
 &lt;td&gt;Corporate, index, derivatives and credit concentration occurred together&lt;/td&gt;
 &lt;td&gt;Forced selling can become self-reinforcing&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Taiwan&lt;/td&gt;
 &lt;td&gt;TSMC earnings credibility and domestic absorption of foreign selling&lt;/td&gt;
 &lt;td&gt;High corporate concentration but weaker single-stock leverage plumbing&lt;/td&gt;
 &lt;td&gt;Vulnerable if TSMC earnings confidence breaks&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Japan&lt;/td&gt;
 &lt;td&gt;Foreign cash-equity demand, governance reform and multi-sector participation&lt;/td&gt;
 &lt;td&gt;Leadership spread across semiconductors, banks, trading houses, automation and defense&lt;/td&gt;
 &lt;td&gt;Yen and global-cycle sensitivity&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Cited industry data indicated that KOSPI margin financing increased from roughly KRW 22.56 trillion to KRW 29.23 trillion during the second quarter, while KOSDAQ margin financing declined from about KRW 10.36 trillion to KRW 8.09 trillion. If confirmed on a consistent basis, this helps explain why KOSPI moved more violently than KOSDAQ. July 13 end-of-day margin balances and forced-liquidation amounts were not yet available.&lt;/p&gt;
&lt;h2 id="5-evidence-against-a-simple-ai-demand-collapse-thesis"&gt;5. Evidence against a simple AI-demand-collapse thesis
&lt;/h2&gt;&lt;h3 id="51-tsmc-june-revenue"&gt;5.1 TSMC June revenue
&lt;/h3&gt;&lt;p&gt;&lt;a class="link" href="https://investor.tsmc.com/english/monthly-revenue/2026" target="_blank" rel="noopener"
 &gt;TSMC&amp;rsquo;s official monthly revenue data&lt;/a&gt; showed June sales of NT$442.68 billion, up 6.2% month over month and 67.9% year over year. Second-quarter revenue was approximately NT$1.270 trillion, up about 12.0% sequentially and 36.1% year over year.&lt;/p&gt;
&lt;p&gt;The June comparison benefited from a weak June 2025 base. First-quarter growth of about 35.1% and second-quarter growth of about 36.1% do not show a doubling of the underlying rate. Yet the absolute monthly record and positive monthly and quarterly growth cannot be dismissed as a base effect. This is strong evidence that GPU, ASIC and leading-edge logic demand is converting into revenue. It is only indirect evidence for SK hynix HBM pricing, LTA margins and yields.&lt;/p&gt;
&lt;h3 id="52-early-july-memory-exports"&gt;5.2 Early-July memory exports
&lt;/h3&gt;&lt;p&gt;A brokerage interpretation of July 1-10 preliminary trade data showed DRAM export value up roughly 29.4% from the comparable June period, dollar-per-kilogram pricing up 15.7%, and implied volume up about 11.8%. The direction suggests both price and volume improved.&lt;/p&gt;
&lt;p&gt;Ten-day preliminary data do not guarantee the full month. Dollars per kilogram are not bit ASP, and the figures cannot be mapped directly to company revenue or operating profit.&lt;/p&gt;
&lt;h3 id="53-sk-hynix-ltas-and-estimate-revisions"&gt;5.3 SK hynix LTAs and estimate revisions
&lt;/h3&gt;&lt;p&gt;A &lt;a class="link" href="https://biz.chosun.com/stock/stock_general/2026/07/13/O2R4UKPBY5G2RJXLUIVOIGNWVU/" target="_blank" rel="noopener"
 &gt;Korea Investment &amp;amp; Securities estimate cited in the press&lt;/a&gt; placed second-quarter 2026 operating profit at KRW 60.4 trillion, roughly 8% below a KRW 65 trillion consensus, and cut 2026 and 2027 operating-profit estimates by about 9% and 11%. The same analysis remained constructive on DRAM and NAND pricing and the durability benefits of long-term agreements.&lt;/p&gt;
&lt;p&gt;The balanced interpretation is that near-term earnings altitude and revision momentum weakened, while LTAs may support volume visibility and profit floors. Without public price floors, cost pass-through, take-or-pay and renegotiation terms, it is not possible to conclude that LTAs eliminate the memory cycle.&lt;/p&gt;
&lt;h3 id="54-cpu-wafers-and-long-term-fab-plans"&gt;5.4 CPU, wafers and long-term fab plans
&lt;/h3&gt;&lt;p&gt;Server CPU growth reinforces the idea that AI servers require more than GPUs, broadening bottlenecks into substrates, sockets, BMCs and conventional DRAM. Forecasts such as a US$100 billion server-CPU market by 2028 and a 53% three-year CAGR rely on aggressive assumptions for agentic-AI demand and CPU pricing and should remain scenarios until orders and supplier results confirm them.&lt;/p&gt;
&lt;p&gt;Micron and GlobalWafers&amp;rsquo; long-term U.S. 300mm wafer agreement, Samsung&amp;rsquo;s reported Yongin schedule acceleration, and SK&amp;rsquo;s M15X and Yongin plans show that memory demand is entering upstream procurement and long-term capacity plans. Those fabs ramp mainly from 2027 through 2030, supporting near-term tightness while increasing the risk of margin normalization after 2029.&lt;/p&gt;
&lt;h3 id="55-hbm-through-2030"&gt;5.5 HBM through 2030
&lt;/h3&gt;&lt;p&gt;Some industry work projects HBM demand rising from roughly 4.8EB today to 26.7EB in 2030 versus supply of 10.6EB. The supply forecast appears to incorporate about 2.65 times current capacity, or roughly 21.5% annual growth, rather than assuming no expansion.&lt;/p&gt;
&lt;p&gt;The directional case is credible because stacking, leading-edge-node migration, yield, TSV, packaging and inspection constraints prevent wafer input from translating one-for-one into effective bit supply. The exact shortage is highly sensitive to token usage, KV-cache efficiency, mixture-of-experts, quantization and offloading assumptions.&lt;/p&gt;
&lt;p&gt;A practical base path is very tight supply through 2027, a growing capacity contribution from 2028, and some broad shortage relief in 2029-2030 while HBM remains structurally tighter than commodity memory.&lt;/p&gt;
&lt;h2 id="6-why-strong-demand-does-not-guarantee-a-strong-stock"&gt;6. Why strong demand does not guarantee a strong stock
&lt;/h2&gt;&lt;p&gt;The market is no longer asking only whether AI demand exists. It is asking how long excess profits can last.&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Funding costs:&lt;/strong&gt; larger AI infrastructure commitments make debt costs and cash-generation differences more important. This implies dispersion among spenders, not necessarily demand disappearance.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;AI inflation:&lt;/strong&gt; higher memory, power and infrastructure prices help suppliers but can lift inflation, bond yields and customer return hurdles. Earnings can improve while equity multiples contract.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Profit-duration debate:&lt;/strong&gt; the key question is whether current margins survive competition, capacity additions and technology shifts.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Emerging-market index concentration:&lt;/strong&gt; heavy TSMC, Samsung Electronics and SK hynix weights create benchmark-flow risk independent of corporate earnings. Global investors can stay long AI while diversifying into India, China, energy or utilities.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="7-separating-ltas-from-a-6x-pb-thesis"&gt;7. Separating LTAs from a 6x P/B thesis
&lt;/h2&gt;&lt;p&gt;If LTAs protect volume and price downside, reduce earnings volatility and extend high ROE, higher valuation multiples are theoretically justified. The market, however, observes estimate cuts immediately while treating cycle elimination as an unproven assumption.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Observable&lt;/th&gt;
 &lt;th&gt;Still unverified&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Second-quarter and 2026-2027 estimate changes&lt;/td&gt;
 &lt;td&gt;LTA floors, cost pass-through, penalties and renegotiation clauses&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Recent forecast errors and ASP expectations&lt;/td&gt;
 &lt;td&gt;HBM4 yields, packaging cost and realized premium&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;A target-price formula using forward book value&lt;/td&gt;
 &lt;td&gt;Long-duration exceptional ROE and the disappearance of downcycles&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;Applying a 12-month forward book value of KRW 643,124 mechanically gives about KRW 1.93 million at 3x P/B and KRW 3.86 million at 6x. The difference reflects profit duration and risk more than a few trillion won of quarterly earnings.&lt;/p&gt;
&lt;p&gt;At zero growth and a 10% cost of equity, a simplified residual-income identity links 3x and 6x P/B to sustainable ROEs of roughly 30% and 60%. With 4% growth, the implied ROEs fall to about 22% and 40%. This is an explanatory sensitivity, not a precise valuation. A 6x multiple requires large excess profits, long duration and structurally lower risk at the same time.&lt;/p&gt;
&lt;h2 id="8-single-stock-leveraged-etfs-and-short-gamma-plumbing"&gt;8. Single-stock leveraged ETFs and short-gamma plumbing
&lt;/h2&gt;&lt;p&gt;Market estimates suggested that single-stock leveraged-ETF assets fell from about KRW 17.4 trillion on June 25 to KRW 11.6 trillion on July 8. A modeled additional 5% shock implied sensitivities of roughly KRW 2.65 trillion for SK hynix and KRW 1.49 trillion for Samsung Electronics. These were modeled exposures, not actual sell orders.&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;Share-price decline
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; lower leveraged-ETF NAV
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; underlying sales to restore target leverage
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; further price decline
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; more program selling and margin liquidation
&lt;/span&gt;&lt;/span&gt;&lt;span style="display:flex;"&gt;&lt;span&gt; -&amp;gt; another round of underlying sales
&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;The July 13 shock may mark the beginning of normalization because it mechanically reduced leverage. It does not prove completion because end-of-day ETF assets, margin balances and forced-liquidation data were unavailable. Japan&amp;rsquo;s hardware selloff also shows that leverage was an amplifier rather than the sole root cause.&lt;/p&gt;
&lt;p&gt;&lt;a class="link" href="https://n.news.naver.com/mnews/article/008/0005385145?sid=101" target="_blank" rel="noopener"
 &gt;Reported regulatory discussions&lt;/a&gt; indicate recognition of the structural risk. Restricting new products could reduce medium-term fragility, but forced conversion, leverage cuts or liquidation of existing products could create a second round of short-term selling. New-product policy and legacy-product treatment must be separated.&lt;/p&gt;
&lt;h2 id="9-what-strengthened-weakened-and-remains-unknown"&gt;9. What strengthened, weakened and remains unknown
&lt;/h2&gt;&lt;h3 id="strengthened"&gt;Strengthened
&lt;/h3&gt;&lt;ol&gt;
&lt;li&gt;AI infrastructure demand is broadening from GPUs and HBM into CPUs, conventional DRAM, wafers, substrates and power networks.&lt;/li&gt;
&lt;li&gt;Limited 2027 commodity-memory capacity and possible LTA prioritization support memory pricing and profit duration.&lt;/li&gt;
&lt;li&gt;Early-July DRAM data suggest simultaneous improvement in price and shipments.&lt;/li&gt;
&lt;li&gt;Long-term fab schedules and upstream contracts indicate demand is being incorporated into plans beyond 2029.&lt;/li&gt;
&lt;/ol&gt;
&lt;h3 id="weakened"&gt;Weakened
&lt;/h3&gt;&lt;ol&gt;
&lt;li&gt;The assumption that good earnings and capex news immediately lift share prices.&lt;/li&gt;
&lt;li&gt;The claim that a low forward P/E automatically defines a bottom.&lt;/li&gt;
&lt;li&gt;The view that an SK hynix ADR premium creates a risk-free rerating of the Korean ordinary share.&lt;/li&gt;
&lt;li&gt;Confidence in any specific short-term price floor.&lt;/li&gt;
&lt;li&gt;The claim that LTAs alone eliminate the memory cycle and immediately justify 6x P/B.&lt;/li&gt;
&lt;li&gt;The idea that the selloff was an isolated Korean leveraged-ETF accident.&lt;/li&gt;
&lt;/ol&gt;
&lt;h3 id="unknown"&gt;Unknown
&lt;/h3&gt;&lt;ol&gt;
&lt;li&gt;Whether July 13 was the final leverage-liquidation low.&lt;/li&gt;
&lt;li&gt;How much LTAs change actual margin floors and pricing upside.&lt;/li&gt;
&lt;li&gt;Whether late-July big-tech AI spending is maintained or raised.&lt;/li&gt;
&lt;li&gt;How strongly post-2029 capacity normalizes excess margins.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id="10-scenario-map"&gt;10. Scenario map
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Scenario&lt;/th&gt;
 &lt;th&gt;Confirmation conditions&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;Bull: flow-driven capitulation&lt;/td&gt;
 &lt;td&gt;New lows stop, program selling collapses, ETF assets and margin exposure fall, foreign cash buying returns&lt;/td&gt;
 &lt;td&gt;Deleveraging gives way to a higher-quality rebound&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Base: volatile bottom-building&lt;/td&gt;
 &lt;td&gt;Program selling slows but cash buying remains weak, new lows alternate with rebounds, markets await CPI, ASML and TSMC&lt;/td&gt;
 &lt;td&gt;Liquidation continues and the low is unconfirmed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Bear: structural repricing&lt;/td&gt;
 &lt;td&gt;New lows despite constructive CPI, ASML and TSMC; big-tech capex slows; estimates fall in sequence; SK hynix lacks relative strength&lt;/td&gt;
 &lt;td&gt;Both 2027 earnings and justified P/B move lower&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;The most important test is whether prices begin to rise on good news. A change in reaction function is more direct bottom evidence than the optimism of the headline itself.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;U.S. semiconductor response&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;SOX outperforms Nasdaq 100 and MU, NVDA, AVGO, AMAT and LRCX reclaim lows&lt;/td&gt;
 &lt;td&gt;Asia over-discounted the risk and Korean leverage was a major amplifier&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SOX falls 3% to 5% with memory, equipment and substrates weak together&lt;/td&gt;
 &lt;td&gt;A global AI-hardware multiple correction is confirmed&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Leveraged AI-infrastructure spenders fall while cash-rich big tech and semis defend&lt;/td&gt;
 &lt;td&gt;Funding-capacity differentiation, not broad AI-demand collapse&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;h2 id="11-what-to-monitor-next"&gt;11. What to monitor next
&lt;/h2&gt;&lt;ol&gt;
&lt;li&gt;The next one to three trading days: new lows, volatility, volume and program-selling intensity.&lt;/li&gt;
&lt;li&gt;Confirmed single-stock ETF assets, margin balances, forced liquidation and foreign cash flows.&lt;/li&gt;
&lt;li&gt;Hormuz traffic, insurance costs, WTI, the U.S. 10-year yield and USD/KRW.&lt;/li&gt;
&lt;li&gt;SOX and the relative strength of MU, NVDA, AVGO, AMAT and LRCX.&lt;/li&gt;
&lt;li&gt;July 14 U.S. CPI composition and the joint response of rates, oil and semiconductors.&lt;/li&gt;
&lt;li&gt;TSMC&amp;rsquo;s July 16 guidance, AI-accelerator demand, CoWoS constraints and capex.&lt;/li&gt;
&lt;li&gt;Late-July AI capex, AI revenue, free cash flow and funding pressure at Microsoft, Meta, Alphabet and Amazon.&lt;/li&gt;
&lt;li&gt;Third-quarter SK hynix HBM4 shipments, yields, ASPs and LTA effects, plus Samsung&amp;rsquo;s HBM qualification and AI5 foundry ramp.&lt;/li&gt;
&lt;li&gt;The distinction between restrictions on new single-stock leveraged ETFs and treatment of existing products.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id="12-evidence-boundary"&gt;12. Evidence boundary
&lt;/h2&gt;&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;Evidence level&lt;/th&gt;
 &lt;th&gt;Inputs&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;High&lt;/td&gt;
 &lt;td&gt;Closing prices, Kiwoom daily bars, official TSMC monthly sales and company releases&lt;/td&gt;
 &lt;td&gt;Establish facts and direction; do not automatically map into supplier earnings&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Medium&lt;/td&gt;
 &lt;td&gt;Brokerage estimates and major financial-media reporting&lt;/td&gt;
 &lt;td&gt;Forecasts and market interpretation, separated from company facts&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Limited&lt;/td&gt;
 &lt;td&gt;Intraday screens, program proxies and ten-day export data&lt;/td&gt;
 &lt;td&gt;Directional signals requiring official close and full-period confirmation&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Hypothesis&lt;/td&gt;
 &lt;td&gt;Confidential LTA terms, 6x P/B assumptions, bottom and regulatory effects&lt;/td&gt;
 &lt;td&gt;Scenario analysis only; insufficient to change a thesis alone&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 July 13 decline was neither a Korea-only market-plumbing accident nor proof that physical AI demand disappeared. A global reduction in crowded AI and electronics-hardware exposure set the direction. Lower SK hynix estimates and funding-cost concerns provided direct catalysts, while Korea&amp;rsquo;s leveraged ETFs, margin exposure and program selling magnified the decline.&lt;/p&gt;
&lt;p&gt;Record TSMC monthly revenue, preliminary memory exports and long-term HBM supply evidence indicate that demand remains alive. The market is nevertheless demanding a higher standard of proof: how long profits last, whether LTAs protect margins, whether AI capex can be funded from cash flow, and how much crowded exposure has actually been removed.&lt;/p&gt;
&lt;p&gt;The next conclusion will come from price behavior rather than the number of optimistic headlines. U.S. semiconductor confirmation, the response to CPI and TSMC, a return of foreign cash buying and weaker Korean program selling will determine whether this was a leverage reset or the start of a deeper structural repricing.&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;Data cutoff: Korea&amp;rsquo;s July 13, 2026 close. Program and selected flow figures are Kiwoom REST proxies pending official KRX confirmation. This article is market-structure research for informational purposes and is not a recommendation to buy or sell any security.&lt;/p&gt;

 &lt;/blockquote&gt;</description></item></channel></rss>