2026-09-04 — SK hynix again tells the SEC no decision has been made on a Solidigm pre-IPO raise

• SK hynix filed a Form 6-K on September 4 reiterating that “no matters have been determined” on a rumored pre-IPO capital raise for its US NAND subsidiary Solidigm, meeting a self-imposed disclosure deadline rather than announcing anything new. The filing traces to an August 5 Korea Economic Daily report that Solidigm was pursuing roughly 5 trillion won (later Korean press put the range at 5-10 trillion won, or roughly $3.6-7.2bn) in pre-IPO funding; SK hynix’s August 6 6-K first denied any final decision and pledged a further update within one month, which put the deadline at today. Today’s filing repeats that nothing is finalized and pushes the disclosure commitment out to December 4, 2026. This is a capital-raise signal, not a shortage-side development — Solidigm is SK hynix’s NAND arm rather than its HBM/DRAM business — but it bears on the thread’s capex-expansion watch line as another memory maker testing capital markets during the squeeze. (SEC EDGAR 6-K, primary, Korea Herald, background on the Nasdaq speculation)

2026-09-01 — Counterpoint Research: CXMT crosses 10% of global DRAM revenue share, first double-digit quarter for a Chinese DRAM maker

• ⚠️ OUT OF WINDOW (Counterpoint’s own report is dated 2026-09-01, two days before this sweep’s window opens) — flagged because it is not recorded on this thread or on CXMT IPO, it speaks directly to this thread’s own watch line (“Samsung/SK Hynix/Micron expansion responses, CXMT’s state-funded push… and who blinks first on capex”), and press coverage of it landed squarely inside my W1 window (TechNode published 09-04, Seoul Economic Daily’s piece is timestamped 2026-09-04T07:00 KST = 2026-09-03T18:00 ET). CXMT’s global DRAM market share by revenue hit 10% in Q2 2026, up from 4% a year earlier and 8% in Q1 2026 — Counterpoint’s own page gives Samsung 38%, SK Hynix 25%, Micron 24%, CXMT 10%, with the top three’s combined share falling from 94% (Q2 2025) to 87% (Q2 2026). Coverage frames the milestone as roughly two years ahead of Counterpoint’s and UBS’s own earlier projection that CXMT would not reach 10% shipment share until 2028. The mechanism named across outlets: Samsung and SK Hynix are pivoting DRAM capacity toward HBM (SK Hynix held 50% HBM share, Samsung’s HBM share rose from 21% to 33% the same quarter) opening a conventional-DRAM vacancy CXMT is filling. (Counterpoint Research, TechNode, Seoul Economic Daily)

2026-08-31 — CXMT’s Q2 gross margin beats Micron and SK Hynix; it also starts mass-producing LPDDR6 first, worldwide

• CXMT posted an 87.59% gross margin for Q2 2026, up 8.44 percentage points from Q1’s roughly 79%, and higher than Micron’s 84.9% non-GAAP gross margin (fiscal Q3) and SK Hynix’s 83.2% (Q2 2026). TrendForce, reporting this 08-31, frames it as the clearest evidence yet that China’s state-backed DRAM entrant is now pricing and earning at the same level as the two incumbents this thread has tracked from the buy side all year — a direct extension of Saturday’s already-recorded H1 revenue/profit beat, not a restatement of it. (TrendForce)

• The same reporting says CXMT has begun mass production of LPDDR6, and multiple outlets (TechNode, Digitimes, Reuters-sourced coverage) confirm this is the first commercial LPDDR6 deployment anywhere, ahead of Samsung and SK Hynix — the chips will ship in Xiaomi’s 18 Fold foldable, launching September 2026, confirmed by Xiaomi founder Lei Jun on Weibo. This is a technology-leadership claim, not just a capacity or pricing one: CXMT was mass-producing LPDDR5 less than a year ago. (TechNode, Digitimes)

• Korean trade press reports spot-market HBM3E (36GB) trading around $2,100 against long-term-agreement pricing equivalent to roughly $350-490 — a 4-5x premium — with HBM4 (16-layer) spot pricing around $3,500 against LTA rates. The piece attributes the gap to HBM4’s lower manufacturing yield consuming more DRAM wafer capacity per unit, and to Samsung’s already-recorded 70%-of-capacity long-term-agreement lock (a July commitment, not new) starving the spot market; it adds that even large customers like Nvidia, Microsoft and Google reportedly cannot get their full contracted volumes. ⚠️ Figures are reported in Korean won and converted; treat as directional rather than exact. (Seoul Economic Daily)

2026-08-30 — CXMT’s first results as a listed company beat its own prospectus by a third, and give the shortage a supply-side number from the Chinese side

• 🕰 CXMT reported first-half 2026 revenue of ¥150.31bn (~$22.4bn), up 873.64% year-on-year, and net profit attributable to shareholders of ¥77.61bn against a ¥2.33bn loss in the same period last year. The number that matters is not the growth rate but the miss against its own guidance: CXMT’s pre-IPO prospectus forecast ¥110–120bn of first-half revenue and ¥50–57bn of net profit, so the company beat the top of its own revenue range by 25% and the top of its profit range by 36%, within weeks of listing. DDR5 alone contributed ¥69.47bn, 46.3% of main-business revenue. This is the first hard Chinese-supply datapoint this thread has carried. Every prior entry here has measured the squeeze from the buy side — Nvidia’s $279bn of forward purchase commitments, Micron’s margin math, Apple’s price negotiations — and priced Chinese memory as a policy question about whether Apple may buy it. CXMT’s own book says the Chinese supply is not marginal capacity waiting on a policy decision; it is already earning at scale off the same DRAM tightness. It also reframes the apple-cxmt-senate-deadline expectation, standing passed-silent since 08-21: a supplier posting these numbers has correspondingly less need of Apple. ⚠️ Dated 08-28, inside a finalized digest-day and caught today; Bloomberg’s own page returned 403 to a direct read, so this rests on SCMP and Benzinga reporting the same filed figures, which agree to the decimal. (SCMP, Benzinga)

2026-08-29 — Nvidia’s own memory bill triples in a quarter, and the price hikes get numbers attached

• Nvidia disclosed its forward purchase commitments jumped from $119B to $279B quarter-over-quarter (+135%); CFO Colette Kress said the increase is “primarily related to the procurement of memory” to support the ramp of its next-generation platforms, naming SK Hynix, Samsung and Micron as the suppliers. This is the same 08-26 earnings call already on this thread’s timeline (margin guidance cut to 74.0%) — the $279B commitment figure is new detail this thread hadn’t yet captured. (24/7 Wall St., 2026-08-27)
• The >15% AI-server price rise this thread logged 08-22 now has supplier-side numbers behind it: SK Hynix raised DRAM pricing roughly 30% sequentially and Samsung roughly 40%+, with NAND climbing faster still (SK Hynix mid-50%, Samsung high-60%) — and TrendForce reports HBM contract prices could rise more than 50% further in 2027. Analysts read Nvidia’s commitment as a durable tailwind for Micron/SK Hynix/Samsung’s HBM franchises, even as the market’s 08-24 demand-risk reading (already on this thread) argued the opposite read of the same price rise. (Investing.com, TrendForce)
• Market reaction split by exposure in the 08-27 session: Micron -3% to $913.62, Western Digital -4% to $449.18, SanDisk -1% to $1,479.23, while SK Hynix rose +1% to $160.30 — SK Hynix’s outperformance attributed to being “most direct” on Nvidia’s HBM pipeline and a newly-listed US ADS structure drawing fresh buyers. All four names remain up sharply year-to-date (Micron +229%, SanDisk +532%).

2026-08-27 — $31bn of Japanese NAND, on a six-year horizon against a shortage felt now

• Kioxia and Sandisk said they will invest more than $31 billion in Japanese NAND flash capacity through 2032 — Yokkaichi and Kitakami plus a new Iwate fab — contingent on Japanese government support and framed explicitly as a response to AI-driven memory demand. This is the supply-side answer to the constraint Nvidia priced into its margin guidance the night before, and the mismatch is the point: a six-year build against a shortage that is moving prices now. It describes the shape of the problem rather than resolving it. (Investing.com, wire, Seeking Alpha)

2026-08-26 — The squeeze shows up in Nvidia’s own margin line

• Nvidia guided Q3 FY2027 gross margin to 74.0% ±50bps, down from 75.0%, with rising HBM and DRAM input costs the stated pressure. Trade reporting adds that Nvidia has told AI-server makers to expect price increases above 15% on systems shipping in early 2027 — ⚠️ that figure traces only to trade coverage, not to any Nvidia statement, and is carried as unverified. This is the first time this thread has seen the memory squeeze priced rather than forecast, and it is priced in the margin of the company best placed to pass it on. (NVIDIA IR, FXStreet)

2026-08-24 — The market re-reads the >15% price rise as demand risk rather than pricing power

• ⚠️ The new fact today is the READING, not the price rise — the price rise itself is already on this thread, dated 08-22. Nvidia’s notice to contract manufacturers that AI-server prices rise more than 15% on systems shipping in early 2027, blamed on DRAM and HBM costs, was logged on the 08-23 page. It resurfaced today as the stated driver of a semiconductor selloff, and that is a different datum: the same disclosure that read as cost pass-through three days ago is now being read as demand-softening risk. Micron, SanDisk and Western Digital each fell more than 5% and the SOXX ETF dropped 3–4% in a session where the Dow rose. (24/7 Wall St., 10:43 ET)

2026-08-23 — Hot Chips answers Micron’s memory-wall math with three separate engineering fixes

• Micron’s Raghu Sreeramaneni told Hot Chips 2026 that compute is scaling roughly 3x every two years while HBM bandwidth scales only about 2x — so the memory wall is widening, not closing. Presented as “Evolving Memory Architectures for AI” on day one of the conference at Stanford (08-23 to 08-25). This is the technical floor under the 08-22 capex story below: if bandwidth structurally trails compute, the memory premium Nvidia blamed its >15% server price rise on is not a cycle to wait out. (ServeTheHome — conference coverage, Hot Chips 2026 programme)
  • Day two produced three separate engineering answers to that same wall. Samsung unveiled zHBM, stacking memory directly atop the compute die, claiming 230% more bandwidth and 70% better power efficiency than a standard HBM4e stack — roughly 100W saved per GPU in a four-stack configuration. (TrendForce)
• SK hynix showed i-HBM packaging claiming >30% lower thermal resistance, and disclosed it is evaluating Intel’s EMIB packaging alongside CoWoS for future HBM-logic integration — a possible SK hynix–Intel packaging tie-up in a market TSMC’s CoWoS currently defines. (TrendForce)
• d-Matrix disclosed Raptor, which fuses DRAM directly beneath compute and drops the PHY layer entirely, claiming SRAM-class bandwidth at roughly a tenth of HBM’s power draw, and hit 105 TB/s in real silicon. Positioned as a complement to HBM rather than a replacement — it is an inference accelerator, not a general part. (ServeTheHome, Wccftech corroborating)
• The analytical point: day one said the wall is widening structurally; day two produced three independent attempts to engineer around it — stack on top of the die (Samsung), re-package the stack (SK hynix), or delete the interface (d-Matrix). That is an industry treating the constraint as permanent rather than cyclical.

2026-08-22 — The memory squeeze arrives in the data center as a capex story

• Nvidia is raising AI server prices more than 15% and naming DRAM and HBM costs as the reason. This thread has followed the three-year HBM deficit through consumer hardware price rises — Apple, Microsoft, handset and PC bills of materials. The 08-22 notice is the same squeeze reaching the buildout itself: Samsung, SK Hynix and Micron now set the marginal cost of AI compute, and Nvidia has said so publicly by passing it through rather than absorbing it. All three price-setters have been watchlist entities on this map since 2026-07-24 and are already tagged in this file’s own frontmatter, so the companies at the centre of this thread carry entity slugs on this bullet. (Bloomberg, Fortune)

2026-08-18 — The shortage produces its first visible casualty on the demand side

• Xiaomi’s profit fell on a worsening memory-chip crunch hitting smartphone demand. This thread has run entirely from the supply side; this is the same transaction seen from the other end. Datacenter HBM and DRAM demand is bidding memory away from consumer devices — Micron crossed $1,000 a share on Monday because it is the seller, and Xiaomi’s margins compressed because it is the buyer that cannot outbid an AI datacenter. (Bloomberg)
• This is the first item where the buildout’s input scarcity measurably taxes a non-AI business’s P&L rather than lifting a supplier’s multiple. If it spreads to other handset and PC makers, this thread needs a demand-side strand: the story stops being who profits from scarcity and becomes who pays for it — which reaches consumer prices, and therefore inflation prints.

2026-08-05 (late catch, added 2026-08-11) — Nvidia tests cutting Rubin Ultra’s own memory to cope with the shortage it’s helping cause

• Nvidia is reportedly testing lower-memory configurations of its next-generation Rubin Ultra accelerator — designs as small as 192GB (stepping back to 8-Hi HBM4) versus the 288GB on the current Vera Rubin, a 33% cut — per an August 4 TrendForce report. Driven by tight HBM wafer allocation across SK Hynix, Samsung and Micron, the same three suppliers this thread already tracks on the supply side. A lower-memory Rubin Ultra means buyers need more GPUs for the same workload — more power, rack space and interconnect cost — the first entry on this thread showing the squeeze reaching into Nvidia’s own next-gen chip spec, not just consumer-hardware pricing or supplier capacity plans. Missed by every digest since 08-05; folded in now on the 08-11 finalize’s coverage-critic pass. (Tom’s Hardware)

2026-08-09 — Apple tests Chinese CXMT memory under a Senate deadline not to

• Apple is testing memory chips from China’s CXMT for China-market iPhones/MacBooks, days after CXMT rejected Apple’s price-cut demands. A bipartisan Senate group (Shaheen, Banks, Schumer and colleagues) has given Apple until 2026-08-21 to publicly commit to rejecting CXMT and YMTC, both Pentagon-designated Chinese-military-linked suppliers; Micron is lobbying against Apple’s plan. CXMT can hold firm on price because Huawei and Xiaomi have already locked up its output at similarly high levels — AI-driven DRAM scarcity flipping the usual buyer-leverage dynamic. HP and Acer already ship CXMT chips outside the US. (eWeek)

2026-08-07 — SK Hynix weighs selling down its China packaging plant

• SK Hynix is exploring a stake sale in its $3B Chongqing chip-packaging plant as export-control tightening around China-facing memory operations continues; potential buyers include Chinese investment funds and semiconductor firms, with SK Hynix possibly retaining a minority stake. The plant handles NAND back-end packaging — separate from the HBM lines that supply Nvidia — and the company says discussions are early and nothing is decided. Cross-refs China’s Stack. (Bloomberg)

2026-08-05 — The 2027 DRAM market closes with no room for a fourth supplier, and Apple’s China hedge fails

• Apple gave up trying to qualify CXMT as a fourth DRAM supplier — months of talks collapsed because CXMT (a Pentagon-listed Chinese military-linked entity) would not price below Samsung/SK Hynix, having already committed its capacity to Huawei and Xiaomi at elevated rates, while Senate pressure (Banks, Schumer) added regulatory-clearance risk; Samsung, SK Hynix and Micron have now fully booked their 2027 output, with existing customers getting only 60-70% of requested volume against advance deposits. (Tech Times)
• CXMT DRAM is already inside HP, Asus and Acer laptops sold outside the US — sold at prices matching Samsung’s (adopted for supply, not discount), and flagged as a fresh risk: Chinese law requires CXMT to report chip vulnerabilities to Beijing’s industry ministry within 48 hours, before notifying the PC makers using the chips. (Tech Times)
• Samsung and SK Hynix are separately testing Chinese chipmaking equipment (AMEC) for their own Chinese-based fabs, hedging against a further tightening of US export controls — testing reported to have started roughly two years ago and now surfacing as the squeeze intensifies. (Reuters)
• 💡 The “who blinks first on capex” question this thread has been tracking now has a concrete answer for Apple: it blinked and lost — it tried the CXMT hedge specifically to route around the Samsung/SK Hynix/Micron oligopoly and couldn’t get a price advantage even from the sanctioned option. The 2027 sellout also sharpens the DoD-ban overhang already on this thread’s watch: CXMT is filling gaps for HP/Asus/Acer and Huawei/Xiaomi while still being too expensive for Apple, not too restricted.

2026-08-01 — Memory prices become large enough to move a hyperscaler’s capex line

• Amazon raised FY2026 cash capex guidance to ~$220B from ~$200B on 07-30, explicitly citing higher memory costs — the first instance in this map of memory-price inflation moving a hyperscaler’s own capex number rather than only its cost line. (CNBC)
• Management said even $220B would not buy enough capacity to meet 2026 demand, and expected the same into 2027-28. (CNBC)

2026-07-31 — Cook: “a hundred year flood on memory pricing”; Samsung converts a NAND line to DRAM

• Tim Cook, on his final earnings call as Apple CEO, put the company’s own voice behind the shortage — called it “a hundred year flood on memory pricing,” confirmed Apple already raised Mac/iPad prices over it, warned the September quarter sees a bigger supply-driven hit than this one, and said “if there were more suppliers, that would be good… it’s unclear on the pricing side” when asked about diversifying beyond Micron/SK Hynix/Samsung — while Apple continues lobbying Washington to clear CXMT/YMTC purchases despite the bipartisan Senate Aug-21 deadline. (MacRumors, Fortune)
• Samsung completed a NAND-to-DRAM line conversion at its Hwaseong fab, shutting an older 2D NAND wafer line and repurposing it for DRAM — secondary reporting (paywalled primary) puts the capacity gain at roughly +15% general-purpose DRAM by end of 2026; a concrete supply-side response landing the same day as Cook’s warning. (Digitimes, paywalled) — confidence: medium, single secondary source on the figure.

2026-07-30 — Samsung’s Q2 answers the pending flag: chip profit ~250x, phones eat the squeeze, shortage guided to 2028

• Samsung’s DS (chip) division posted ₩89.2-89.5T ($62B) operating profit, a ~250-fold YoY jump — while the Mobile/MX division posted its first-ever operating loss (~₩0.7T on ₩33.2T revenue), citing “elevated component cost pressures across the industry.” Samsung’s own phone business is now eating the DRAM/NAND squeeze its chip arm profits from — the consumer-price-echo thesis, inside one company’s P&L. ([Korea Herald] / AP)
• Samsung locked ~70% of memory output into long-term agreements with five major customers (reported names include AWS, Microsoft, Google) and guided the shortage to extend through 2028 — widened past the “three-year deficit” figure traced to SK Hynix’s Q1 call. ⚠ Provenance split: “through 2028” is the figure carried near-uniformly across Reuters/Bloomberg/WSJ/Yonhap, almost certainly earnings-call commentary rather than release text — strong-secondary, not primary-document-confirmed.
• Bipartisan senators (Banks, Schumer, others) sent Apple a letter demanding it not buy CXMT/YMTC memory chips, with an August 21 deadline — warning that once a China-market part clears Apple’s qualification, “extending it worldwide is a single procurement decision away.” Apple has reportedly already requested to buy from CXMT and purchased (unused) YMTC chips. Converges this thread’s China-capacity strand with the Apple consumer-price echo into one live political fight. ([AppleInsider], corroborated Bloomberg/The Information/ 9to5Mac/Business Standard)

2026-07-28 — The best print in the cycle’s history couldn’t hold a bid

• Anthropic asked SK Hynix for chip supply to build its own custom AI silicon, and separately struck supply agreements with both Samsung and SK Hynix — SK Group chair Chey Tae-won disclosed the ask at a San Francisco AI summit (Bloomberg, 07-25); Dario Amodei confirmed the Samsung + SK Hynix supply deals days later without giving terms (volume, components, or duration). Widens the squeeze’s demand side past hyperscalers to frontier AI labs themselves — SK Hynix also participated in Anthropic’s May Series H round. (Bloomberg · Digitimes)
• Samsung’s chip talent is defecting to SK Hynix over a bonus gap reported as wide as 100x — ~200 Samsung engineers have moved to SK Hynix since December (SK Hynix pays out 10% of operating profit as a profit-sharing pool, fattened by record HBM margins); Samsung won an 18-month injunction this month against two ex-employees joining SK Hynix, citing chips as protected “national core technology.” A labor-market echo of the same squeeze — the profit differential is now visibly reallocating headcount, not just capital. (MIT Technology Review)
• Then the quarter itself landed. SK Hynix Q2, primary-sourced: revenue 79.3187T won, operating profit 60.5426T won (+557% YoY, 76% margin), net profit 93.9226T won. A record in absolute terms — and a miss against the 63.7-64.1T consensus this thread’s ledger entry tracked as “~$43.7B.” Revenue missed too (~83.6-83.9T expected). The net-profit beat is non-operating gains, not operations. (SK hynix)
• HBM4 mass shipments began in Q2, “full ramp” guided for 2H26; ~10 customers signed to long-term supply agreements. Management called Blackwell/HBM3E demand still strong and said Rubin — which needs HBM4 — is running later than expected. The shortage thesis survives the miss; the timing shifts right.
• ⚠ A capex claim this thread is NOT adopting: secondary outlets reported capex raised to the ~40T-won high end (~45T/$31B). The release itself says “capital expenditure discipline,” phased “based on customer demand and investment efficiency.” Primary and secondary disagree in direction, not just degree — flagged pending the 07-30 Samsung breakdown.
• The market’s answer (07-29 KST session): SK Hynix opened up, ran +3.74%, then closed −9.61% at 1,401,000 won. Samsung +4.55% → −5.23%. Micron had already fallen 8.85% to $820.53 in the US session.

2026-07-28 — Capacity-race backstory: the trio, plus a same-day shock

• The trio’s HBM race, mapped. Backward crawl on Samsung/SK Hynix/Micron finds SK Hynix still the incumbent (~55% HBM share, first to clear Nvidia’s HBM4 Vera Rubin qualification, commanding a reported 60-70% of that volume) but Samsung projected by UBS to overtake it on HBM bit-capacity share in 2027 (41% vs 39%) — Samsung’s first-ever lead — even though it’s still stuck at Nvidia’s “paid evaluation” stage for HBM4 as of 07-17, no confirmed volume order. Micron is #3 and the most exposed to Nvidia’s on-again-off-again favor (excluded from early Vera Rubin slots in Feb, reinstated by June) but posted the trio’s standout quarter (Q3 FY26: $41.5B revenue vs $9.3B a year ago, 84.6% gross margin) while running a $200B US buildout as the only US-based DRAM maker.
• The “three-year HBM deficit” claim, sourced. It traces to SK Hynix’s own Q1 2026 earnings call (04-23) — CFO Kim Woo-hyun and VP Park Joon-deok guiding that demand will outpace supply for at least three years. Company guidance, not an outside analyst estimate — the CXMT-thread framing that repeated it (via a 07-27 Tech Times piece) draws from this same-origin figure, not a fresh source. DRAM’s 2026 YTD price trajectory stays genuinely unsettled across sources — the only claims that clear medium confidence are outcome-side (Samsung’s confirmed Q2 sequential DRAM +44%/NAND +53%; SK Hynix’s ~72-77% operating margins), not a clean industry percentage.
• “Black Tuesday” — the squeeze gets a same-day shock. CXMT’s Shanghai STAR-market debut (07-27, +466% to ~$489B) triggered a 2026-07-28 Seoul selloff: KOSPI -10.84% with a circuit breaker, SK Hynix -14.65% — Korean financial press attributed it directly to investors “refocus[ing] on… China’s growing competitiveness in the memory… industries,” compounding a Wall Street chip selloff and Middle East tension. The sharpest, most immediate market pricing of CXMT-as-threat found in either this crawl or last week’s CXMT-focused one — worth checking whether it holds past SK Hynix’s 07-29 and Samsung’s 07-30 earnings prints, or gets partly reversed. Finding: artifacts/findings/memory-trio-2026-07-28.md

2026-07-23 — The supply side raises guidance too

• TSMC raised 2026 capex guidance to $60–64B plus an additional $100B Arizona investment; CEO C.C. Wei says demand robust “through 2029–2030” — the supply-side echo of the same squeeze. Chip stocks (incl. Nvidia, -2.2%) still slid Friday on profit-taking despite the beat.

2026-07-20 — Capital-side marker set

• CXMT set July 27 for an up-to-$9.8B Shanghai IPO — Asia’s largest 2026 listing, funding Chinese DRAM/HBM capacity into the squeeze (now tracked as CXMT IPO). (MarketScreener)

← Backstory

2026-06-25 — Consumer price echo

• Apple price hike attributed to the DRAM/NAND squeeze — The Neuron / Rundown AI led with it; we tracked the buildout but missed the demand-side cost echo (the gap that opened this thread).