Exclusive Analysis Hub: Proprietary Korea Market Screens and Flow Work

Korea Invest Insights

Purpose

This hub collects Korea Invest Insights posts built from proprietary or internally constructed datasets. These are not simple news summaries. They use ETF constituent scans, foreign and institutional flow, real-money classifications, breadth, liquidity and relative-strength screens.

Question	Data used	Goal
Where is money moving?	Foreign, institutional, retail and sub-investor flow	Separate raw buying from quality and persistence
Where can ETF flow move prices?	ETF constituents, weights and market-cap-adjusted exposure	Identify mechanical flow and second-line redistribution candidates
Why is the market narrow?	ADR, turnover, relative strength and flow	Separate leadership from average-stock weakness

Recent Exclusive Analysis

Date	Topic	Key takeaway	Post
2026-07-31	The AI demand was real, and the bill went to memory: a July earnings season wrap	An independent wrap of the July season. All three clouds accelerated (Google +82%, Azure +43%, AWS +36.7%) and every management team called out supply constraints, yet the market began grading capex on same-quarter revenue proof, rewarding only Microsoft (+15%, its best day since 2008) and Amazon. The hidden protagonist is memory pricing: it became Amazon’s stated reason for raising capex, Apple called it a hundred-year flood, and the recipients on the other side are Samsung (89.5tn won quarterly operating profit) and SK Hynix (60.5tn won at a 76% margin). The same month stacked two record listings, SK Hynix’s $26.5bn Nasdaq ADR and CXMT’s $8.6bn Shanghai debut, while Kimi K3 left behind not price destruction but a reframing: bigger models eat more memory. Verdicts on the four worries: token demand refuted by measurement, frontier labs at a fork, the capex-ROI paradox intact, and post-2028 memory profitability undecided, a contest between the LTA-prepayment structure and the CXMT capacity variable.	July Earnings Season Wrap
2026-07-29	Rebound or terminal? Eight doubts, seven discriminators, and a 48-hour verdict	We dissect why the market doubts AI capex can last into four clusters and eight reasons, grading the evidence. The weight sits on two: funding descending to vendor financing (measured, so the discount is legitimate) and the missing end payer (a fear that contradicts the gap ratio narrowing 8x to 4.6x and token revenue growing +100-250%). The market is looking at 1999’s financial structure and 1996’s demand data at once, and pricing the 1999 side because positioning broke. Scoring crash-amid-chaos history on seven discriminators gives rebound 5, terminal 0, undecided 2; tops are made in euphoria, bottoms in chaos. The coordinate is the overlap of October 1998 and mid-1999. The CDS dashboard (Oracle 215bp, NVIDIA 82bp) is largely hedging technicals, making a GE-2018-style retracement the base rate, and the season’s strongest single trigger is Microsoft’s equity-method line (the $11.5bn back-solve dates to Jul-Sep 2025; April’s print was effectively zero). The two undecided cells get filled within 48 hours by the FOMC and earnings.	Rebound or Terminal?
2026-07-29	Why does NVIDIA want to be OpenAI’s guarantor? Anatomy of the $250bn backstop	We dissect the WSJ report that NVIDIA would guarantee up to $250bn of lease and construction debt for OpenAI’s 10GW Piketon, Ohio campus (SB Energy, phase 1 800MW in 2028) plus a separate $350bn of chip financing. OpenAI’s lens: ~$25bn annualized revenue, ~$14bn annual losses, no rating and foundation governance mean debt markets said no, and every alternative (more equity, IPO, the Oracle path, an SPV, GPU-backed debt) ultimately requires someone’s investment-grade credit. NVIDIA’s lens: a guarantee is the cheapest-looking choice, no cash out, but it equals 159% of equity and 71x the disclosed program cap. Success is defined as the guarantee expiring undrawn; Lucent lost 99.3% on 24%-of-revenue exposure. We list the market’s five concerns (the circle completed, the debt market’s refusal, accounting opacity, concentration repricing, the historical base rate) and six dispelling conditions. A turnaround dispels the concerns about this deal, not about this structure.	Anatomy of the $250 Billion Backstop
2026-07-28	Anatomy of Black Tuesday: is China’s DUV mass-production report a major negative or noise	The day KOSPI fell 9.65% and tripped the year’s eighth circuit breaker, triggered by a report of China mass-producing domestic immersion DUV tools (Yuliangsheng, NA 1.35, comparable to ASML’s 2008 platform) and CXMT’s 466% debut. We steelman the negative case (export-control premium erosion, the migration insurance on a ~1,400-tool stockpiled fleet, the subsidy playbook) and the noise case (25 units over two years is under a fifth of ASML’s annual immersion output, the three commercial gates undisclosed, zero sell-side cuts), then judge on four axes. Equipment names 60:40 genuinely negative (right direction, excessive speed), Samsung 30:70, SK Hynix 15:85 (no HBM transmission channel; -12.89% was flow contagion). The crash attributes 20 to information and 80 to positioning, evidenced by beta-proportional losses and rising estimates. Discriminators: teardown-grade spec verification, CXMT G5 shipments, MATCH Act passage, Japan’s parts controls, and as ever the DRAM contract price.	Anatomy of Black Tuesday
2026-07-27	Cisco did not die of missing demand: the dot-com vs AI comparison rewritten with the monetization gap ratio	End demand was real in 1999 too. What cut Cisco down 89% was the myth metric (doubling every 100 days), marginal-buyer credit (CLEC junk bonds and $25bn of vendor financing) and price deflation betraying revenue (bandwidth -90%, dark fiber under 5% lit). We propose the monetization gap ratio, infrastructure spend over end AI revenue, as the single discriminating number. Dot-com held at 6-10x until rupture; AI has narrowed from 8x (2024) to 4.6x (2026), a 1996 signal. Against that, OpenAI’s $1.4tn commitment stack (~35x coverage) and NVIDIA’s reported talks to guarantee $250bn of Ohio 10GW campus debt plus $350bn of chip financing are a 1999 signal. Verdict: a mid-1999 analog, with vendor financing historically 12-18 months ahead of rupture. The labor audit finds job-category replacement proven (23% of US layoffs AI-attributed), profit conversion unproven (Gartner) and full replacement narrative-stage. For Korea: suppliers fell harder than Cisco in 2001 (JDS Uniphase, Nortel ~-99%), so buyer-mix credit quality joins the DRAM contract price as the discriminator.	Cisco Did Not Die of Missing Demand
2026-07-26	Dissecting $950bn: three things the San Francisco AI Declaration changed, and one it did not	The declaration was issued by President Lee Jae-myung on behalf of Korea, not the US, and no US government officials appear on the confirmed attendee list. The $950bn is SK’s $750bn plus Samsung-Broadcom’s $200bn, mostly LOIs and MOUs, with Reuters and Fortune explicitly calling the Samsung deal a non-binding statement of intent. It is a multi-year notional figure larger than the entire $889.3bn global memory market forecast for 2026. With HBM at 35-55% of accelerator cost, Korea holding ~79% of HBM and ~67% of DRAM, and memory at ~30% of hyperscaler capex, roughly 20-25% of every AI capex dollar accrues to Korea and almost all of it is memory. What is new: that share locked through 2030, Samsung Foundry winning Broadcom as an anchor tenant (a segment still losing KRW 2.8tn a quarter could flip sign), and Korea joining the circular financing loop where NVIDIA invests in Naver and Naver buys NVIDIA. None of the declaration’s four pillars claims the model layer, and HyperCLOVA X moves from disqualified Qwen weights to NVIDIA Nemotron. The market has not priced any of it after Friday’s oil-driven 5.72% crash.	Dissecting $950 Billion
2026-07-23	Re-ranking 19 Korean AI data-center stocks: direct exposure is not the same as a good entry	We separate operators, facility and power suppliers, and server components, then score direct exposure at 40%, foreign and real-money flow at 35%, and valuation and entry position at 25%. GnCenergy and LS ELECTRIC lead on directness but are no-chase names after sharp gains, high valuation and foreign selling. The current risk-adjusted order is LG Electronics, Iljin Electric, Samsung SDS, GST and SK Gas.	Re-Ranking 19 Korean AI Data-Center Stocks
2026-07-23	AI fundamentals are solid, the problem is rates, the trigger is oil: from Brent at $94 to Korean memory	Of Brent at $94, physical supply-demand justifies $74-78; the other $16-20 is war premium. Inventories rose 1.4M barrels and Hormuz stays open. That fear component travels through the August CPI ($10/bbl = +0.2pp) and the September FOMC (~68% hike odds) into AI multiples. The 4.7% yield runs on three engines: oil, policy repricing and real yields (30Y TIPS at 2.95%, highest since 2008), with AI CAPEX sitting on both sides of the rate as cause and casualty in a self-limiting loop. Cloud profit deceleration (2027 +15-22%) is accounting-certain; only the slope is open. The chain reaches Korean memory through multiple, flows and cost channels, but 2026-2027 earnings are set by contracts, so selling Samsung and SK Hynix on rates is a category error. Verdict: the 48-hour triple reading from the small hours of July 30.	From Oil to Korean Memory
2026-07-23	Alphabet Q2: Cloud +82% ends the demand debate, negative FCF starts the cash debate	Cloud grew 82% (consensus +64%), backlog reached $514bn (+$52bn) and existing customers consume 50%+ above commitments, cutting the odds of a post-2028 demand cliff further. The strongest evidence is behavioral: short of capacity, Alphabet began renting compute from SpaceX at ~$920M a month. The same print delivered the first negative quarterly FCF (-$5.9bn), a 2026 CAPEX guide raised to up to $205bn and a ~$257bn 2027 consensus. Verdict: the demand debate is effectively over, the FCF turnaround slips to a conditional 2028-2029, and for memory this is a volume tailwind while ASP and peak margins remain separate questions. Next scorecards: Microsoft July 30 and Amazon July 31.	Alphabet’s Q2
2026-07-22	FADU’s Q2 surprise and Gen6 enterprise SSD controller moat	Q2 revenue of KRW 73.16bn beat consensus by 16.1%, while KRW 16.27bn of operating profit beat by 89.2%. The 22.2% operating margin matters more than the revenue beat. Gen5 revenue and the OCP FDP listing prove commercialization, but Gen6 qualification and production, 64TB/128TB QLC adoption and lower customer concentration remain unconfirmed. The current moat scores 3.0/5, narrow but expandable.	FADU’s Q2 Surprise and Its Real Moat
2026-07-19	Who burns all those tokens? NVIDIA’s customer map, sovereign AI and Codex at 9 million	The final weak link of the AI CAPEX debate, end demand, is starting to get numbers. NVIDIA’s 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 hyperscale’s $38bn and quarterly growth flipped, 31% vs 12%. Sovereign AI revenue topped $30bn in FY2026, more than tripling. Meritz’s calls on sovereign purchases and a mid-July server DRAM re-tightening are cross-confirmed by TrendForce’s +13-18% contract-price forecast, Inventec’s 40-week lead times and Micron meeting only 50-66% of key-customer demand. Codex added three million users in the three days after the GPT-5.6 release to reach nine million, shifting the demand unit from MAU to agent execution time. Probability updates wait for the Jul 30 earnings window and NVIDIA’s late-August print.	Who Burns All Those Tokens?
2026-07-19	NVIDIA Vera CPU expansion and Simmtech: SoCAMM2 is a module-unit thesis, not only a bit-demand thesis	TrendForce’s reported SoCAMM capacity reduction is a supply response that lowers bits per CPU while increasing total Vera CPU and module shipments. Simmtech’s PCB volume is therefore more sensitive to Vera CPU units × module count × share than to aggregate DRAM bits. Its official product exposure is confirmed, but supply to all three memory makers and leading share remain analyst estimates. With the stock at KRW 107,400 on July 16 and foreign/program selling still active, the rating is Conditional Buy after support and flow confirmation.	NVIDIA Vera CPU Expansion and Simmtech
2026-07-19	Weekly semiconductor deep dive: stronger earnings, lower multiples and the 2028 pivot	TSMC, ASML and memory prices confirm strong 2026-2027 physical demand, while the same capacity plans reserve a 2028 supply response. Integrating HBM LTAs, China’s three roles, open-model efficiency and Korea’s flow shock produces a 25/40/25/10 scenario map. The central path is strong earnings with a lower multiple, judged through contract prices, 2028 EPS, big-tech monetization and relative strength.	Weekly Semiconductor Deep Dive
2026-07-19	SK Hynix Chairman Chey Tae-won’s two months: the company strengthens, the margin peak passes	We decompose SK Hynix Chairman Chey Tae-won’s remarks and actions from the June 2 Computex to the July 17 Jeju forum through information edge and vested interest. He holds top-tier information on demand direction and supply constraints but not on fair value or long-term margin. His words strongly support an AI memory volume supercycle, yet a roughly KRW 40 trillion new-share issuance at the peak, a doubling of capacity and a token-cost-cut goal point to a volume-conversion strategy rather than perpetuating excess margins. The company thesis strengthens while the stock’s margin thesis weakens. The call is Hold, Conditional Add, No Chase, and the single discriminator remains the DRAM contract price.	SK Hynix Chairman Chey Tae-won’s Two Months
2026-07-18	Will AI memory demand exceed expectations? Demand scenarios and the supply map	AI memory demand is likely to meet or exceed today’s high expectations (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. We lay out six upside variables (GPU and ASIC together, spread beyond HBM, SOCAMM2, agents, China’s low-end-chip strategy, usage over efficiency) and six downside risks (CAPEX ROI, efficiency, utilization, power, demand destruction, China supply), then connect the single-wafer competition across memory layers (HBM consuming roughly 3x the wafers of DDR5), the site-by-site expansion maps of Samsung, SK Hynix and Micron, and the substance of Chairman Chey’s expansion argument. The conclusion is a shortage thesis through 2027 and re-verification from 2028.	Will AI Memory Demand Exceed Expectations?
2026-07-18	China’s memory localization and Korea: decomposing the big three’s China exposure	Semiconductors are 38.7% of Korea’s H1 exports and China is the largest market. Decomposed from disclosures, Samsung is at 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 pressuring global ASP. CXMT’s RMB 57.9bn raise brought Alibaba Cloud, Xiaomi and NIO in as shareholders, securing funding and customers at once. The conclusion is a discount variable for 2028 earnings and terminal multiples, not a 2026 earnings risk.	China’s Memory Localization and Korea
2026-07-18	Is China’s AI API pricing sustainable? Verifying the cost structure through disclosures	DeepSeek V4-Pro’s output costs 34 times less than GPT-5.6 Sol. Hong Kong disclosures split the answer. Zhipu’s API gross margin improved to 18.9% (from 3.3%), clearing marginal inference cost, and demand still exceeded supply after an 83% price increase in March 2026. Yet total gross profit covers only 9.3% of R&D (MiniMax 7.9%), so fully loaded cost is not recovered. The cost-advantage ranking runs model architecture and batching/caching > cloud utilization > thin target margins > power and domestic NPUs, with the electricity gap worth only about 2% of compute price. Unlike EVs, switching costs are low, so any winner-take-all emerges in the cloud layer rather than in models. A low-price oligopoly at 60% is the most likely path.	Is China’s AI API Pricing Sustainable?
2026-07-17	US House letter seeking restrictions on Chinese memory procurement	The July 16 bipartisan House Select Committee letter is a policy request, not enacted regulation. It asks for stronger YMTC controls, an expedited CXMT Entity List review, procurement restrictions for AI, data centers, federal IT and critical infrastructure, and coordination with Korea, Japan and the EU. If implemented, the larger effect is not 2026 HBM volume but a lower 2027-2028 Chinese supply-overhang discount. The benefit is more direct for Samsung’s broad portfolio and works as duration protection for SK Hynix’s HBM advantage.	US House Letter on Chinese Memory Procurement
2026-07-17	China’s open-model convergence: value-chain redistribution, not demand collapse	Chinese open models converging on US frontier performance while served on domestic Chinese inference infrastructure is value-chain redistribution rather than an AI demand collapse. DeepSeek V4-Pro proved the efficiency case at 27% FLOPs and 10% KV cache per token, yet TSMC raised 2026 CAPEX to $60-64bn and Microsoft’s token usage rose 30% QoQ. Cost per unit of intelligence has fallen; total silicon spend falling has not been demonstrated. Relative preference runs Samsung > SK Hynix, Micron and SanDisk > NVIDIA, Meta and Amazon > pure model vendors. CXMT’s HBM sits 1.5-2 generations behind, and the diffusion of Chinese model technology is separate from revenue diffusion for Chinese API vendors.	China’s Open-Model Convergence and Value-Chain Redistribution
2026-07-17	Semiconductor fair value: pricing with FCFE and normalized earnings	Samsung at 3.9x and SK Hynix at 4.3x on 2028 consensus means the market doubts the duration of those earnings, not their existence. Back-solving a normalized 8x PER implies the market only allows EPS 52% and 46% below consensus respectively. A share price is not set by either 2028 or 2029 EPS but by 2026-2028 FCFE plus a normalized 2029 terminal value. Probability-weighted fair values come to KRW 385,000 for Samsung (+51%), KRW 1,950,000 for SK Hynix (+6%) and $1,140 for Micron (+34%). With three-year FCF covering 26% of value at Samsung versus 17% at SK Hynix, risk-adjusted attractiveness ranks Samsung, Micron, then SK Hynix.	Semiconductor Fair Value with FCFE and Normalized Earnings
2026-07-17	Kimi K3, the AI price curve and semiconductor demand	Kimi K3 combines 2.8T parameters, a 1M-token context and 16 active experts out of 896 at $3 input and $15 output per million tokens, matching Sonnet 5’s standard price. Kimi Linear-style cache efficiency and quantization reduce GPU and HBM intensity per request, while 64+ accelerator supernodes and open-weight self-hosting can increase total hardware deployment. The near-term efficiency risk for NVIDIA and HBM competes with medium-term demand for server DRAM, enterprise SSDs, networking and multi-model cloud infrastructure. The July 27 weights, license and external throughput are the proof points.	Kimi K3 Resets the AI Price Curve
2026-07-17	Semiconductor bull/bear integrated thesis: four physical clocks and one stock-price clock	The semiconductor debate is no longer about whether AI demand exists but about time, price and cost of capital. Because the four physical clocks (price, investment, construction, monetization) and the stock-price clock run at different speeds, strong results and CAPEX raises can become selling catalysts. We develop the bull (TSMC CAPEX $60-64bn, HBM crowding-out, oligopoly and LTAs) and the bear (pull-in payback, supply response, Oracle FCF -$23.7bn and CoreWeave’s cost of capital, SOXX -19%) each in their strongest form, and contrast how both sides read eleven identical pieces of evidence in opposite directions. 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
2026-07-14	IBM and Ericsson show memory prices reaching the real economy	IBM’s preliminary Q2 print missed at $17.2bn revenue and $2.93 adjusted EPS, and the stock fell more than 17% premarket. One stated reason is decisive: CEO Krishna wrote that in the last few weeks of June, clients shifted capex toward servers, storage and memory to secure supply-constrained infrastructure ahead of expected price increases. The same day, Ericsson warned on margin pressure from AI-driven component costs. This is bullish evidence that memory demand is confirmed by end-customer behavior rather than supplier claims, and simultaneously a warning that the cost is paid in IT budgets and a possible Q3 pull-in payback.	Why IBM’s Earnings Miss Is Evidence of Memory Strength
2026-07-14	The supply gap spilling beyond HBM: three Hana Securities reports as one story	Three Hana Securities reports (July 3 package substrates, July 8 ASICLAND, July 14 GigaDevice) synthesized into one frame. The supply gap from the big-three’s shift to HBM and DDR5 spills into package substrates (Daeduck Electronics, Simmtech, Haesung DS, where larger and higher-layer boards erode effective capacity), eSSD controllers (ASICLAND, FADU, with SK Hynix picking ASICLAND for Gen6), and legacy memory (GigaDevice, SLC NAND ASP up ~5x and legacy DRAM ~10x). It is evidence the memory boom has broadened beyond HBM, and the checkpoints are substrate LTA filings, FADU revenue recognition, and China’s capacity-fill pace.	The Supply Gap Spilling Beyond HBM
2026-07-14	Is any stock more attractive than Samsung on a 6-month view? Axis diversification	No single stock replaces Samsung Electronics outright. Synthesizing two independent screens (a risk-adjusted stock screen and a six-axis screen with a memory-uncorrelation lens), the answer is diversification by axis, not replacement. It converges on four axes: non-memory diversification (Hyundai Rotem, top foreign buying at KRW 330.5bn over 20 days; KB Financial, KRW 1.2tn buyback cancellation), consumer absolute return (Samyang Foods, Dalba Global at 53% ROE), asymmetric semiconductor alpha (Hana Materials, Daeduck Electronics, Korea Circuit), and relative value (Samsung preferred at a 30.7% discount). The practical implication is trimming Samsung’s ~63% concentration while layering independent alpha.	Beyond Samsung: 6-Month Diversification Picks
2026-07-14	SK Hynix Q2 earnings cut versus held target prices	Mirae Asset and Korea Investment & Securities both cut SK Hynix Q2 operating profit (Mirae KRW 62.3tn, KIS KRW 60.4tn). This is not a demand slowdown but a normalization: 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. KIS’s KRW 3.8m ties to its formula better; Mirae’s KRW 4.2m implies a more honest quant target near KRW 3.77m.	SK Hynix Q2 Earnings Cut but Target Prices Held
2026-07-13	HBM 2030 supply-demand deep research: cross-checking the 26.7EB demand model	We independently reproduce the widely-shared 2030 claim of 26.7EB HBM demand versus 10.6EB supply (a 2.5x shortage) from the published formula. The arithmetic reproduces, but it is a bull scenario that needs 24x tokens, 5x model scale, 4x context and 70% KV residency to hold at once. Synthesizing three lenses (demand-model structure, Goldman/DeepSeek/TurboQuant counter-evidence, Big 3 capacity schedule), tight through 2027 and supply relief from 2028 are robust, while a precise 2.5x in 2030 is bull-biased. The point is to separate confidence in direction from uncertainty in magnitude.	HBM 2030 Supply-Demand Deep Research
2026-07-10	Reverse-engineering the 2027 consensus dispersion: the worst case is already the price	Samsung at KRW 285,000 divided by the lowest 2027 EPS estimate of KRW 24,323 gives 11.72x; SK Hynix at KRW 2,180,000 over the lowest estimate of KRW 186,357 gives 11.70x. Both trade at the street’s most bearish scenario times a normal mid-cycle multiple. The buy trigger is not “cheap versus the average” but two consecutive months of low-estimate upgrades plus high/low dispersion narrowing below 3x, judged against late-July big-tech 2027 CAPEX commentary and SK Hynix’s Q2 LTA price-floor disclosure.	The Worst Case Is Already the Price
2026-07-09	US-China agentic inference split and the SRAM opportunity	The US is combining HBM and SRAM/LPU to reduce token cost per watt, while China is working around leading-edge logic and HBM constraints through optical mesh and parallel scale-out. Korean listed-equity opportunities are cleaner in Samsung HBM4E/SRAM-LPU foundry, SK Hynix HBM4, HD Hyundai Electric power equipment and Hanmi packaging tools than in China optical modules.	US-China inference infrastructure split
2026-07-09	Samsung and SK Hynix 2028E profit valuation	On MarketScreener 2028E net income, Samsung and SK Hynix trade at about 4.1x and 4.4x 2028E P/E and look very cheap. But after supply, HBM LTA and historical drawdown scenarios, the probability-weighted upside narrows to +36-81% for Samsung and +23-64% for SK Hynix. The bottleneck is not the multiple, but whether 2028E is normalized or peak earnings.	Samsung and SK Hynix 2028E valuation
2026-07-09	AI token value and memory value added	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 physical bottleneck, but its token-dollar value share is small. The key test is not low P/E, but whether HBM4 pricing, blended ASP and cloud backlog conversion can support KRW 100-140tn of mid-cycle net income.	AI token value and memory value added
2026-07-08	Samsung 2Q26 signals and noise	Samsung’s provisional 2Q26 operating profit of KRW 89.4tn, DRAM ASP +47%, NAND ASP +66%, legacy-memory price spikes and hyperscaler capex strengthen the investment case. But July 7-8 foreign selling, rates and FX pressure, ASML/TSMC high-bar risk and weak market breadth argue against adding before the next proof points.	Samsung 2Q26 signals and noise
2026-07-08	Samsung selloff and the closest NVIDIA earnings analogue	Samsung’s two-day post-print drawdown after 2Q26 looks less like NVIDIA’s August 2024 one-day sell-on and more like NVIDIA Q4 FY26 in February 2026. NVIDIA’s large post-earnings drawdowns did not recover on strong numbers alone. They needed a catalyst that killed the specific fear, such as a Fed cut, tariff pause or GTC 2026 $1T demand comment. For Samsung, the next proof point is late-July Big Tech CapEx guidance.	Samsung earnings selloff and NVIDIA analogue
2026-07-07	Big Tech financing relay and the memory bottleneck	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 recipient is the memory/HBM bottleneck, with late-July calls needed to confirm the 2027 capex slope and continued absorption of memory-cost pressure.	Big Tech financing relay
2026-07-07	Late-July Big Tech calls and the memory thesis	The Alphabet, Microsoft, Meta and Amazon calls are less about total capex and more about the 2027 capex slope plus memory-price durability. The scenario 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 still needs DS core profit, 3Q pricing and HBM4/HBM4E confirmation.	Late-July Big Tech earnings and the memory thesis
2026-07-04	Dukovany and the Westinghouse settlement	The Westinghouse IP risk that could have blocked Dukovany 5 and 6 has fallen after the January 2025 global settlement and June 2025 contract signing. But this is not free European market access for Korean nuclear exports. The next question is Westinghouse tolls, localization, fixed-price and schedule risk, and actual margins.	Dukovany and the Westinghouse settlement
2026-07-04	SK Square and SK Hynix cash-flow conversion	SK Square is not the investment vehicle for the whole SK Group. It is better framed as a listed AI and semiconductor capital allocator built on SK Hynix equity value. The key is not Hynix net income by itself, but whether Hynix FCF turns into cash dividends and SK Square allocates that cash to buyback cancellation or Hynix-linked investments, compressing the NAV discount.	SK Square investment memo
2026-07-04	SK Gas LNG cold energy and AI data-center cooling option	SK Gas has not disclosed a confirmed commercial cold-energy project. The confirmed point is that an SK Gas-linked engineer published the technical logic for using LNG cold energy in AI data-center cooling on SK Group’s DevOcean channel. The physics of roughly 0.2 MWh of cold energy per ton of LNG and the Ulsan KET / SK-AWS data-center location are relevant, but target PUE, tariff, capex, payback and SK Gas margin attribution remain undisclosed.	SK Gas LNG cold energy
2026-07-04	AI data-center power bottleneck map	Splitting AI data-center power demand into generation and backup power, fuel cells and distributed power, ESS/UPS, power conversion and transformers, cables and bus ducts, and cooling/EPC shows the cleanest direct evidence in GNC Energy, VinaTech and Seojin System. Sanil Electric, Iljin Electric, Gaon Cable and SK Gas remain follow-up diligence candidates, while Hanwha Engine and HD Hyundai Marine Engine still need listed-entity proof.	AI data-center power bottleneck map
2026-07-04	SK Hynix ADR, common-share price transmission and leveraged ETF plumbing	SKHY can broaden U.S. AI-memory investor access, but in the short run the key is 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
2026-07-04	SK Gas vs SK Oceanplant as second-line AI data-center power candidates	After SK Group’s AI data-center investment announcement, the market selected power bottleneck directness rather than the whole group. SK Gas is tied to Ulsan GPS, KET LNG and the Ulsan AI data-center power structure, with 2026F PER 3.9x and PBR 0.52x. SK Oceanplant remains a second-line offshore-wind infrastructure option, but it needs chart repair and earnings confirmation.	SK Gas vs SK Oceanplant
2026-07-04	GNC Energy AI data-center backup power	GNC Energy should be read as an AI data-center backup-power EPC candidate, not a generic AI power stock. LG Uplus AIDC cooperation, the KRW 29.748bn Samsung SDS Gumi AI data-center generator contract, data-center backup-generator demand and five-day institutional buying of KRW 7.66bn explain the rally. But after a 45.5% five-day move, foreign selling and a wide upper wick argue for waiting for a pullback.	GNC Energy rally analysis
2026-07-04	SK Hynix F-1/A and F-6	On the latest SEC filings, SKHY is not just an ADR wrapper. It packages HBM, Custom HBM, eSSD, Yongin Fab 1 and Cheongju P&T7 capex for U.S. AI-memory investors. The F-6 confirms 1 ADS = 0.1 common share, and the board-authorized new-share ceiling is 17.79mn shares, 2.50% of existing shares. But ADS count, offer price, net proceeds and effectiveness are still unavailable, so this remains a watchlist event.	SK Hynix F-1/A and F-6
2026-07-04	FADU 2Q26 earnings preview	Combining 1Q financials, 2Q supply contracts, June amendments and local consensus data gives a midpoint of KRW 70.5B revenue and KRW 11.8B operating profit. A consensus beat looks likely, but the contract total should not be pulled fully into 2Q, so this is a moderate beat rather than a mega surprise.	FADU 2Q26 earnings preview
2026-07-04	VinaTech and Bloom Energy power-buffer chain	VinaTech should not be read as an AI data-center power-generation stock. It sits in the power-quality layer that can buffer transient load, voltage movement and peak current between Bloom SOFC units and AI servers. The KRW 41.2B Bloom contract is 50.12% of FY2025 revenue, but the real thesis needs repeat POs, system margins and customers beyond Bloom.	VinaTech and Bloom Energy
2026-06-30	H1 2026 AI infrastructure bottlenecks and narrow markets	Korea doubled at the KOSPI level while the median stock was still down, and the U.S. rally was less about all mega-cap tech than memory, storage, servers, equipment and power. H2 is about separating recurring-revenue bottlenecks from option-like names exposed to leverage and ETF plumbing.	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, but by June 29 intraday Samsung was below its pre-event price and SK Hynix was nearly back to zero. The key for Samsung 2Q26 is not reported OP of KRW 87-92tn alone, but Core OP of KRW 102-107tn after adding back catch-up bonus accruals and 3Q OP durability near KRW 100tn.	Samsung 2Q26 Preview
2026-06-28	HBF and HBC commercialization calendar	HBF and HBC are real AI-memory bottleneck themes, but they are milestone trades before they are revenue trades. The 2026 signals are HBF sample delivery and Qualcomm AI200 shipment; 2027 is about HBF customer qualification and AI250/HBC Gen 1; 2028 and beyond is revenue. The clean exposure is SanDisk for HBF, Qualcomm for HBC and Samsung Electro-Mechanics as Korea’s confirmed component exposure, while FADU in HBF and Hanmi Semiconductor in HBC are weak classifications.	HBF And HBC Commercialization Calendar
2026-06-22	MSCI developed Watch List probability and the KB Financial trade	At the June 24 MSCI review, immediate developed-market inclusion is ~1%; the real event is Developed Markets Watch List re-entry. Synthesizing two analyses (probability frame + Red Team), Watch List is 30-35% and the base case is positive monitoring (~53%). The cleanest payoff is not chasing semis but KB Financial’s conditional re-rating at ~0.87x 2026F P/B, entered after confirming the announcement wording.	MSCI Watch List Probability and the KB Financial Trade
2026-06-22	Samsung FCF return and DS bonus buyback flow	Samsung is not a high-dividend stock, but with FY2026E net income near KRW 300.2tn, the 50% FCF return policy has meaningful option value. The DS bonus treasury-share purchase is not cancellation, but it can still create KRW 20-22tn of company purchase demand, roughly 1% of common market cap.	Samsung Is Not A Dividend Stock
2026-06-22	HBM, HBF and HBC: how AI memory technologies differ	The three are not the same type. HBM (bandwidth wall, mass production ★★★★★), HBF (capacity wall, simulation stage ★★, first sample H2 2026), HBC (Qualcomm’s accelerator architecture, not a memory chip ★, 2027 sample target). The three are complements, not competitors: hot data (KV cache) goes to HBM, cold data (frozen weights) goes to HBF, low-cost inference goes the HBC direction. No technology threatens HBM’s working-memory slot today. Meaningful revenue contribution from HBF/HBC is earliest 2027-2028.	HBM, HBF and HBC: How to Tell AI Memory Technologies Apart
2026-06-21	Who pays for the 2027 semiconductor consensus	The 2027E consensus for Samsung, SK hynix, Micron and NVIDIA rests on hyperscaler CAPEX, not consumer electronics. Big-four 2027E CAPEX of ~$782.2bn against ~$119.9bn of FCF means it is payable on paper but with a thin buffer, and NVIDIA’s FY2028 ~$551.7bn revenue is ~70.5% of that CAPEX, so demand beyond the big four is required. Government / sovereign AI is supplementary; PC and smartphone OEMs are already in the can’t-pay zone.	Who Pays For The 2027 Semiconductor Consensus
2026-06-21	Q2 KRW-weakness earnings surprise candidates	FRED USD/KRW and May/early-June export data show a supportive FX setup for Q2 exporters. The alpha is not generic exporter exposure, but names where dollar revenue, won costs, improving volume or pricing, and under-reflected consensus line up, led by Kolmar Korea, PSK Holdings and Wonik QnC.	Q2 Earnings: Exporters That Can Still Benefit From KRW Weakness
2026-06-21	TechWing HBM Cube Prober	TechWing is no longer just an HBM theme name. Samsung supply, the first SK hynix order and Micron evaluation could reclassify it from memory handler into HBM inspection platform, but entry still needs support at KRW 58,000-60,000 or a reclaim of KRW 65,300.	TechWing HBM Cube Prober analysis
2026-06-21	KOSPI 60-day disparity overheat framework	KOSPI’s +28.6% 60-day disparity is not a top call. It is a partial risk-budget reduction signal. Since 2025, disparity-overheat signals still produced positive average 5-day and 20-day forward returns, but 60-day disparity above +20% lifted the probability of a 10-session -5% pullback from a 16% baseline to 63%. Because the effective sample is only eight episodes, use it to reduce chase intensity, not as a mechanical sell rule.	KOSPI 60-Day Disparity At +28.6%
2026-06-21	CXMT IPO and memory price risk	CXMT’s IPO is not an immediate HBM price-collapse signal. It is a structural supply event that can cap client DDR5, LPDDR and consumer NAND pricing. The key equity question is whether HBM and AI server memory mix can absorb client DRAM/NAND risk.	CXMT IPO And Memory Price Risk
2026-06-20	How hard is it to beat pure KOSPI?	In 2026, pure KOSPI was not the average Korean market; it was a concentrated Samsung Electronics / SK Hynix AI-memory benchmark. In a 2,730-stock Monte Carlo, a random equal-weight 20-stock portfolio beat KOSPI only 1.0% of the time; a 30-stock portfolio only 0.5%.	How Rare Is It To Beat The Pure KOSPI Benchmark?
2026-06-19	Warsh Fed and expensive money	The June FOMC changed market habits more than it changed rates: less forward guidance, a weaker Fed put, and more selective capital moving toward AI infrastructure, power, defense, energy, physical bottlenecks and proven cash flow.	The Era of Expensive Money
2026-06-13	Target-price gap and comfort-entry candidates	A large target-price gap is not enough. Combining target revisions, smart money and 20/50-day trend leaves KB Financial and Hana Financial as the cleaner comfort-entry candidates.	A Large Target-Price Gap Is Not Enough
2026-06-13	Korea-U.S. policy capital and Team Korea nuclear	Breaks the $350bn policy-capital event into nuclear beta, execution risk and follow-on alpha across O&M, controls, instrumentation and grid equipment.	Korea’s $350 Billion U.S. Investment Law
2026-06-13	Semiconductor Top 50 relative alpha	Samsung and SK Hynix are core exposure, but two-month alpha candidates are second-line equipment, substrate and AI storage names such as Korea Circuit, HPSP, TES, VM and FADU.	Korean Semiconductor Top 50: Which Names Can Beat Samsung and SK Hynix?
2026-06-13	Semiconductor ETF exposure	Separate absolute ETF ownership, market-cap-adjusted ETF sensitivity and under-owned ETF gap candidates.	Samsung and SK Hynix Are 90.8% of Korean Semis
2026-06-13	ETF-led market volatility	Use ETF rebalance flow, big-figure pullbacks and KOSPI200 expiry squeeze signals as separate playbooks.	ETF Flow Is Leading the Korean Market
2026-06-12	AI supercycle duration	Agent demand, IPO funding and memory/storage bottlenecks keep the AI cycle longer than a simple rate-sensitive capex trade.	Why the AI Supercycle Is Getting Longer
2026-06-12	Theme ETF rebalance flow	Track cap redistribution pressure in Korean theme ETFs and second-line semiconductor equipment candidates.	Korea Theme ETF Rebalance Flow
2026-06-11	Quality laggards after Samsung/SK crowding	Find quality names with rising EPS and strong 1Q results but lagging prices.	Quality Names to Revisit in Q3
2026-06-05	Sam-Ha-Ma parity follow-up	An AI chip and memory P/E map showing Micron’s premium and the Korean memory discount.	Sam-Ha-Ma Parity Follow-Up
2026-06-03	Sam-Ha-Ma forward P/E parity	Samsung Electronics and SK Hynix derated versus Micron without EPS damage.	Samsung-Hynix-Micron Forward P/E Parity
2026-06-03	Real-money flow framework	Split institutional flow into short-term broker flow and real-money buying from funds, private funds and pensions.	Real Money Flow Framework
2026-06-03	Liquidity and ADR	The market has liquidity, but breadth is weak and leadership is narrow.	Korea Market Liquidity and Narrow Leadership
2026-05-16	Samsung versus TSMC-style PER reclassification	Tests whether Samsung can move from a memory-cycle multiple to an AI platform multiple.	Can Samsung Electronics Get 15x PER?
2026-05-05	AI PCB system bottleneck thesis	Reframes PCB, substrates and CCL not as a linear “next theme after GPUs,” but as the common bill-of-materials bottleneck across rack-scale AI systems where GPUs, CPUs, DPUs, NICs, switch ASICs and memory modules expand together.	AI PCB and Substrate Thesis
2026-05-05	Korea AI PCB ecosystem, 10 companies	Maps Samsung Electro-Mechanics, Isu Petasys, Daeduck Electronics, Simmtech, Korea Circuit, TLB, Doosan Electronic BG, Kolon Industries, Pamicell and Taesung across FC-BGA, MLB, SoCAMM, CCL, low-dielectric materials and equipment.	Korea AI PCB Ecosystem: 10 Companies

Analysis Angles

ETF And Passive Flow

AI PCB And System Bottlenecks

AI PCB and Substrate Thesis: GPU, CPU, NIC and CCL Demand Are One System Bottleneck
Korea AI PCB Ecosystem: 10 Companies Across FC-BGA, CCL, SoCAMM, Materials and Equipment

Policy Capital And Event Reading

Korea’s $350 Billion U.S. Investment Law: What Is the Nuclear Opportunity for Team Korea?

Relative Valuation And Cycle Frames

Flow Quality

Stock Selection Screens

How To Use This Hub

Start with the dataset and method. Then ask whether the price move is supported by flow, whether the flow is real-money or program-like, whether the theme is turning into revenue, and what would invalidate the idea.