AI HBM Hub: Samsung Electronics, SK Hynix and Korea Semiconductor Exposure

Korea Invest Insights

One-Line Thesis

Korean semiconductor exposure is not one trade. It combines SK Hynix’s HBM leadership, Samsung Electronics’ memory and foundry recovery option, and KOSPI index concentration around Samsung and SK Hynix. HBM is both a stock-level catalyst and the central earnings engine behind Korea’s market re-rating.

Quick Answers

Search Question	Current Answer	Read
What did the July earnings season leave behind for Korean memory	Memory pricing became an earnings variable outside the memory companies for the first time. Amazon named memory costs as the reason for raising capex to $220bn, Apple called the move a hundred-year flood, and IBM’s mainframe revenue fell 42% as client budgets shifted to memory pre-buys. The recipients on the other side are Samsung (89.5tn won quarterly operating profit) and SK Hynix (60.5tn won at a 76% margin). SK Hynix’s miss was not a demand problem but the product of LTA fixed pricing (about half of revenue) and deferred HBM4 shipments, the lessons of 2018 written into accounting as contract structure. With the $26.5bn Nasdaq ADR and CXMT’s listing landing the same month, capital bet on Korea’s present and China’s future simultaneously; CXMT’s Q1 profit came from bits +11% and ASP +57%, price-made earnings, so the real tail risk is that it reverts to loss-making expansion when 2028 capacity lands into falling prices.	July Earnings Season Wrap
Is this crash mid-cycle or the end, and where does Korean memory stand	Scoring seven discriminators (estimate direction, valuation, utilization, credit markets, nature of selling, policy, anchor-customer economics) gives rebound 5, terminal 0, undecided 2. The decisive one is estimate direction: every historical terminal came with downgrades, yet consensus rose through this crash (+2.6-4.4%), and memory’s forward PER of 4.7-5.6x is the opposite of terminal-group extremes. Korean memory’s implied persistence of 38-42% (own back-calculation) is a price that denies everything beyond 2028, and the keys to repricing it are the prepayment (contract liability) jump and 3Q contract prices at Samsung and SK Hynix, plus the OpenAI reading on Microsoft’s call (the equity-method line alongside Azure and backlog). The terminal checklist, estimate downgrade turn, first major chain bankruptcy, credit issuance shutdown, has zero items lit. The discipline: do not sell into chaos; sell the day estimates roll over.	Rebound or Terminal?
What does NVIDIA’s OpenAI guarantee mean for Korean memory	If the $350bn chip financing executes, the credit quality of the HBM orders inside those chips becomes tied to this guarantee structure, making it the largest single case for the buyer-mix watch frame (FCF buyers versus financed buyers). The deal’s discriminating variable converges to one number: how many times over OpenAI’s revenue has grown by the 2028 start-up (4x from $25bn is precarious, 6-8x works). If it succeeds, the guarantee expires undrawn and HBM revenue flows per contract; if it fails, a contingent liability worth 159% of NVIDIA’s equity crystallizes and the HBM revenue behind Ohio-bound chips becomes revenue contingent on the bridge surviving. Galaxy’s 9.875% coupon and Oracle’s BBB- downgrade mark the current position on that spectrum. And even a turnaround leaves the structure replicating toward Anthropic and the neoclouds, so the system-wide vendor-financing stock and the memory orders inside it stay on the watch list.	Anatomy of the $250 Billion Backstop
Should you sell Samsung and SK Hynix on China’s domestic DUV report	Check the stock-level channel first. HBM’s bottleneck is stacking, TSV, base die and customer qualification, not lithography, so SK Hynix’s profit core has no transmission channel, and Samsung’s 2026-2027 contract prices are locked. The report’s real new information is insurance that the ~1,400-tool stockpiled ASML fleet can keep migrating nodes despite controls, a terminal-value variable that raises the post-2028 commodity-glut odds — an axis we classified ten days ago in the China localization piece and one already largely priced into current multiples. Black Tuesday’s -11.81% (Samsung) and -12.89% (SK Hynix) were roughly 80% positioning: estimates rose and no target was cut through the episode. Watch items: teardown-grade verification of the domestic tool, CXMT G5 wafer shipments, the MATCH Act’s service clause, Japan’s parts controls; the discriminator remains the DRAM contract price.	Anatomy of Black Tuesday
If AI is the dot-com, is Korean memory Cisco	No: the sharper analog is Cisco’s component suppliers, and they (JDS Uniphase, Nortel) fell ~-99% in the 2001 rupture, deeper than Cisco’s -89%, because sequencing ruptures amplify up the value chain through the inventory bullwhip. What differs now is the buyer. Cisco’s marginal customers in 1999 were junk-bond-funded CLECs; memory’s main buyers today are hyperscalers generating $300-400bn of annual FCF. But NVIDIA’s reported talks to guarantee $250bn of OpenAI’s data-center debt plus $350bn of chip financing, and debt-funded neoclouds, signal 1999-style credit seeping into the buyer mix. An LTA protects the price in a world where the counterparty stays solvent, not that world itself. Watch items: the DRAM contract price, the quarterly monetization gap ratio (narrowing 8x to 4.6x), and the FCF-buyer versus financed-buyer mix.	Cisco Did Not Die of Missing Demand
What do the San Francisco AI Declaration and $950bn mean for Korean semis	It extended duration, not volume. Korea already captures roughly 20-25% of every AI capex dollar through memory (HBM is 35-55% of accelerator cost, Korea holds 79% of HBM, memory is 30% of capex), and Samsung’s KRW 89.4tn and SK Hynix’s KRW 64.1tn in quarterly operating profit are the proof. What this week added is that share locked on paper through 2030. The least-priced item is Samsung Foundry: a segment losing KRW 2.8tn a quarter at 6.5% share just won Broadcom as its second anchor tenant, having posted its first monthly profit since 2023 in June. Against that, most of the deals are LOIs and MOUs, SKT’s 2GW exceeds the 1.5GW projected for all Korean data centers, and Samsung, SK and TSMC all expanding at once raises the 2028 margin-normalization risk. The discriminator remains the DRAM contract price.	Dissecting $950 Billion
Should you sell Samsung and SK Hynix when oil and rates rise	No: that is a category error. Oil-driven rate rises reach Korean memory through four channels: multiple compression, dollar-strength flows, power and materials costs, and earnings, and only the first three answer to rates while earnings are set by contracts. The 3Q server DRAM contract forecast of +13-18% and the 4Q negotiations on 2027 HBM pricing are what move profits, and Alphabet’s print strengthened that demand backdrop. At low-to-mid single-digit forward PERs, memory should carry less rate beta than 27x US growth stocks, so falling in sympathy on oil-spike days is the work of flows and fear. The exception is escalation (Brent above $110), where low multiples are no shield. Invalidation test: if oil breaks below $80 while the US 10Y holds above 4.5%, the culprit is structural, not oil, and the trigger thesis is rejected.	From Oil to Korean Memory
What does Alphabet’s print tell memory investors	Strongly positive on volume, neutral on price. Cloud +82%, a $514bn backlog, consumption 50% above commitments and renting SpaceX compute at ~$920M a month all weaken the post-2028 demand-cliff premise. About 60% of infrastructure CAPEX goes to servers, and with TPU system sales and rented compute the HBM and server DRAM demand channels now run three ways. Axion CPUs, data and security workloads and 500+ customers processing over a trillion tokens a year thicken the server DRAM and eSSD case. But the first negative quarterly FCF (-$5.9bn) and a CAPEX guide raised to $205bn show the market now grades on cash, and memory supply also grows through 2027-2029, so volume must not be conflated with ASP or peak margins. A better demand backdrop for the 4Q negotiations on 2027 HBM pricing.	Alphabet’s Q2
Did FADU’s Q2 surprise prove an AI-storage moat?	It proved Gen5 commercialization and profitability. Q2 revenue was KRW 73.16bn, operating profit KRW 16.27bn and operating margin 22.2%, with operating profit 89.2% above consensus. Gen6 production, 64TB/128TB QLC adoption and customer diversification remain unconfirmed, leaving a 3.0/5 narrow but expandable moat.	FADU’s Q2 Surprise and Its Real Moat
Beyond hyperscalers, who actually uses AI infrastructure	NVIDIA’s books are starting to answer. The hyperscale share has held near 50% for seven quarters, but in the Feb-Apr 2026 quarter non-hyperscale (ACIE) revenue of $37bn reached parity with $38bn and quarterly growth flipped, 31% vs 12%. Sovereign AI revenue topped $30bn in FY2026, tripling, and Saudi HUMAIN locked 600,000 GPUs over three years. That demand tightens server DRAM outside the LTAs: TrendForce sees 3Q contract prices up 13-18% and Inventec reports 40-week lead times. Codex adding three million users in the three days after GPT-5.6’s release is the first hard datapoint on the end-usage weak link. Still, no sovereign direct memory purchase has been disclosed and Stargate volumes remain a non-binding LOI, so the verdict moves to the Jul 30 earnings window and NVIDIA’s late-August print.	Who Burns All Those Tokens?
Is NVIDIA’s Vera CPU expansion and lower SoCAMM2 capacity bullish for Simmtech?	Separate bits from module units. TrendForce’s reported redesign lowers memory capacity per module because LPDRAM supply is tight, while increasing total module shipments and Vera CPU production. Simmtech supplies a PCB for each module, so the change does not automatically reduce substrate demand. NVIDIA’s official material also expands Vera beyond NVL72 into standalone CPU racks with up to 256 CPUs and major OEM systems in H2 2026. Still, Simmtech’s supply to all three memory makers and leading share are analyst estimates, and persistent foreign/program selling keeps the stock a conditional rather than an immediate buy.	NVIDIA Vera CPU Expansion and Simmtech
Did the third-week-of-July selloff end the memory supercycle?	The official evidence does not show a demand break. TSMC and ASML results plus DRAM and NAND pricing support strong 2026-2027 earnings. Yet ASML capacity expansion, HBM LTAs, Chinese commodity supply and open-model efficiency can shorten the duration of excess profits from 2028. The current distribution is 25% prolonged scarcity, 40% strong earnings with lower multiples, 25% efficiency and supply normalization, and 10% financing or demand break. Contract prices, 2028 EPS, big-tech AI profitability and relative strength matter more than labeling the selloff as flow-only.	Weekly Semiconductor Deep Dive
How should SK Hynix Chairman Chey Tae-won’s remarks feed into an investment view	He holds information public investors cannot easily reach on customers’ long-term supply demands and production constraints, but on fair value and long-term margin he carries the conflict of being the largest shareholder and the issuer of about KRW 40 trillion in new shares. Trust the direction, discount the intensity. His words strengthen the AI memory volume supercycle, yet supply anything, cut token cost and double capacity read as a pledge not to perpetuate today’s excess margin. So the company thesis strengthens while the stock’s margin thesis weakens. The call is Hold, Conditional Add, No Chase, with additions gated on FY2028 earnings holding, long-term contracts converting to real volume and post-expansion ROIC defense. The discriminator is the DRAM contract price.	SK Hynix Chairman Chey Tae-won’s Two Months
Will AI memory demand grow beyond our expectations	Meeting or exceeding today’s high expectations is likely (strong growth ~80%), but a large beat is not yet the base case (~35%). Probability-weighting demand into base 45%, beat 35% and miss 20%, the key to a beat is not server unit sales but accelerator shipments and memory content per accelerator both beating at once. Upside comes from GPU and ASIC growing together, spread beyond HBM (SOCAMM2, eSSD) and always-on agent inference; downside from CAPEX ROI doubts, efficiency gains and China supply. On supply, HBM consumes roughly 3x the wafers of DDR5, so memory layers compete for a single wafer, and Chairman Chey’s expansion argument reads as market preservation rather than a price war. The conclusion is a shortage thesis through 2027 and re-verification from 2028.	Will AI Memory Demand Exceed Expectations?
Is the US House letter seeking Chinese-memory procurement restrictions bullish for Samsung and SK Hynix?	It is not enacted regulation but a July 16 bipartisan policy request from two House Select Committee lawmakers. If the CXMT Entity List review, AI and data-center procurement restrictions, and allied coordination are implemented, the larger effect is a lower 2027-2028 Chinese supply-overhang risk rather than more 2026 HBM volume. Samsung benefits more directly through mobile and commodity DRAM plus NAND, while SK Hynix gains duration protection for its HBM advantage and multiple.	US House Letter on Chinese Memory Procurement
Do Kimi K3 and linear attention reduce HBM demand?	They can reduce KV cache per request, but that does not automatically reduce total HBM demand. Kimi Linear’s up-to-75% KV-cache saving and 6.3x throughput come from a 48B research model, not measured K3 production. K3 still needs 2.8T weights and recommends 64+ accelerators. If open-weight self-hosting and agent workloads grow faster than efficiency, total HBM demand rises, while long-lived context spills into server DRAM and enterprise SSDs, expanding the full memory hierarchy.	Kimi K3 Resets the AI Price Curve
How dangerous is China’s localization for Samsung and SK Hynix?	Disclosed China exposure runs Samsung 30.1% (company-wide), SK Hynix 24.3% (sales subsidiary) and Micron 10.1% (customer HQ), on bases too different to compare directly, so they are proxies. The real risk is not lost China revenue but the second-order effect of displaced volume flowing overseas and pressuring global DRAM and NAND ASP, which is why Micron is not safe despite the lowest direct exposure. CXMT’s RMB 57.9bn raise brought buyers in as shareholders, securing funding and customers at once. Still, the case for a 2026-2027 earnings collapse is thin; this belongs in 2028 earnings and terminal multiples as a discount variable. The decisive signal is not Chinese domestic share but contagion into global contract prices.	China’s Memory Localization and Korea
Is China’s ultra-low API pricing sustainable, and what does it mean for semis?	Hong Kong disclosures split the answer. Zhipu’s API gross margin of 18.9% (from 3.3%) clears marginal inference cost, but total gross profit covers only 9.3% of R&D (MiniMax 7.9%), so fully loaded cost is not recovered. The low pricing is therefore held up jointly by efficiency gains and capital raising, and since those proceeds flow back into training and compute infrastructure, it does not imply falling total semiconductor demand. Cheap APIs do lift inference volume, favoring server DRAM, NAND and networking while reducing HBM intensity per token. The decisive variable is not the rate of price decline but the growth rate of total token usage.	Is China’s AI API Pricing Sustainable?
Are Chinese open models bad news for Korean memory?	Not simply bad news, but redistribution. Cheap inference and open-model diffusion lift not just HBM but total server DRAM, NAND and commodity memory, which favors Samsung’s broader exposure, while MoE and KV compression cut HBM per token and pressure the scarcity multiple of SK Hynix’s pure HBM exposure first. Hence Samsung > SK Hynix on relative preference. US-China separation drives duplicate investment across both blocs, supporting total infrastructure demand and protecting SK Hynix’s HBM premium longer inside the allied bloc. CXMT’s HBM sits 1.5-2 generations behind, making it a conditional option for 2028 and beyond rather than a present supply shock.	China’s Open-Model Convergence and Value-Chain Redistribution
What is fair value for Samsung, SK Hynix and Micron?	Samsung at 3.9x and SK Hynix at 4.3x on 2028 consensus is a price that doubts the duration of earnings, not their existence. A share price is set by neither 2028 nor 2029 EPS alone but by 2026-2028 FCFE plus a normalized 2029 terminal value, and at an 11% required return one won in three years is worth 0.73 won today, so 2029 is not a distant future. Probability-weighted fair values are KRW 385,000 for Samsung (+51%), KRW 1,950,000 for SK Hynix (+6%) and $1,140 for Micron (+34%). SK Hynix already falls below its current price from the second scenario onward, making it the most dependent on sustained excess demand, while Samsung has the largest buffer with three-year FCF covering 26% of value.	Semiconductor Fair Value with FCFE and Normalized Earnings
The industry looks strong, so why are semiconductor stocks weak?	Because four physical clocks (price, investment, construction, monetization) and one stock-price clock run at different speeds. When price and investment run ahead while construction and monetization lag, the market discounts future idle capacity, depreciation and cost of capital rather than current earnings, so strong results and CAPEX raises become selling catalysts. Developing both bull and bear in their strongest forms, the dispute is not whether demand exists but its duration, the speed of supply response, and incremental profitability. The central path is a fundamental bull coexisting with a valuation bear (55%), as the scarcity cycle shifts into a capital-intensity cycle.	Four Physical Clocks and One Stock-Price Clock
Is there evidence memory demand is more than a supplier claim?	Yes. IBM’s preliminary Q2 print on July 14 missed consensus and the stock fell more than 17% premarket, and one of CEO Krishna’s stated reasons was that “in the last few weeks of June, we saw clients shift their quarterly capex spend toward servers, storage, and memory purchases to secure supply-constrained infrastructure ahead of expected price increases.” That is testimony from a buyer of memory, against its own interest, which makes it high-credibility. The same day Ericsson warned on margin pressure from AI-driven component costs. The offset: IT budgets get crowded out, and the late-June pull-in leaves a Q3 payback risk.	Why IBM’s Earnings Miss Is Evidence of Memory Strength
Is the HBM supercycle benefit spilling beyond HBM?	Yes. Synthesizing three Hana Securities reports, the supply gap from the big-three’s shift to HBM and DDR5 spills three ways. Package substrates (Daeduck Electronics, Simmtech, Haesung DS, where DDR5’s larger, higher-layer boards erode effective capacity, with LTAs as the shortage signal), eSSD controllers (ASICLAND as SK Hynix’s Gen6 partner, FADU-bound Gen6 mass production in H2 2027), and legacy memory (GigaDevice, SLC NAND ASP up ~5x and legacy DRAM ~10x). The boom has broadened, but the checkpoints are substrate LTA filings, FADU revenue recognition, and China’s capacity-fill pace.	The Supply Gap Spilling Beyond HBM
Brokers cut SK Hynix Q2 earnings, so why hold the target prices?	Mirae Asset (KRW 62.3tn) and Korea Investment & Securities (KRW 60.4tn) both cut Q2 operating profit, but this is a normalization, not a demand slowdown: LTAs and a high HBM mix mean the surging spot price is not fully reflected in blended ASP. Target prices held because they derive from 12-month-forward BPS times a 6x target P/B, not Q2 earnings, and the market has not accepted that re-rating, punishing the stock -15.37% on July 13. At roughly 3x 12MF BPS today, brokers price the re-rating while the market prices normalization.	SK Hynix Q2 Earnings Cut but Target Prices Held
Is HBM really 2.5x short in 2030?	We independently reproduce the widely-shared 2030 claim of 26.7EB demand versus 10.6EB supply (a 2.5x shortage) from the published formula. The arithmetic reproduces, but it is a bull scenario needing 24x tokens, 5x model scale, 4x context and 70% KV residency at once. Synthesizing the demand-model structure, Goldman/DeepSeek/TurboQuant counter-evidence and the Big 3 capacity schedule, tight through 2027 and supply relief from 2028 are robust, while a precise 2.5x in 2030 is low-probability. Separate confidence in direction from uncertainty in magnitude.	HBM 2030 Supply-Demand Deep Research
Is the market really oversold versus the 2027 consensus?	On average estimates it looks cheap, but the same math held at the 2018 peak. Reverse-engineering shows both Samsung and SK Hynix trade at 11.7x the street’s lowest 2027 EPS estimate, meaning the market has already adopted the worst case as its base case. The verdict comes from late-July big-tech 2027 CAPEX commentary, SK Hynix’s Q2 LTA price-floor disclosure, and the direction of the lowest estimates.	The Worst Case Is Already the Price
What does the US-China split in agentic inference infrastructure change for Korean semiconductors?	The US is optimizing token throughput per watt by combining HBM with SRAM/LPU, while China is working around leading-edge logic and HBM limits through optical mesh and parallel scale-out. The cleaner Korean opportunity is not China optical modules, but HBM, power equipment, advanced packaging and Samsung’s SRAM/LPU foundry option sold into the US-style AI bottleneck.	US-China inference infrastructure split
How cheap do Samsung and SK Hynix look on 2028E earnings?	On MarketScreener 2028E net income, Samsung and SK Hynix trade at about 4.1x and 4.4x 2028E P/E. But a scenario model using supply, HBM LTAs and historical drawdowns lowers the probability-weighted upside to +36-81% for Samsung and +23-64% for SK Hynix. The key is whether 2028E is normalized or peak earnings.	Samsung and SK Hynix 2028E valuation
How much of one AI token dollar does memory capture?	One current AI token dollar splits roughly into 45-55 cents of model-layer value added, 10-16 cents for cloud, about 13 cents for GPUs and about 2.3 cents for memory. Memory is a real AI physical bottleneck, but its economic share is small. The key is not P/E alone, but HBM4 contract pricing, ASP durability, cloud backlog conversion and whether mid-cycle net income can reach KRW 100-140tn.	AI token value and memory value added
After Samsung’s 2Q26 preliminary result, what is signal and what is noise?	KRW 89.4tn of operating profit, DRAM ASP +47%, NAND ASP +66%, legacy-memory price spikes and hyperscaler capex strengthen the investment case. Foreign selling, high rates and FX, ASML/TSMC high-bar risk and weak Korean breadth weaken the entry timing. The stance is not thesis abandonment, but strong hold and no incremental buying until the next proof points.	Samsung 2Q26 signals and noise
Which NVIDIA quarter does Samsung’s 2Q26 selloff resemble?	Samsung’s July 7-8 two-day drawdown of -12.7% is closer to NVIDIA Q4 FY26 in February 2026 than to the August 2024 one-day sell-on. NVIDIA’s big post-print selloffs eventually recovered, but strong numbers alone were not enough. The rebound required a catalyst that directly killed the fear behind the selloff. For Samsung, late-July Big Tech CapEx guidance is the equivalent proof point.	Samsung selloff and NVIDIA analogue
What does the Big Tech financing relay say about the memory thesis?	The Amazon, Meta and Alphabet bond/equity financing relay lowers the probability of an absolute 2026 AI capex cut. But hyperscaler equities also carry FCF, debt and dilution risk. The cleaner read-through is memory and HBM bottlenecks, with late-July calls needed to confirm the 2027 capex slope and whether memory-cost pressure keeps being absorbed.	Big Tech financing and the memory bottleneck
What can late-July Big Tech earnings calls change for the memory thesis?	The key is not total capex, but the 2027 capex slope and memory-price durability. For Alphabet, Microsoft, Meta and Amazon, the probability map is 50% strengthening, 35% neutral and 15% weakening, with structural damage around 4%. Samsung’s 2Q26 preliminary result cleared the first gate, but the July 30 call needs to confirm DS core profit, 3Q pricing tone and HBM4/HBM4E progress.	Late-July Big Tech earnings and the memory thesis
Is SK Square an alternative way to own the SK Hynix HBM cycle?	Partly, but the core is not simple Hynix share-price beta. The key is whether SK Hynix FCF becomes cash dividends and whether SK Square allocates that cash to buyback cancellation or Hynix-linked AI/semi investments. SK Square is better read as a listed AI and semiconductor capital allocator built on SK Hynix equity value, not as the investment vehicle for the whole SK Group.	SK Square investment memo
How can SK Hynix’s ADR change the common share and leveraged ETF flows?	SKHY can broaden U.S. AI-memory investor access, but in the short run the key is not volatility dampening. Watch ADR-common-share arbitrage, offer discount, first-week premium or discount and Korea’s single-stock 2x ETF close rebalancing. The clean stance is watch the common share and ADR, avoid leveraged wrappers.	SK Hynix ADR price transmission
What does SK Hynix’s SKHY ADR preparation change for HBM investors?	As of the June 30 F-1/A and July 1 F-6, SK Hynix is preparing a Nasdaq ADS listing under `SKHY`. The F-6 confirms 1 ADS = 0.1 common share, while the board-authorized new-share ceiling is 17.79mn shares, 2.50% of existing shares. The key issue is not dilution alone, but U.S. AI-memory long-only access and the packaging of HBM, Custom HBM and eSSD capex into a U.S.-listed format.	SK Hynix F-1/A and F-6
Was H1 2026 a broad AI/HBM bull market?	No. Korea’s KOSPI rose +101.1%, but KOSDAQ was down -1.0% and the median Korean stock was down -12.6%. In the U.S., the stronger signal was the shift of AI bottlenecks beyond GPUs into memory, storage, servers, equipment and power.	H1 2026 Review
For Samsung 2Q26, should investors focus on reported OP or Core OP?	The key is not reported OP of KRW 87-92tn alone, but Core OP after adding back the catch-up accrual for performance-based compensation. Micron’s FY3Q26 surprise was reflected in Korean memory names on June 25, but most of it was erased by June 29 intraday. Samsung is back to a price that barely embeds Micron’s read-through; the next test is Core OP of KRW 102-107tn and 3Q OP durability near KRW 100tn.	Samsung 2Q26 Preview
When do HBF and HBC become stock triggers?	They are milestone trades before they are revenue trades. For HBF, watch second-half 2026 samples, early-2027 inference-device samples and 2028-plus revenue. For HBC, AI200 is the 2026 pre-HBC shipment, AI250/HBC Gen 1 is the mid-2027 sample target, and revenue is more likely 2028 or later. The clean exposure is SanDisk for HBF, Qualcomm for HBC, and Samsung Electro-Mechanics as Korea’s confirmed component exposure; FADU in HBF and Hanmi in HBC are weak classifications.	HBF And HBC Commercialization Calendar
How do HBM, HBF and HBC differ, and are they substitutes?	They are not the same type. HBM (bandwidth) and HBF (capacity) are memory components; HBC is Qualcomm’s accelerator architecture name, not a memory chip. Maturity differs sharply: HBM is in mass production (★★★★★), HBF is simulation-stage (★★, first sample H2 2026), HBC is a blueprint (★, 2027 sample target). The three are complements, not competitors: hot data (KV cache) goes to HBM, cold data (frozen weights) goes to HBF. No technology threatens HBM’s working-memory role today; HBF/HBC revenue contribution is earliest 2027-2028.	HBM, HBF and HBC: How to Tell AI Memory Technologies Apart
Can demand actually pay for the 2027 memory/HBM consensus?	The real payer is hyperscaler CAPEX, not consumer electronics. Big-four 2027E CAPEX of ~$782.2bn against ~$119.9bn of FCF is payable on paper but with a thin buffer. Government / sovereign AI is supplementary; PC and smartphone OEMs are in the can’t-pass-through, can’t-pay zone. Memory earnings are more sensitive to ASP and HBM scarcity persistence than to CAPEX itself.	Who Pays For The 2027 Semiconductor Consensus
What is SK Hynix’s HBM market share?	A practical 2026 baseline is roughly 50% of total HBM revenue/share, with potentially higher exposure in Nvidia’s HBM4 ramp if allocation reports hold.	SK Hynix HBM Market Share 2026
How large is Samsung Electronics in KOSPI?	For Korea ETF investors, Samsung Electronics is not just one holding. In major Korea index products, its weight typically sits in the low-20s to low-30s percentage range.	Samsung Electronics Weight in KOSPI
Samsung or SK Hynix: which is cleaner HBM exposure?	SK Hynix is the cleaner public HBM exposure. Samsung is a broader semiconductor and electronics compounder with HBM recovery, DDR5/eSSD, foundry and device exposure.	SK Hynix Deep Dive · Samsung Electronics Deep Dive
Can Samsung C&T work as a lagging Samsung Electronics proxy?	Yes, but it is not a high-beta substitute. As of Jun 1, 2026, Samsung C&T’s YTD stake-value pass-through was 51.7%, its YTD beta to Samsung Electronics was 0.83, and a 60% pass-through recovery implies roughly KRW 489,000.	Samsung C&T Samsung Electronics proxy / NAV gap trade
Is SK Hynix’s higher P/E versus Samsung a new normal?	The 2026E annual P/E inversion is unusual but explainable through the HBM bottleneck premium. The 12-month-forward view is closer to parity, and the cleaner relative-value alpha is Samsung’s HBM4E catch-up rather than chasing SK Hynix.	Samsung vs SK Hynix Forward P/E Inversion
What does Samsung’s HBM4E 12-high sample shipment mean?	It is evidence that the HBM catch-up path is alive, but not yet confirmed revenue before customer qualification, mass production and EPS revisions. For Hanmi, it raises Samsung TC-bonder option value rather than confirming an order.	Samsung HBM4E 12-high samples
Who is ahead in the HBM4E 12-high race, Samsung or SK hynix?	Samsung’s technical re-entry signal is stronger after its earlier sample announcement, but SK hynix still leads in HBM share, NVIDIA lock-in and production learning. The first proof window is 3Q-4Q 2026 customer qualification, mass-production shipment and 2027 allocation commentary.	HBM4E 12-high race: Samsung’s re-entry vs SK hynix supply lock-in
Why is Samsung shareholder return understated by dividend yield?	Samsung is not a high-dividend stock, but FY2026E net income near KRW 300.2tn makes the 50% FCF return policy more valuable. The DS bonus treasury-share purchase is not cancellation, but it can add KRW 20-22tn of company purchase demand.	Samsung Is Not A Dividend Stock
Is CXMT’s IPO an HBM price-collapse signal?	No. CXMT’s IPO can cap pricing power first in client DDR5, LPDDR and consumer NAND, but HBM remains a separate pricing structure driven by customer qualification, stacking yield, packaging, long-term agreements and AI roadmaps. The key question is whether HBM and AI server memory mix can absorb client DRAM/NAND risk.	CXMT IPO And Memory Price Risk
What does Jensen Huang’s “all three vendors qualified” HBM comment change?	It suggests Samsung Electronics, SK hynix and Micron are all in the qualified / production structure for Vera Rubin HBM4. Samsung gets HBM-discount compression, while SK hynix sees monopoly-premium dilution and TAM expansion at the same time.	Jensen Huang’s bigger HBM4 comment
What does theme ETF rebalancing mean for Korean semiconductor equipment names?	On Jun 12, KR Theme ETF Rebalance Flow v1 scanned 31 ETFs, 30 valid ETFs and 291 constituent rows, producing 69 candidates. The strongest signal was cap redistribution inside semiconductor TOP10 / AI semiconductor ETFs: Leeno, EO Technics, Soulbrain, DB HiTek and Hanmi were buy-pressure candidates, while SK Hynix, Samsung Electro-Mechanics and LS ELECTRIC were cap-trim watches.	Korea theme ETF rebalance flow
How should investors respond when ETF flow increases semiconductor volatility?	Confirm semiconductor equipment redistribution candidates on T+1/T+3 rather than chasing on the event day. For Samsung and SK Hynix, watch foreign cash equity flow and big-figure breakout/pullback patterns more than ETF cap pressure. Treat KOSPI200 expiry squeezes as separate event flow.	ETF flow is leading the Korean market
Where is Korean semiconductor ETF flow most sensitive?	Samsung Electronics and SK Hynix are 90.8% of the top-50 semiconductor market-cap proxy, but ETF-flow sensitivity is higher in Leeno, DB HiTek, EO Technics, Wonik IPS, HPSP and ISC. The under-owned ETF gap watchlist is TCK, Daeduck Electronics, Korea Circuit and Doosan Tesna.	Korean semiconductor ETF exposure strategy
Which Korean semiconductor names could beat Samsung and SK Hynix over the next two months?	Megacap memory remains core exposure, but relative alpha may come from under-owned second-line equipment, substrate and AI storage names. The compressed list is Korea Circuit, HPSP, TES, VM and FADU.	Korean semiconductor Top 50 two-month alpha candidates
When can KOSDAQ semiconductor equipment recover after Samsung/SK Hynix crowding?	The better condition is not a Samsung/SK Hynix collapse, but five to ten sessions of sideways mega-cap memory after good news. If Micron’s June 24 print does not trigger new highs and retail capital returns to KOSDAQ, test, equipment and metrology names such as Leeno, HPSP and Park Systems become the first watchlist.	KOSDAQ recovery after Samsung/SK Hynix crowding
What does Hanmi’s June 1 IR notice mean for the HBM equipment thesis?	It expands the story from HBM TC bonders toward 2.5D packaging, AI system semiconductors, OSAT and HBF. But it is not an order, so investors still need flow confirmation, real customer / production data, and a recovery of the latest CEO purchase-price zone around KRW 315,000.	Hanmi IR notice and TC-bonder TAM expansion
Do Samsung and SK hynix look cheap again versus Micron?	Yes. As of Jun 1, 2026, Samsung/Micron forward PER is 0.82x and SK hynix/Micron is 0.80x. The discount is less about Korean memory weakness and more about Micron re-rating faster.	Sam-Ha-Ma Parity
Does the broader AI chip basket also show Korean memory as too cheap?	Yes. In the AI Chip & Memory Forward P/E map, Samsung and SK hynix sit near 6x 2028E P/E, the lowest cluster in the basket. Low P/E is not automatically bullish, but if EPS durability holds, the discount versus Micron’s U.S.-listed AI-memory premium looks wide.	Sam-Ha-Ma parity follow-up
Is Micron’s higher forward P/E versus SK Hynix a technology premium?	Not really. SK Hynix still leads on technology, references and current HBM profitability, while Micron earns a U.S.-listed AI-memory scarcity premium plus an HBM4, SOCAMM2 and Gen6 SSD platform narrative.	SK Hynix vs Micron
Is FADU Korea’s Sandisk beta?	Foreign flow says the relationship is forming: foreigners net bought about KRW 445.1bn of FADU since May, but return correlation is not yet as structural as SK Hynix-Micron.	FADU and Sandisk AI storage beta
What does FADU 2Q26 need to prove for the AI-storage thesis?	The updated base case is KRW 69-72bn of revenue and KRW 11-12.5bn of operating profit. That is above the KRW 60.5bn / KRW 9.3bn consensus, but it is not a mega-surprise because the KRW 150.8bn gross disclosed 2Q contract amount is not all recognized in 2Q.	FADU 2Q26 earnings preview
Can FADU become a better AI infrastructure bottleneck than memory?	Possibly, but it must prove P (price), Q (volume), C (cost) and new segments such as Gen6, PMIC and CXL together. The June 2 note is the long-duration framework; the July 4 preview resets the near-term 2Q proof bar more conservatively.	FADU P, Q and new-segment check
Why is HBM a bottleneck from a chip-design perspective?	AI performance is less about headline FLOPS and more about how close data sits to compute. HBM is the common memory bottleneck for GPUs and custom ASICs, and it pulls FC-BGA, PCBs and power-integrity parts into the same thesis.	AI chip design and the data-movement bottleneck
What do Marvell and Broadcom earnings mean for HBM?	If custom AI chips and AI Ethernet networking stay strong, HBM demand broadens beyond Nvidia GPUs into Broadcom XPUs, Google TPUs, OpenAI accelerators and Marvell custom silicon.	Marvell and Broadcom earnings preview for Korea’s AI bottlenecks
What does Marvell Q1 FY2027 mean for Korean semiconductors?	The key is not a simple HBM-only trade. Custom XPU, optical interconnect and scale-up networking push the bottleneck into FCBGA, MLCC, silicon capacitors and test sockets.	Marvell Q1 FY2027 and Korean semiconductors
What do Marvell’s trillion-dollar story and Broadcom’s call mean for HBM investors?	Strong custom XPU and AI Ethernet/networking demand broadens HBM exposure beyond Nvidia GPUs into Marvell custom silicon, Broadcom XPUs, TPUs and OpenAI accelerators. But Korea’s first-order bottlenecks are FC-BGA, AI network PCBs, Si-Cap and power integrity.	Marvell’s trillion-dollar story and the Broadcom read-through
Has higher-rate pressure already broken the HBM / AI-memory cycle?	Not yet. The observed fact is not AI-demand collapse, but a yellow light: rates, capital costs and data-center financing are starting to cap AI infrastructure multiples. For memory, watch HBM pricing, EPS durability, customer inventory and capex guidance cuts.	AI Supercycle Midgame
Why can Samsung Electronics look cheap while Samsung Electro-Mechanics looks expensive inside the same AI cycle?	Multiples are not driven by AI exposure alone. They reflect earnings duration, pricing power, customer lock-in, capex burden and peak-earnings doubt. Samsung offers HBM catch-up plus AI memory-hierarchy optionality at a lower multiple, while SEMCO’s MLCC / FC-BGA / Si-Cap thesis is already priced aggressively.	AI Infrastructure Multiple Map
Does SEMCO’s KRW 2.8M report prove it is stronger than memory?	Not yet. The June 1 target is 2029F EPS of KRW 93,242 multiplied by 30x, so it pulls a long-term earnings path into today’s valuation. SEMCO must prove 2027-2029 ASP durability, Si-Cap repeat orders and high OPM to deserve a premium versus memory.	SEMCO KRW 2.8M Follow-Up
How does SEMCO’s KRW 100T market cap connect to the HBM thesis?	As HBM packages grow, power integrity and FC-BGA become parallel bottlenecks. SEMCO at KRW 100T is a signal that the market is starting to price the layer below HBM, not just HBM itself.	SEMCO at KRW 100T
How do AI server passive components connect to HBM?	Larger GPU/HBM packages create bigger transient power swings. MLCCs, silicon capacitors and inductors become the power-integrity layer that lets HBM and GPU packages run reliably.	AI server passive-component bottleneck
What does Korea’s global No. 6 market-cap headline mean for HBM?	It is not a buy button. It is a signal to verify Samsung + SK hynix AI-memory earnings durability and KOSPI concentration risk.	Why Korea Part 5: Global No. 6 Market Cap
Does foreign selling in Samsung and SK Hynix kill the HBM thesis?	Not by itself. Most 2026 foreign selling was concentrated in the two memory mega-caps, but domestic liquidity absorbed the supply. The discipline is to wait for foreign re-buying or ownership recovery before chasing.	Korea Foreign Investor Flow Analysis
Why is KOSPI foreign ownership high while Samsung and SK Hynix ownership is low?	KOSPI ownership is market-value weighted, so it can stay high after a price rally. Samsung and SK Hynix share-count ownership is already near 2026 lows, making the five-day foreign-selling speed the key timing signal.	KOSPI ownership versus Samsung / SK Hynix
Why focus on second-line semiconductor candidates if KOSPI is still strong?	Korea’s aggregate 20-day ADR has fallen to 67.3, so this is narrow leadership rather than a broad market. Instead of chasing HBM mega-caps, the next signal is turnover and flow in second-line candidates such as HPSP, SFA Semiconductor and Hana Micron.	Korea ADR at 67 and narrow leadership
What is the real question behind Samsung’s DS stock-bonus buyback?	Two questions matter: whether the KRW 200T condition refers to annual or cumulative DS operating profit, and whether the stock-bonus buyback lands closer to KRW 12.6T or KRW 80T.	Samsung stock-bonus buyback analysis
How can foreign investors access Korean AI memory stocks?	Start with the access map, then move from KOSPI heavyweights into HBM, equipment and substrate names. Broker availability depends on country and account type.	Korea Stocks for Foreign Investors Hub
Why is KOSPI being re-rated in 2026?	HBM earnings, governance reform and global allocation away from crowded US exposure are working together to compress Korea’s discount.	Korea 2026 Re-Rating
Why did Samsung jump +14.4% and SK Hynix +10.6% on May 6?	₩3.1T foreign net buying on Samsung in a single session, re-pricing “AI memory bottleneck at NVIDIA / TSMC discount.” Spread to AI substrate (Daeduck +9.62%, Simmtech +6.35%) and front-end equipment (Wonik IPS, Eugene Tech, KC Tech).	Korean Semis Rally May 6: Samsung + SK Hynix + Substrate + Equipment

Latest Sector Updates

Date	Topic	Read
2026-07-22	FADU’s Q2 surprise and Gen6 eSSD moat: compares KRW 73.16bn of revenue, KRW 16.27bn of operating profit and a 22.2% margin with the prior preview, then evaluates controller-firmware integration, OCP FDP, FC6161, QLC, customer concentration, revised EPS and flows. The moat scores 3.0/5, with Gen6 production and a QLC customer design win as the next proof points.	FADU’s Q2 Surprise and Its Real Moat
2026-07-19	NVIDIA Vera CPU expansion and Simmtech: NVIDIA confirms 36 Vera CPUs and 54 TB of LPDDR5X in NVL72 plus standalone racks scaling to 256 CPUs. TrendForce’s reported SoCAMM capacity cut is a supply response designed to raise total CPU and module shipments, not a demand cut. Simmtech’s demand equation is closer to Vera CPU units × modules × PCB × share than to DRAM bits. At KRW 107,400 on July 16, 20-day institutional buying of KRW 236.2bn was still opposed by KRW 161.1bn of foreign selling, so the business thesis and price bottom remain separate questions.	NVIDIA Vera CPU Expansion and Simmtech
2026-07-19	Weekly semiconductor deep dive: integrates the July 13-19 TSMC and ASML results, memory pricing, HBM LTAs, Chinese-memory policy, Kimi K3 efficiency and Korea’s semiconductor selloff. Physical scarcity remains intact for 2026-2027, but the market is discounting 2028 supply, efficiency and customer profitability. The central path is strong earnings coexisting with lower multiples.	Weekly Semiconductor Deep Dive
2026-07-17	US House letter seeking Chinese-memory procurement restrictions: verifies the official request for stronger YMTC controls, an expedited CXMT Entity List review, procurement limits for AI, data centers, federal IT and critical infrastructure, and allied coordination. It is not yet regulation, but implementation would primarily shrink the 2027-2028 Chinese supply-overhang discount, directly protecting Samsung’s broad portfolio and extending the duration of SK Hynix’s HBM advantage.	US House Letter on Chinese Memory Procurement
2026-07-09	US-China agentic inference infrastructure split: The US is moving toward HBM plus SRAM/LPU to reduce token cost per watt, while China works around advanced logic and HBM constraints through optical mesh and parallel scale-out. The Korean landing is cleaner in Samsung’s HBM4E/SRAM-LPU foundry option, SK Hynix HBM4, HD Hyundai Electric power equipment and Hanmi HBM packaging tools than in China optical modules.	US-China inference infrastructure split
2026-07-09	Samsung and SK Hynix 2028E profit valuation: compares MarketScreener 2028E point estimates with a supply, HBM LTA and historical drawdown scenario model. The stocks look cheap on headline numbers, but the real question is whether 2028E is normalized or peak earnings; relative asymmetry is slightly better in Samsung	Samsung and SK Hynix 2028E valuation
2026-07-09	AI token value and memory value added: decomposes one AI token dollar across models, cloud, GPUs, memory, foundry, power and applications. Memory is an HBM bottleneck, but its token-dollar economic share is small, so the key test is HBM4 contract premium and KRW 100-140tn mid-cycle net income, not low P/E alone	AI token value and memory value added
2026-07-04	SK Hynix ADR price transmission and leveraged ETF plumbing — analyzes SKHY not simply as a listing catalyst, but as a flow channel connecting 000660 common shares, ADR-common-share arbitrage, U.S. price discovery and Korea’s single-stock 2x ETF rebalancing. The medium-term access story is positive, but the first test is offer price and first-week premium or discount	SK Hynix ADR price transmission
2026-07-04	SK Hynix F-1/A and F-6 — reads the SKHY ADR process not as a simple listing wrapper, but as a U.S. AI-memory investor access event. The F-6 confirms 1 ADS = 0.1 common share; the board-authorized new-share ceiling is 17.79mn shares; the filing packages Yongin Fab 1, Cheongju P&T7, EUV capex, HBM, Custom HBM and eSSD into a U.S. prospectus format	SK Hynix F-1/A and F-6
2026-07-04	FADU 2Q26 earnings preview — combines 1Q financials, Q2 supply contracts, June amendments and revenue-recognition cutoffs to estimate Q2 revenue of KRW 69-72bn and operating profit of KRW 11-12.5bn. The conclusion is a clear beat versus consensus, but not a mega-surprise	FADU 2Q26 earnings preview
2026-06-30	H1 2026 AI infrastructure bottlenecks and narrow markets — Korea’s index surge came with weak breadth, while U.S. AI leadership moved from GPU-only narratives toward memory, storage, servers, equipment and power. H2 discipline is to separate recurring-revenue bottlenecks from option-like themes	H1 2026 Review
2026-06-29	Samsung 2Q26 Core OP preview — Micron’s FY3Q26 surprise was reflected in Samsung and SK Hynix for one session, then largely erased. The next test is not headline OP but Core OP after adding back bonus accruals and whether 3Q OP can stay near KRW 100tn	Samsung 2Q26 Preview
2026-06-22	Samsung FCF return and DS bonus buyback flow — reframes Samsung’s shareholder return away from headline dividend yield and toward FY2026E net income of KRW 300.2tn, the 50% FCF return policy, and a possible KRW 20-22tn after-tax DS bonus treasury-share purchase flow	Samsung Is Not A Dividend Stock
2026-06-22	KOSDAQ recovery after Samsung/SK Hynix crowding — argues that KOSDAQ needs sideways Samsung Electronics and SK hynix, not a mega-cap memory selloff. The key window is after Micron’s Jun 24 print and the July Samsung/SK hynix Q2 expectation reset, with retail capital and KOSDAQ semiconductor test/equipment/metrology turnover as the confirmation checks	KOSDAQ recovery after Samsung/SK Hynix crowding
2026-06-21	CXMT IPO and memory price risk — reads CXMT’s STAR Market IPO and China DRAM supply expansion not as an immediate HBM bear case, but as a structural supply event that can cap client DDR5, LPDDR and consumer NAND pricing. The 2026 risk is price-momentum peak-out; the 2027 risk is bifurcation between HBM / AI server memory and client DRAM / NAND	CXMT IPO And Memory Price Risk
2026-06-18	HBM4E 12-high race — separates Samsung’s earlier HBM4E sample announcement and SK hynix’s 12-high sample supply into product specs, NVIDIA and AMD customer channels, current share and the 3Q-4Q 2026 proof window	HBM4E 12-high race: Samsung’s re-entry vs SK hynix supply lock-in
2026-06-12	Why the AI supercycle is getting longer — agent demand, IPO funding, memory and storage — agent-driven token demand → expanding big-tech financing → SpaceX/Anthropic-scale IPOs opening the public market as a third funding relay, extending the cycle’s lifespan. With Nomura’s KRW 590,000 Samsung and KRW 5,000,000 SK Hynix targets as a milestone, the strongest and still-cheapest assets are Korean memory at 0.80-0.82x relative P/E and early-stage storage reclassification	Why the AI supercycle is getting longer
2026-06-12	Korea theme ETF rebalance flow — the first KR Theme ETF Rebalance Flow v1 run scanned 31 ETFs, 30 valid ETFs and 291 constituent rows, producing 69 candidates. The Jun 12 signal was stronger in semiconductor equipment/materials cap redistribution than nuclear/SMR: Leeno, EO Technics, Soulbrain, DB HiTek and Hanmi ranked as buy-pressure candidates, while SK Hynix, Samsung Electro-Mechanics and LS ELECTRIC were cap-trim watches	Korea theme ETF rebalance flow
2026-06-13	ETF flow is leading the Korean market — connects theme ETF redistribution, cap trims, KOSPI big-figure breakout/pullback patterns and the KOSPI200 expiry semiconductor squeeze monitor into one response framework. The key is to separate core semiconductor theses from event flow and confirm ETF candidates on T+1/T+3	ETF flow is leading the Korean market
2026-06-13	Korean semiconductor ETF exposure strategy — combines the Korean top-50 semiconductor market-cap proxy with full ETF exposure data to separate Samsung/SK Hynix 90.8% concentration, absolute ETF ownership, market-cap-adjusted ETF sensitivity and under-owned ETF gaps. Mega-cap memory is core exposure; Leeno, DB HiTek, EO Technics, Wonik IPS, HPSP and ISC are ETF-sensitive trades; TCK, Daeduck, Korea Circuit and Doosan Tesna are gap candidates	Korean semiconductor ETF exposure strategy
2026-06-13	Korean semiconductor Top 50 two-month alpha candidates — combines 1,137 Naver ETF constituent scans with local flow data to compress the watchlist for names that could outperform Samsung and SK Hynix over two months: Korea Circuit, HPSP, TES, VM and FADU	Korean semiconductor Top 50 two-month alpha candidates
2026-06-06	AI Supercycle Midgame — argues that the AI memory rally is still earnings-backed, but long rates, data-center financing and hyperscaler FCF are now a yellow-light cap on AI infrastructure multiples. Red flags are capex guidance cuts, HBM price reversal and financing stress	AI Supercycle Midgame
2026-06-05	Jensen Huang’s HBM4 three-vendor qualified comment — reads the comment as Samsung, SK hynix and Micron entering Vera Rubin HBM4 production competition, then separates Samsung HBM-discount compression, SK hynix leadership-premium transition and Hanmi TC-bonder read-through	Jensen Huang’s bigger HBM4 comment
2026-06-05	Sam-Ha-Ma parity follow-up — combines the AI Chip & Memory Forward P/E map with 2Q run-rate OP multiples. The conclusion: Micron deserves a premium, but Samsung and SK hynix near 6x 2028E P/E look too cheap if EPS durability holds	Sam-Ha-Ma parity follow-up
2026-06-03	Sam-Ha-Ma parity — compares Samsung Electronics, SK hynix and Micron on forward PER and finds Samsung/Micron at 0.82x and SK hynix/Micron at 0.80x. The conclusion: SK hynix is buyable in a first tranche, while Samsung is the larger HBM4E/HBM4 catch-up candidate if proof arrives	Sam-Ha-Ma parity
2026-06-02	Marvell’s trillion-dollar story and the Broadcom read-through — interprets Jensen Huang’s Marvell “next trillion-dollar company” call-out and Broadcom’s AI semiconductor call through custom XPU, AI networking, optical interconnect and 3.5D packaging bottlenecks, then maps the Korea read-through to Samsung Electro-Mechanics, Korea Circuit, ISU Petasys, Samsung Electronics and SK Hynix	Marvell’s trillion-dollar story and the Broadcom read-through
2026-06-02	FADU P, Q and new-segment check — sets the original AI-storage bottleneck framework around price, volume, cost and new segments. The July 4 preview updates the near-term earnings bar after June amendments to KRW 69-72bn of Q2 revenue and KRW 11-12.5bn of operating profit	FADU P, Q and new-segment check
2026-06-02	Hanmi Semiconductor IR notice and TC-bonder TAM expansion — interprets the June 1 IR notice as a 2H26 overseas new-market message rather than an order, then checks whether Hanmi can expand from HBM TC bonders into 2.5D packaging, AI system semiconductors, OSAT and HBF, with KRW 300,000, the KRW 315,000 insider purchase zone, foreign buying and easing financial-investment selling as confirmation triggers	Hanmi IR notice and TC-bonder TAM expansion
2026-06-01	Samsung HBM4E 12-high samples — what is the market watching? — reframes Samsung’s official sample shipment into customer qualification, mass-production timing, HBM laggard-discount compression, Hanmi Semiconductor’s TC-bonder option value, and short-selling / flow confirmation signals	Samsung HBM4E 12-high samples
2026-06-01	SEMCO KRW 2.8M follow-up — 2029 is far, and it must prove stronger duration than memory — checks Mirae Asset’s June 1 formula of 2029F EPS KRW 93,242 × 30x and reframes the debate as a 2027-2029 proof test for MLCC +10%, FC-BGA +20%, Si-Cap repeat orders and high OPM	SEMCO KRW 2.8M Follow-Up
2026-06-01	Is Samsung C&T a lagging Samsung Electronics proxy? — tests the trade using 51.7% Samsung Electronics stake-value pass-through, 0.83 YTD beta, 1.10 recent-20D beta, and the choice between Samsung common, Samsung C&T and Samsung Electronics 2x ETFs	Samsung C&T Samsung Electronics proxy / NAV gap trade
2026-05-31	Is FADU Korea’s Sandisk beta? — tests whether FADU is being discovered as a Korean high-beta proxy for AI storage/NAND using KRW 445.1bn of foreign net buying since May, institution selling and FADU-Sandisk return correlations	FADU and Sandisk AI storage beta
2026-05-31	SK Hynix vs Micron — HBM technology premium or U.S. listing premium? — compares two trillion-dollar AI-memory stocks: SK Hynix as the technology/reference/HBM-profitability leader and Micron as the U.S.-listed scarcity proxy with an HBM4, SOCAMM2 and Gen6 SSD story	SK Hynix vs Micron
2026-05-31	Samsung vs SK Hynix Forward P/E inversion — decomposes SK Hynix’s slight 2026E P/E premium into HBM bottleneck value and Samsung’s HBM laggard discount. The conclusion: annual P/E has inverted, 12MF is closer to parity, and the cleaner alpha is Samsung’s HBM4E catch-up trade	Samsung vs SK Hynix Forward P/E Inversion
2026-05-31	AI Infrastructure Multiple Map — why Samsung Electronics looks cheap and Samsung Electro-Mechanics looks expensive — decomposes why GPU, HBM, CPU, MLCC and FC-BGA deserve different multiples inside the same AI capex cycle, using pricing power, LTAs, customer lock-in, capex burden and peak-earnings doubt. Bottom line: Samsung Electronics has the cleaner relative-value case; SEMCO is a thesis with chase risk	AI Infrastructure Multiple Map
2026-05-31	Goldman token demand vs J.P. Morgan memory ASP peak-out — Goldman sees tokens up 24x and cost/token down 60-70%/yr; J.P. Morgan sees DRAM/NAND ASP YoY growth slowing from 2027. Split by P, Q and C they don’t conflict: alpha moves from 2026 memory beta → 2027 price-momentum peak-out → 2028-2030 token-cost-reduction stack	Goldman token demand vs J.P. Morgan memory ASP peak-out
2026-05-30	AI token futures and cost per token — from a performance race to a cost race — Shanghai Futures Exchange is early-designing AI-token-price-linked futures while CME and ICE prepare GPU compute futures; once AI usage gets a market price, the axis shifts from performance to cost per token. The conclusion is to buy not generic AI stocks but the bottlenecks that lower cost per token (custom ASIC → memory / KV-cache → packaging), with Samsung Electronics the most balanced Korea pick	AI token futures and cost per token thesis
2026-05-29	Dell Q1 FY2027 and Korea AI-server margin read-through — the print was a big beat, but AI-server gross margin compressed from 21.6% to 18.1%. Pricing power and durable margin sit upstream with memory (Samsung / SK Hynix) and high-value package/passive components (SEMCO), not with the server OEM	Dell Q1 FY2027 and Korea AI-server margin
2026-05-28	Marvell Q1 FY2027 and Korea semiconductor read-through — Q1 revenue $2.418B, Q2 guide $2.70B ±5%, custom XPU / optical / scale-up networking strength translated into SEMCO FCBGA / MLCC / SiCap, Samsung / SK Hynix memory and test sockets	Marvell Q1 FY2027 and Korean semiconductors
2026-05-27	Korea ADR at 67 and narrow leadership — KOSPI and KOSDAQ 20-day ADRs have both fallen into the 60s; the average stock is weak while money is compressed into AI infrastructure, back-end semis, substrates, shipbuilding and defense	Korea ADR at 67 and KOSPI-KOSDAQ breadth analysis
2026-05-27	Did Samsung signal a memory supercycle through 2028? — reframing the DS stock-bonus issue around the annual-vs-cumulative KRW 200T condition and the possible KRW 12.6T-80T stock-buyback range	Samsung stock-bonus buyback analysis
2026-05-26	AI server passive-component bottlenecks below HBM — a non-engineer explainer for MLCCs, silicon capacitors and inductors as the power-integrity layer that lets GPU/HBM packages run reliably	AI server passive-component bottleneck
2026-05-26	KOSPI foreign ownership versus Samsung / SK Hynix ownership — KOSPI market-value foreign ownership remains high, but Samsung at 48.32% and SK Hynix at 51.62% are already near 2026 lows; the note lays out the stabilization triggers	KOSPI ownership versus Samsung / SK Hynix
2026-05-26	SEMCO at KRW 100T and the power-integrity bottleneck below HBM — using Hyundai Motor and Murata as benchmarks to test whether silicon capacitors and FC-BGA are receiving their own AI multiple	SEMCO at KRW 100T
2026-05-24	Korea market cap near global No. 6 and AI-memory concentration — cross-checking WFE, CEIC, KRX and a market-cap proxy to explain why the headline is mainly a Samsung + SK hynix AI-memory re-rating	Why Korea Part 5: Global No. 6 Market Cap
2026-05-24	Foreign-flow distribution in Korea’s memory mega-caps — breaking down why KRW 84.8T of the KRW 89.2T YTD foreign net selling came from Samsung Electronics and SK hynix	Korea Foreign Investor Flow Analysis
2026-05-24	After NVIDIA, the AI semiconductor bottleneck — translating Dwarkesh/Reiner Pope’s chip-design discussion into an investment frame: AI performance is moving from headline FLOPS into data movement, HBM, FC-BGA, power integrity and testing	AI chip design and Korea’s lower-stack bottlenecks
2026-05-23	Marvell and Broadcom earnings preview — using Broadcom’s $10.7B AI semiconductor guide and Marvell’s 74% data-center revenue mix to test whether Korea’s AI read-through expands from HBM into custom ASICs, Ethernet, optical links and Samsung Electro-Mechanics silicon capacitors	Marvell / Broadcom and Korea’s AI bottlenecks
2026-05-17	US-China summit and technology decoupling — separating Korea’s short-term memory upside from long-term standard fragmentation risk through the H200 refusal, Huawei CloudMatrix and ITIF’s USD 21B Korea absorption estimate	US-China tech decoupling and Korean semiconductors
2026-05-17	Google I/O and NVIDIA earnings preview — separating AI use-case expansion from infrastructure CAPEX confirmation, and what NVIDIA’s Q2 guide and 75% gross margin signal for Korean HBM, substrates and MLCC	Korean semiconductor event preview
2026-05-16	Can Samsung Electronics deserve 15x PER? — the gap between Samsung at ~5x and TSMC at 19-22x, HBM4/foundry/packaging integration and the realistic 8-10x re-rating path	Samsung Electronics vs TSMC re-rating thesis
2026-05-12	AI citizen dividend and semiconductor tax-windfall debate — KOSPI morning selloff and rebound, separating Samsung/SK Hynix EPS risk from policy multiple risk	KOSPI May 12 selloff and rebound
2026-05-11	Citi Samsung Electronics TP ₩460,000 — 1Q DS margin 65.7%; the claim is that the memory-cycle frame itself is wrong this time	Samsung Citi TP ₩460,000 — Memory Re-rating
2026-05-09	Why Korea Part 3 — Samsung / SK hynix ₩300tn+ profit pool upgrading Korean fiscal capacity	Why Korea Part 3
2026-05-07	May 6 rally — Samsung +14.4% / SK Hynix +10.6%, ₩3.1T foreign buying, two-stage spread to substrate + equipment	Korean Semis Rally May 6

Start Here

Step	Question	Read
1	How should investors think about Samsung’s KOSPI weight?	Samsung Electronics Weight in KOSPI: 2026 Index Concentration Explained
2	Where does SK Hynix stand in HBM and AI memory?	SK Hynix HBM Market Share 2026: AI Memory Investor Guide
3	Is SK Hynix’s P/E inversion a Samsung catch-up signal?	Samsung vs SK Hynix Forward P/E Inversion
4	How does Samsung’s HBM4E sample connect differently to the common stock and equipment names?	Samsung HBM4E 12-high samples
5	How does Jensen Huang’s HBM4 three-vendor comment connect differently to Samsung, SK hynix and Hanmi?	Jensen Huang’s bigger HBM4 comment
5A	Who is better positioned in the HBM4E 12-high generation, Samsung or SK hynix?	HBM4E 12-high race: Samsung’s re-entry vs SK hynix supply lock-in
5+	How can semiconductor theme ETF cap rebalancing create second-line equipment flow?	Korea theme ETF rebalance flow: semicap redistribution buy pressure and mega-cap trim pressure
5+A	Which semiconductor names are more sensitive when ETF exposure is adjusted for market cap?	Samsung and SK Hynix Are 90.8% of Korean Semis: Where ETF Flow Is More Sensitive
5+B	Which second-line names could beat Samsung and SK Hynix over the next two months?	Korean Semiconductor Top 50: Which Names Can Beat Samsung and SK Hynix?
5++	How far can Hanmi’s IR notice expand the TC-bonder TAM?	Hanmi IR notice and TC-bonder TAM expansion
6	How do Marvell and Broadcom expand the bottlenecks below HBM?	Marvell’s trillion-dollar story and the Broadcom read-through
7	Are Samsung and SK hynix cheap again versus Micron?	Sam-Ha-Ma Parity: Samsung and Hynix Look Cheap Again Versus Micron
8	Can FADU open price, volume and a new AI storage segment at the same time?	FADU P, Q and new-segment check
9	Is Micron’s premium a HBM technology premium or a U.S. listing premium?	SK Hynix vs Micron: HBM Technology Premium or U.S. Listing Premium?
10	What is the lagging proxy after Samsung Electronics’ rally?	Samsung C&T Samsung Electronics proxy / NAV gap trade
11	What are Samsung’s AI, HBM and foundry options?	Samsung Electronics 2026: AI, HBM and Foundry Deep Dive
12	Why is SK Hynix central to AI infrastructure?	SK Hynix: The HBM Leader Powering the AI Revolution
13	Why is Korea being re-rated?	Korea 2026: Why KOSPI +49% YTD Is a Re-Rating, Not a Rally
14	Is Korea’s global No. 6 market-cap headline a buy signal?	Why Korea Part 5: Global No. 6 Market Cap
15	Is foreign selling a HBM thesis break or position reduction?	Korea Foreign Investor Flow Analysis
16	How much ownership-based foreign selling pressure remains?	KOSPI ownership versus Samsung / SK Hynix
17	Did Samsung signal a memory supercycle through 2028?	Samsung stock-bonus formula analysis
17A	What changes if Samsung shareholder return is read through FCF and DS bonus buyback flow instead of dividend yield?	Samsung Is Not A Dividend Stock
18	Does SEMCO’s KRW 2.8M report prove stronger duration than memory?	SEMCO KRW 2.8M Follow-Up

Read by Thesis

HBM and AI Memory

KOSPI and Index Concentration

AI Hardware Supply Chain

FAQ

What is the difference between Samsung Electronics and SK Hynix as HBM investments?

SK Hynix offers cleaner HBM exposure and stronger visibility into Nvidia-linked demand. Samsung combines HBM recovery potential with a broader memory cycle, foundry optionality and device exposure.

Why does Samsung’s KOSPI weight matter?

Samsung is the largest single component in Korean equity benchmarks. Buying Korea through ETFs often means taking meaningful exposure to Samsung, SK Hynix and the semiconductor cycle.

Is the KOSPI re-rating only about semiconductors?

No. HBM is the strongest earnings engine, but Korea’s 2026 re-rating also reflects governance reform, rising shareholder-return pressure and global allocation away from crowded US exposure.

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.

Direct-Answer FAQ Pages

For quick answers to natural-language questions about SK hynix and HBM:

What’s SK hynix’s HBM market share in 2026? → FAQ in the SK hynix HBM analysis
Is SK hynix publicly traded? / Who owns SK hynix? → same FAQ block above
Does SK hynix supply HBM to NVIDIA? → same FAQ block above
Is Samsung Electronics publicly traded? / Does Samsung make Apple’s iPhone screens? → FAQ in the Samsung Electronics deep dive