Markets desk

Tue 1 Sep 2026 07:30 HKT / 08:30 KST

Last cash is Mon 31 Aug. Paper only. Shorts UNKNOWN (data.krx login-walled). Evening X tape (bullish memory tightness) left in place.

watchlists AJ US omit — no new @alojoh after RIVN; Outlook still dark. MRVL + NVDA still the last follow-ups Y. Korea attractive list empty besides Hynix desk.

news No new Asia-news pass this flash. Leftover Yonhap July IP / KED zHBM / Friday KOSPI dump cards stay below.

X Evening 21:30 HKT tape already live — not wiped. Sentiment still bullish memory ASP/tightness, medium (Samsung 70% LTA / Pyeongtaek S5 / SKHY Yongin-Honam + possible Kioxia JV / TrendForce Q3 PC DRAM +18–23% / CXMT HBM3E / Rubin CPX HBM4). Home unread.

DART No new 000660. After-close Mon keeps still on the board (Agent AI offering withdrawn; E-Gis buyback then cancel; IPARK Yangjae 2; HS Hwaseong Bichwitown). English + why on cards.

funding 07:00 HKT Hynix marks only (not a full majors pass). SKHYUSDT ~164 vs xyz:SKHY 163.8; SKHYNIXUSDT ~1225 vs xyz:SKHX 1224.6. Two books stay split. Fade still vetoed. Paper.

Hynix Mon 31 cash 000660 1,674,000 (+1.27% / +21k) V 3,980,661. For −211,935 / Inst −314,294 (Naver labeled KRX). shorts UNKNOWN. SKHY RTH 164.58 / AH 163.75. FX Yonhap 15:30 1,368.60. prem vs RTH +34.55% (1:10). KIND/DART: still 28 Aug FTC large-business-group status only. Seoul paper rule did NOT fire (+1.27% / +34.55% vs ≤−8% / ≥+40%).

Withdraws Mar rights issue 3.99m sh / ₩18.74bn (82% of shares); will re-open funding

What: Board withdrew the 25 Mar 2026 shareholder-allotment plus leftover public offering. Planned 3,990,609 ordinary shares / ₩18,735,909,255 for debt repayment and working capital. FSS registration ping-pong: filed 26 Mar, amendments 21 Apr and 21 May, extra correction demand 8 Jun. They say the timetable broke and they will look at other funding that fits the balance sheet. Companion amendment rcp 20260831001162 zeros the material-fact report fields. Close ₩3,625 (−2.16% vs ₩3,705), high ₩3,890, low ₩3,585, volume 4,072, turnover ₩15.0m, 4,883,930 shares, mcap ₩17.70bn (Naver close). Withdrawn paper was 81.7% of listed shares / 106% of close mcap. Printed 15:55 after the 15:30 cash close.

Why mixed 3-8%, p≈40%: pulling an 82% rights issue takes a dilution overhang off a ₩17.7bn KOSDAQ microcap, which is the bull case. Mixed not clean-up because they still need the debt-repay / working-capital cash and say they will re-open another raise, and FSS has been blocking this deal since March so the street had five months to treat it as dead. 3-8% not 8%+ because Monday already closed −2.16% on 4k shares and leftover is Tuesday’s open. 40% because the ₩18.74bn and 3.99m share count are unambiguous, but a ₩18bn name on 4k shares can gap either way on the “other funding” line.

DART

Buyback trust ₩3.0bn (~6.9% / 707,547 sh) via Hana 31 Aug–2 Mar; cancel after trust

What: Board 31 Aug: Hana Securities buyback trust ₩3,000,000,000. Window 31 Aug 2026–2 Mar 2027. Purpose: share-price stability and shareholder value. Planned ordinary 707,547 sh sized off 28 Aug close ₩4,240 (implied ₩3,000,000,000-ish). Holding period after purchase 6 months or more; they intend to cancel the lot after the trust ends (separate board later). No existing treasury. Outside directors 3 present / 3 absent. Dividendable-profit limit ₩4,023,936,655 before the ₩3.0bn contract. Close ₩4,220 (−0.47% vs ₩4,240), high ₩4,335, low ₩4,035, volume 21,822, turnover ₩89.9m, 10,277,800 shares, mcap ₩43.37bn (Naver close). ₩3.0bn is 6.91% of close mcap / 6.88% of shares. Printed 15:54 after the 15:30 cash close.

Why up 3-8%, p≈45%: ₩3.0bn / 6.9% of a ₩43bn KOSDAQ name with explicit cancel intent is a real six-month bid plus share-count shrink, not a rumor. Bullish because Hana can dribble-buy into a ₩90m tape and they start from zero treasury. 3-8% not 8%+ because it dribbles six months, cancel is a later board, 3 of 6 outside directors missed, and Monday cash already closed. 45% because the ₩3.0bn and 707,547 are unambiguous, leftover is Tuesday’s open, and ₩3bn uses most of the ₩4.02bn dividendable-profit limit so the program is sized to the cap.

DART

Seoul Yangjae 2 reconstruction contractor ₩900.2bn (21.7% of 2025 sales); unsigned until the construction contract

What: Selected as contractor for the Seoul Yangjae District 2 reconstruction project. Expected contract ₩900.2bn (21.7% of 2025 consolidated sales ₩4,147.0bn). Association selected them 29 Aug (Sat); fact confirmed 31 Aug after the AGM result. Scope: 2,344 units / 13 buildings / B3–40F plus amenities. Amount is pre-contract (locks at the construction contract), VAT excluded, can slip. Close ₩23,400 (+1.08% vs ₩23,150), open ₩22,800, high ₩23,400, low ₩22,650, volume 127,915, turnover ₩2.959bn, 65,907,330 shares, mcap ₩1.542tn (Naver close). ₩900.2bn is 58.4% of close mcap. Printed 15:41 after the 15:30 cash close. Parent HDC 012630 filed the same as a subsidiary notice (rcp 20260831800444). NXT after-hours ₩23,200.

Why up 1-3%, p≈40%: ₩900bn / 21.7% of last-year sales clears the 10% KRX bar on a ₩1.54tn KOSPI builder. Bullish because it is a named Seoul reconstruction contractor win, not a rumor. 1-3% not 3-8% because the KRW is still expected until the construction contract, cash is multi-year progress on 2,344 units, this is a ₩1.54tn name, and Monday cash already closed +1.08% so leftover is Tuesday’s open. 40% because the ₩900.2bn and 21.7% are on the page, but an unsigned construction contract and a Saturday association vote can still slip, and some of the street may have seen the Saturday AGM.

DART

Bichwitown alley-housing contractor ₩129.6bn (19.7% of 2025 sales); unsigned until the construction contract

What: Selected as contractor for the Bichwitown alley-housing improvement. Expected ₩129.6bn (19.7% of 2025 consolidated sales ₩659.5bn). Association selected them 29 Aug (Sat); fact confirmed 31 Aug after the AGM result. Scope: 95 units / 18F+7B plus amenities. Amount locks at the construction contract, VAT excluded, can slip. Close ₩10,900 (+0.37% vs ₩10,860), high ₩10,900, low ₩10,630, volume 4,321, turnover ₩46.0m, 10,150,000 shares, mcap ₩110.6bn (Naver close). ₩129.6bn is 117% of close mcap. Printed 15:39 after the 15:30 cash close. Same name already kept the signed Myeonmok Bondong amendment ₩162.6bn at 14:33.

Why up 3-8%, p≈35%: ₩130bn / 19.7% of last-year sales on a ₩111bn KOSPI builder is a second same-day orderbook print, not a rumor. Bullish because it is a named alley-housing contractor win on top of this morning’s ₩163bn signed amendment. 3-8% not 8%+ because the KRW is still expected until the construction contract, 95 units is small vs this morning’s 512-unit amendment, there is no down payment, and Monday cash already closed +0.37% on 4k shares after the 14:33 keep. 35% because the ₩129.6bn and 19.7% are on the page, leftover is Tuesday, and an unsigned Saturday AGM can still slip.

DART

Ship-engine supply to affiliate HD HHI ₩87.09bn (21.6% of 2025 sales); advance yes

What: Confirmed ₩87,091,155,000 ship-engine supply. 21.6% of 2025 consolidated sales ₩402,396,221,904. Not a large corporation. Counterparty: HD Hyundai Heavy Industries (affiliate). Domestic. Period 31 Aug 2026–24 Jul 2028 (can slip with the yard’s build schedule). Advance yes: down payment plus remainder on the engine-delivery schedule. Award 31 Aug. KRW uses the 31 Aug first-notice FX ₩1,376.5/USD (implied ≈ USD 63.3m). Live ₩52,100 (−1.88% vs ₩53,100), open ₩51,700, high ₩52,200, low ₩50,100, volume 124,535, turnover ₩6.406bn, mcap ₩1.767tn, 33,921,495 shares (Naver ~15:36 KST). ₩87.09bn is 4.93% of live mcap.

Why up 1-3%, p≈35%: ₩87bn / 21.6% of last-year sales clears the 10% KRX bar, and there is an advance, so it can print. Bullish because it is a signed two-year engine book, not a rumor. 1-3% not 3-8% because the buyer is the captive parent yard (run-rate, not a new customer), ₩87bn is only 4.9% of a ₩1.77tn name, cash dribbles to Jul 2028, and the 15:29 print landed into the 15:30 cash close with the tape already −1.88% on 125k shares. 35% because the ₩87.09bn and 21.6% are unambiguous, but leftover is Tuesday’s open rather than today, and a captive-yard engine print is easier to fade than a third-party yard order.

DART

Buyback trust ₩2.0bn (~1.7% / 524,934 sh) via KB 31 Aug–2 Mar; hold, not cancel

What: Board 31 Aug: KB Securities buyback trust ₩2,000,000,000. Window 31 Aug 2026–2 Mar 2027. Purpose: share-price stability, shareholder value, and employee compensation. Planned ordinary 524,934 sh sized off 28 Aug close ₩3,810 (implied ₩2,000,000,000-ish). Holding period after purchase left blank — not cancel. Existing treasury (dividendable): 472,039 ordinary (1.53% of listed). Outside directors 4 present / 0 absent. Live ₩3,935 (+3.28% vs ₩3,810), volume 167,877, turnover ₩642m, mcap ₩121.3bn, 30,842,625 shares (Naver ~15:20 KST). ₩2.0bn is 1.65% of live mcap / 1.70% of shares.

Why up 1-3%, p≈30%: ₩2.0bn / 1.7% of a ₩121bn KOSDAQ name is a real six-month bid into the tape, not a rumor. Bullish because KB can dribble-buy for price support. 1-3% not 3-8% because it is only ~1.7% of mcap, hold not cancel, purpose also covers employee compensation, they already sit on 1.53% treasury, and the 15:11 print is live with the tape already +3.28% on ₩642m turnover. 30% because the ₩2.0bn and 524,934 share count are unambiguous, but same-session leftover after +3.28% is small and the program is modest versus today’s printed volume.

DART

Loan overdue ₩50.78bn (104.9% of equity) after workout halt; unpaid bonds ₩11.89bn; rehab review

What: After the 28 Aug halt of the creditor-bank joint-management (workout) that started 29 Apr, the stay on creditor rights ended. Loan principal+interest overdue ₩50,781,668,367 (principal ₩50,713,828,330 + interest ₩67,840,037) = 104.87% of stated equity ₩48,423,022,041 (2024 capital + later capital changes because the 2025 audit was an opinion refusal). Occurrence dated 28 Aug; FX on FX debt ₩1,380.30. They say they will promptly review and pursue a court rehab filing. Companion filing rcp 20260831800277 at 14:32: unpaid 33rd/34th unsecured bonds ₩11,885,052,544 (principal ₩11,495,000,000 + interest ₩390,052,544) = 24.54% of the same equity. 33rd ₩9.5bn issued 29 Apr 2025, unpaid from 17 Apr 2026; 34th ₩1.995bn issued 26 Jun 2025. Combined ₩62.67bn. Live ₩2,620 (flat vs ₩2,620), 0 volume, 0 turnover, 6,723,603 shares, mcap ₩17.62bn, lower limit ₩1,835 (Naver ~15:00 KST). Combined overdue is 356% of live mcap.

Why down 8%+, p≈55%: this is default plus a rehab path, not a working-capital miss. ₩50.8bn overdue is 105% of book and 288% of a ₩17.6bn KOSPI microcap, and the ₩11.9bn of unpaid bonds sit on top. Bearish because the 28 Aug workout halt lifted the stay and they are pointing at court rehab. 8%+ is the floor (KOSPI −30% limit is ₩1,835). Not leftover 8% from a live session: Monday has printed 0 shares, so none of this is in the tape. 55% because the KRW totals and the 28 Aug date are unambiguous, but the workout halt itself already printed Friday so the street had a session of warning, and a zero-volume name can halt or limit-down rather than trade the 8%+.

DART

Myeonmok Bondong 3 alley-housing job ₩77.4bn → ₩162.6bn (26.5% of 2024 sales); 247→512 units

What: Amendment of the 22 Dec 2025 (award 19 Dec) construction contract for the Myeonmok Bondong District 3 alley-housing improvement. Amount ₩77,373,686,403 (12.6% of 2024 consolidated sales ₩612,780,034,467) → ₩162,550,611,049 (26.5%). Counterparty: Myeonmok Bondong District 3 Alley-Housing Improvement Association. Seoul Jungnang-gu Myeonmok-dong 107-33. Scope 247 units / 15F+3B / GFA 32,418㎡ / 30 months from start → 512 units / 27F+3B / GFA 71,743㎡ / 35 months from start. No down payment; paid from unit sales. VAT excluded. Amount and period can slip with construction. Live ₩10,820 (−0.37% vs ₩10,860), high ₩10,860, low ₩10,630, mcap ₩109.8bn, 10.15m shares, volume 3,288, turnover ₩35.3m (Naver ~15:00 KST). ₩162.6bn is 148% of live mcap.

Why up 3-8%, p≈40%: ₩163bn / 26.5% of last-year sales on a ₩110bn KOSPI builder is a real orderbook scale-up, not a rumor. Bullish because the job roughly doubled (247→512 units). 3-8% not 8%+ because it is an eight-month-old amendment not a new award, there is no down payment, cash is unit-sale progress over 35 months, the buyer is a housing association, and the 14:33 print is live with the tape already −0.37% on 3k shares. 40% because the ₩162.6bn and 26.5% are unambiguous, but same-session leftover is small and the 512-unit permit may already have leaked.

DART

North America DRI-plant equipment to Tenova ₩15.87bn (15.74% of 2025 sales); 20% down

What: Confirmed ₩15,873,450,000 (USD 11,500,000 at ₩1,380.30 on 28 Aug) equipment supply for a North America DRI plant project. 15.74% of 2025 separate sales ₩100,854,566,106. Not conditional. Counterparty: TENOVA S.p.A. Region: US. Period 28 Aug 2026–30 Oct 2027. Advance yes: 20% / progress 70% / remainder 10%. In-house, not outsourced. Award 28 Aug. No same-type contract in the last 3 years. Amount and period can change with project progress. Live ₩12,300 (+8.85% vs ₩11,300), high ₩12,300, mcap ₩181.9bn, volume 11,289, turnover ₩133m (Naver 13:50 KST). ₩15.87bn is 8.7% of live mcap.

Why up 3-8%, p≈45%: ₩15.9bn / 15.7% of last-year sales on a ₩182bn KOSDAQ equipment name clears the 10% KRX bar, and 20% down is cash. Bullish because it is a signed Tenova DRI-plant book into the US, not a pipeline rumor. 3-8% not 8%+ because ₩15.9bn is 8.7% of mcap, cash dribbles to Oct 2027, the KRW can slip with construction, and the 13:48 print is live with the tape already +8.85% at the day's high on 11k shares. 45% because the USD/₩ and 15.74% are unambiguous, but the award was 28 Aug (three-day lag) so some of today's +8.85% may already be the leak, and leftover same-session is smaller than a cold open.

DART

Taean smart-city drone-experience content ₩549m (17.74% of 2025 sales); no prepay

What: Drone adventure content development and install inside the drone experience hall of the 2023 Taean-gun small smart-city project. Confirmed ₩549,450,000 (ex-VAT). 17.74% of 2025 consolidated sales ₩3,097,045,412. Not conditional. Counterparty: Daon I&C (drone swarm-control platform), 2025 sales ₩8.05bn, no relation. Chungnam Taean-gun. Period 31 Aug–31 Dec 2026. No advance. Mix of in-house and outsourced. Award 31 Aug. Live ₩2,035 (+11.20% vs ₩1,830), high ₩2,245, mcap ₩22.7bn, volume 432,795, turnover ₩883m (Naver 13:50 KST). ₩549m is 2.4% of live mcap and below today's ₩883m turnover.

Why up 3-8%, p≈35%: 17.7% of last-year sales clears the 10% KRX bar on a ₩22.7bn KOSDAQ XR name, so it can print. Bullish because it is a signed four-month content job, not a rumor. 3-8% not 8%+ because ₩549m is only 2.4% of mcap, there is no down payment, it is a municipal tourism slice of a 2023 project via Daon I&C, and the 13:45 print is live with the tape already +11.2% (high +22.7%) on ₩883m of volume — larger than the contract. 35% because the 17.74% figure is unambiguous, but same-session leftover after +11% is small and today's turnover already dwarfs the PO.

DART

Buyback trust ₩3.0bn (~5.3% / 530,973 sh) via Samsung 31 Aug–2 Mar; hold, not cancel

What: Board 31 Aug: Samsung Securities buyback trust ₩3,000,000,000. Window 31 Aug 2026–2 Mar 2027. Purpose: share-price stability and shareholder value. Planned ordinary 530,973 sh sized off 28 Aug close ₩5,650 (implied ₩2,999,997,450). Holding period after purchase left blank — not cancel. Existing treasury: dividendable 584,337 sh (5.8%) + other 14,044 (0.1%) = 598,381 (5.98% of listed 10,002,634; par ₩500; capital ₩5,001,317,000). Of that, 275,153 open-market + 309,184 already in a prior trust + 14,044 other. Remaining dividendable-profit limit ₩84,807,861,031. Outside director 1 present / 0 absent; auditor present. Live ₩5,780 (+2.30% vs ₩5,650), high ₩6,100, mcap ₩57.8bn, volume 188,204, turnover ₩1.102bn (Naver 13:30 KST). ₩3.0bn is 5.19% of live mcap / 5.31% of shares.

Why up 3-8%, p≈40%: ₩3.0bn / 5.3% of a ₩57.8bn KOSDAQ name is a real six-month bid into the tape, not a rumor. Bullish because Samsung can dribble-buy for price support. 3-8% not 8%+ because it is hold not cancel, they already sit on 6.0% treasury including 309k in an existing trust, and the 13:13 print is live with the tape already +2.3% (high +8.0% at ₩6,100). 40% because the ₩3.0bn and 530,973 share count are unambiguous, but same-session leftover after +2.3% is smaller and this is an add-on trust rather than a first program.

DART

Yangsan Central housing-association mixed-use build ₩164.9bn (25.56% of 2025 sales); halted for reverse-split

What: Construction award for the Yangsan Central Regional Housing Association mixed-use new-build at Gwangju Buk-gu Yangsan-dong 209-192. Confirmed ₩164,892,814,000 (ex-VAT). 25.56% of 2025 consolidated sales ₩645,114,465,728. Not conditional. Counterparty: Yangsan Central Regional Housing Association. Period 31 Aug 2026–31 Oct 2030 (50 months from groundbreaking; actual start may slip). No down payment. Cash every two months. Award 31 Aug. No same-type contract with this counterparty in the last 3 years. Last print ₩783 (flat, 0 volume) — halted since 12 Aug for a share consolidation until the day before new-share listing (KOSDAQ halt notice 7 Aug). Last cash 11 Aug ₩783, 100.89m shares, mcap ₩79.0bn (Naver 12:20 KST). ₩164.9bn is 209% of frozen mcap. Also KOSDAQ administrative-issue watch (price below ₩1,000, 5 Aug).

Why up 8%+, p≈20%: ₩165bn / 25.6% of last-year sales on a ₩79bn KOSDAQ builder is a real four-year housing book, not a rumor. Bullish because it fills the orderbook through Oct 2030. 8%+ because ₩165bn is 209% of the frozen mcap and none of it is in the tape (zero volume since 11 Aug). Not leftover 8% from a live session: no down payment, cash dribbles 50 months, and the buyer is a regional housing association. 20% because the KRW size and 25.56% ratio are unambiguous, but the next cash session is still halted for the reverse-split (expires the day before new-share listing, date not in this filing), so today almost certainly does not gap.

DART

Six 115k-dwt crude tankers ₩592.9bn (61.4% of 2025 sales) for a Greek owner; option +2 ships

What: Construction award for six 115k DWT crude oil tankers. Confirmed ₩592,894,062,000 (USD 429,540,000 at ₩1,380.3 on 28 Aug). 61.41% of 2025 consolidated sales ₩965,388,430,763. Not a large corporation. Counterparty: Greek shipowner, name withheld as a business secret until 28 Feb 2027. Period 28 Aug 2026–30 Jun 2030. No down payment; progress billing. Award 28 Aug. If option conditions hit they will file another two-ship contract. Live ₩14,980 (+8.32% vs ₩13,830), mcap ₩935.9bn, 62.48m shares, volume 153,244, high ₩14,980 (Naver ~11:31 KST). ₩592.9bn is 63% of live mcap.

Why up 8%+, p≈55%: ₩593bn / 61% of last-year sales on a ₩936bn yard is a real six-ship crude book, not a rumor. Bullish because it fills the orderbook through mid-2030. 8%+ because ₩593bn is 63% of mcap and the 11:28 print is already +8.3% at the day's high. Not leftover 8% from here: no down payment, cash dribbles four years, the Greek name is blank until Feb 2027, and a chunk is already in the tape. 55% because the USD/₩ and six-ship count are unambiguous, but same-session leftover after +8.3% can fade and the extra two ships are still an option.

DART

Open-market buyback 371,747 sh (~4.4% / ~₩1.00bn) 1 Sep–30 Nov; hold, not cancel

What: Board 31 Aug: KOSDAQ open-market buy of 371,747 ordinary shares (KRW amount left blank; sized off Friday close ₩2,690 → implied ~₩1.00bn). Period 1 Sep–30 Nov 2026. Purpose: shareholder value via price stability. Not cancel. Shinhan Securities. Existing treasury 1,295,518 sh (15.43%) after July buys. Daily cap 37,174. Issued 8,396,593 → 4.43% of shares. Live ₩2,705 (+0.56%), mcap ₩22.7bn, volume 28,336 (Naver 11:22 KST). 관리종목.

Why up 3-8%, p≈40%: ₩1.00bn / 4.4% of a ₩22.7bn 관리종목 is a real bid into a thin book that is defending the KOSDAQ ₩20bn mcap floor, not a cancel. 3-8% not 8%+ because it dribbles three months under a 37k daily cap, they already sit on 15.4% treasury from July, and the tape is already +0.56%. 40% because the share count is unambiguous but same-session residual is small and this is a repeat program.

DART

Spain K9 SPH export execution with Indra; amount withheld until 28 Nov; 20%+5% advances

What: Signed product-supply execution contract (labeled Derivative Contract) for Spain K9 self-propelled howitzers with INDRA SISTEMAS S.A. Amount and end date withheld as a business secret until 28 Nov 2026, then re-file. 2025 consolidated sales ₩26.70tn. Large corporation. Award 28 Aug. Advances: 20% after signing plus 5% by 31 Dec 2026. Confirms the 25 Mar / 24 Apr / 23 Jul rumor-clarification (unconfirmed) filings and replaces the 27 Jul investment-judgment confidentiality re-disclosure. Live ₩1,142,000 (−1.30%), ₩5.89tn cap.

Why up 1-3%, p≈45%: this is the confirmation the Mar–Jul unconfirmed rumor tape was waiting for — a signed Spain K9 execution with Indra, and 25% of the still-secret contract due as advances by year-end. Bullish because it turns a rumor into a cash-in-the-door export. 1-3% not 3-8% because this is a ₩5.9tn mega-cap, the KRW size is still blank, and three prior rumor-clarifications plus the 27 Jul filing already telegraphed the package. 45% because the signature is unambiguous but the session is already −1.3% into the 10:27 print and the street cannot re-rate on a number it cannot see until 28 Nov.

DART

5th CB put: buy/cancel ₩5.18bn face today (7.6% of mcap); ₩3.22bn still out; strike ₩7,020

What: Holders exercised the put on the 5th private unsecured CB (issued 30 May 2024, maturity 30 May 2029). Company paid ₩5.18bn for ₩5.18bn face on 31 Aug, OTC, own cash, and will cancel. Conversion ₩7,020 (last reset 3 Mar 2026). Remaining outstanding ₩3.22bn. Live ₩3,870 (−2.52%), ₩67.9bn cap. Strike is 81% above the live tape, so they put rather than convert. ₩5.18bn is 7.6% of mcap.

Why down 3-8%, p≈50%: this is a cash call, not a conversion. ₩5.18bn leaving a ₩68bn KOSDAQ constructor is 7.6% of the equity value, and the leftover ₩3.22bn CB still sits. Bearish because cash walks out the door on an out-of-the-money put; cancelling the notes only removes paper that was not going to convert at ₩7,020 vs ₩3,870 anyway. 3-8% not 8%+ because it is not a new dilutive issue. 50% because the put is unambiguous and printing live, but the name is already −2.5% and some overhang-relief bid can offset.

DART

DAPA naval spare-parts PO ₩1.71bn (13.04% of 2025 sales) through Oct 2027; no prepay

What: Confirmed ₩1.711bn (USD 1.240m at ₩1,380.30 on 28 Aug) naval spare-parts PO (turbocharger, engine, non-aircraft) to DAPA. 13.04% of 2025 sales ₩13.12bn. 28 Aug 2026–29 Oct 2027. No advance; pay as delivered. Outsourced, not in-house. Repeat DAPA work in the last 3 years. Live ₩22,300 (−3.88%), ₩128.3bn cap. ₩1.71bn is 1.3% of mcap.

Why up 3-8%, p≈45%: 13% of last-year sales clears the 10% KRX bar on a ₩128bn KOSDAQ name, so it can print. Bullish because it is a signed DAPA spare-parts book, not a pipeline rumor. 3-8% not 8%+ because ₩1.7bn is only 1.3% of mcap, there is no down payment, it is outsourced, it runs 14 months, and DAPA is a repeat customer. 45% because the 13% figure is unambiguous, but the 10:22 print landed while the tape is already −3.9%, so residual same-session is smaller than a cold open.

DART

2nd take-private tender: ₩900 cash for residual 16.08m sh (31.04%); 31 Aug–19 Sep, settle 22 Sep

What: Parent TKG Taekwang will buy every remaining ordinary share — 16,076,041 sh (31.04% of 51,794,579) at ₩900 cash. Window 31 Aug–19 Sep, settle 22 Sep. Buys all tendered stock, no minimum. They already hold 35,718,538 sh (68.96%) after the April ₩900 tender; 100% if this fills. Purpose is voluntary delisting. 1H26 NI −₩5.01bn. Friday close ₩600, ₩31.1bn cap, PBR 0.37. Pre-open, untraded. Bid is +50% vs Friday.

Why up 8%+, p≈65%: cash bid ₩900 is 50% above Friday ₩600 on a ₩31bn thin book, so tender arb pulls the first 09:00 print toward the bid. 8%+ is the floor; the KOSDAQ ±30% limit caps Monday at ₩780 (still 13% below ₩900). 65% because the filing is a formal second take-private and 09:00 is first tape, but April's same ₩900 bid already failed to hold and leftover holders may skip again.

DART

1 death at Cheongju Dongnam public-housing site (fall 11 Aug, died 29 Aug); partial work-stop

What: Worker fell 11 Aug while taking down safety railings to put up vertical nets at a rest area on the Cheongju Dongnam public-housing job; died in treatment 29 Aug. 1 dead, 0 injured. Cheongju labor office ordered a partial work-stop at that site the same day. Police and MoEL still investigating. Parent Kolon 002020 filed the same as a sub notice. Friday close ₩9,920, ₩252bn cap. Pre-open, untraded.

Why down 3-8%, p≈55%: this is Serious Accident Punishment Act tape on a small public-housing builder, not lost output from one site. 3-8% not 8%+ because the stop is partial and there is no bid ban. 55% because 09:00 is the first cash print and the death is unambiguous, but the fall was 18 days earlier and there is no CEO charge yet.

DART

6-month public-sector bid ban on ~98% of sales (₩255.9bn / ₩261.0bn)

What: PPS sanctioned them as an unfair contractor. Barred from national, local, and public-institution bids 4 Sep 2026–3 Mar 2027. Those customers are ₩255.9bn of ₩261.0bn sales (98%). 2025 audit was an opinion refusal, so the sales base is 2024. They may appeal.

Why down 8%+, p≈70%: this is the new-contract pipeline, not one lost customer. For a ₩97bn name that is operational. 8%+ is the floor because 98% of sales plus a thin book, printed after close so Monday is first tape. 70% not 90% because an appeal could leak a stay and the 2025 opinion-refusal already had hair on the name.

DART

53rd CB fully converted: 2.4m shares (~13.9%) at ₩2,500 / ₩6.0bn; list 11 Sep

What: Holders converted the entire ₩6.0bn 53rd private CB into 2.4m new shares at ₩2,500. That is 13.9% of the 17.3m share count. Listing 11 Sep. Zero left unconverted. Close was ₩1,481, below the ₩2,500 strike — they converted anyway. Session already +9.2%.

Why down 8%+, p≈55%: 14% new paper on a ₩26bn microcap is a float shock into list week. Bearish because supply hits a thin book, even though the strike is out of the money. 8%+ is list-week, not necessarily Monday. 55% because listing is two weeks out and today’s +9% already mixed the tape.

DART

₩5.39bn cash for 40% of Canada newco NaviComm (17.7% of equity) + ₩3.0bn short-term debt

What: ₩5.39bn cash for 40% of NaviComm Inc., a Canada company set up 10 Aug with blank financials, related to parent Kyungdong Navien. That is 17.7% of Commax equity. Same day they raise short-term debt ₩29.2bn → ₩32.2bn for working capital. Pay-in 2 Sep. No put, not a backdoor listing. Close ₩2,885 flat, ₩126bn cap.

Why mixed 1-3%, p≈35%: headline is global distribution, but it is related-party cash into a two-week-old shell with no P&L, plus extra leverage. Not control (40%) and not a third-party operating buy. ₩5.4bn is ~4% of mcap so 1-3% not 8%+. 35% because the sign is mixed and Monday can go either way.

DART

₩36.0bn cash for 8.01% of affiliate HLB Innovation (024850) via 3rd-party; 12.1% of equity, pay 7 Sep

타법인주식취득. 에이치엘비이노베이션 2,764,127주 / ₩36.00bn. 자기자본 대비 12.09% · 자산 7.53%. 취득후 8.01%. 제3자배정 현금, 납입 7 Sep. 목적 계열사지분·기업가치. 대상 2025 매출 ₩32.3bn / NI −₩33.7bn. 내재가 ₩13,025 vs 024850 close ₩14,760. Close 067630 ₩2,645 (+0.19%) mcap ₩322.5bn vol 180k — printed 16:00 after cash close, Monday open.

DART

Cancel already-held treasury 231,974 sh (~5.30%) / book ₩11.73bn on 4 Sep

주식소각결정. 기취득 자사주 231,974주 / 발행 4,378,709주 → 5.30%. 장부 ₩11.73bn. 신규 매입 없음. 소각예정 4 Sep (변동 가능). 상법 343조① 이사회, 자본금 불변. 모회사 자회사공시 20260828800502 동일. Live ₩55,800 (+3.72%) mcap ₩244.3bn vol 8.3k — printed after cash close, Monday open.

DART

Cancel already-held treasury 275,837 sh (~6.53%) / book ₩16.69bn on 4 Sep

주식소각결정. 기취득 자사주 275,837주 / 발행 4,224,646주 → 6.53%. 장부 ₩16.69bn vs live ₩28.77bn. 신규 매입 없음. 소각예정 4 Sep. 상법 343조① 이사회, 자본금 불변. 자회사 KPX케미칼도 같은 패키지. Live ₩104,300 (+2.56%) mcap ₩440.6bn vol 5.9k — after cash close, Monday open.

DART

Cancel-purpose buyback trust ₩15.0bn (~3.6% / ~996k sh) via Kiwoom through 28 Feb 2027

자기주식취득신탁계약체결결정. ₩15.00bn / 키움증권 / 2026-08-28~2027-02-28. 목적 소각. 취득예정 996,016주 (27 Aug 종가 ₩15,060 기준). 상장 27.46m주 → ~3.63%. 시총 ~₩415bn → ~3.6% mcap. 취득전 자사주 0. 이사회 28 Aug. Printed into cash close — next open is the cleaner tape.

DART

3rd-party raise 1.51m sh / ₩3.00bn (~8.1% dil) @ ₩1,981 (−10%); 1yr lockup; new-biz WC

유상증자결정. 보통주 1,514,385주 @ ₩1,981 = ₩2,999,996,685. 증자전 18,639,195주 → 희석 ~8.12%. 제3자배정. 기준주가 ₩2,201 · 할인 10%. 운영자금(신규사업) ₩3.00bn. 납입 23 Sep · 상장예정 21 Oct. 1년 보호예수. 배정: 주건테크 / SHEEN POWER / JOY BILLION. Live ₩2,375 (+10.21%) mcap ~₩44.3bn — session already ripping so residual same-session messy vs overnight dilution reset.

DART

Parent put on Robotics CPS: 6yr IPO clock, IRR 4% base / 15% listing penalty; max exercise >10% assets

풋백옵션등계약체결결정. 원익로보틱스 CPS 13,707,750주(₩350.0bn) 전부에 대한 부모 현금 풋. 그로스솔루션1호 28 Aug 예정 + KDB 납입 전. 상장 6년(1년 연장 1회). 콜 우선(만료 후 6개월); 미행사 시 풋 IRR 4%. 상장 위약 IRR 15% vs FMV 높은 쪽; 기타 위약 IRR 10%. 최대 행사가액이 직전연도 연결자산 10% 초과. Live ₩21,550 (+6.42%) mcap ₩1.664tn — session high ₩24,700 so CPS pop already in tape.

DART

Sub Wonik Robotics ₩350.0bn 3rd-party CPS (KDB + Growth Solution); parent put option pending

종속 원익로보틱스 제3자배정 CPS 13,707,750주 @ ₩25,533 = ₩350.0bn. 그로스솔루션1호 + 산업은행(첨단전략산업기금). 시설 ₩183.9bn + 운영/R&D ₩86.1bn + 타법인 ₩80.0bn(미정). 납입 20 Nov (변경 가능). 부모 풋옵션 별도 공시 예정. Live ₩20,600 (+1.73%) mcap ₩1.59tn.

DART

Nantong Huadong crane-control PO ₩13.74bn (11.78% of 2025 sales) through Sep 2027

단일판매ㆍ공급계약. ELECTRICAL CONTROL SYSTEM FOR CRANE. NANTONG HUADONG HEAVY MACHINERY (중국). 확정 ₩13,743,233,010 (USD 9,956,700 @ 1,380.30). 2025 연결매출 ₩116.63bn 대비 11.78%. 기간 28 Aug 2026–15 Sep 2027. 선수금 10% TT / 선적 70% / 시운전 후 20%. 자체+외주. 3년 동종 해당. Live ₩44,950 (+1.93%) mcap ₩231.5bn — session high ₩47,400 so partial in tape.

DART

Changwon A-lot: ₩27.0bn of ₩34.5bn land sale received today; close 4 Sep

기재정정 유형자산양도(원 15 Apr 2025). A필지 창원 신촌동 71-1 42,308.9㎡. 장부 ₩59.66bn / 2024 연결자산 ₩513.40bn = 11.62%. 실거래 ₩34.50bn. 금아하이드파워. 2차 중도금 ₩27.00bn 당일 수령, 잔금 ₩0.60bn 4 Sep. 관련 B필지 71-5 실거래 ₩20.40bn 처분일 4 Sep. Live ₩2,595 mcap ₩40.5bn.

DART

GLM-5.3 full weights are live — Friday’s Upcoming-release countdown is closed

HF API zai-org/GLM-5.3: gated=false, 141 FP8 safetensors shards, ~753.3B params, 50,116 downloads, lastModified 29 Aug 09:51 UTC. License tagged other, not MIT. The 14 Aug z.ai blog still says weights “soon in two weeks.”

Monday is the first HK/KR cash session with the flagship 753B downloadable. Flash was already priced Thursday; Friday’s pack still had the repo gated. Not a clean MIT dump.

Hugging Face

Z.AI 02513 faded Friday after Thursday’s Flash spike; weight drop is after the close

2513.HK Fri close HK$1,090 (−6.03% / −70; open 1,155, high 1,171, low 1,063, vol 3.86m). Thu close HK$1,160 (+12.62%). GLM-5.3 lastModified is Sat 17:51 HKT, after Friday HK cash.

Thursday’s +12% Flash rally faded into Friday. Monday is the first session that can reprice flagship open weights. No Monday print at pack time.

Yahoo 02513

MiniMax lifts Alibaba Cloud ceiling 220% to US$1.2bn as H1 compute burns

SCMP: MiniMax raised the 3-year Alibaba Cloud purchase ceiling 220% to US$1.2bn. 2026 spend cap nearly triples to US$300m from US$115m after using two-thirds of the old budget by end-June; 2027 US$400m and 2028 US$500m. H1 revenue +283% to US$116.6m. 00100 Fri HK$300.40 (−4.51%).

Alibaba Cloud capturing MiniMax training+inference spend even as 00100 sold off with the HK AI complex. Mixed-to-bearish MiniMax unit economics; overlay on 9988 cloud.

SCMP

Alibaba / MiniMax / ByteDance take 8 of 10 on Artificial Analysis text-to-video

SCMP: Alibaba Wan 3.0 ranks 1st on Artificial Analysis T2V (with audio), ahead of Google Gemini Omni Flash; a fal post-train of MiniMax H3 is 3rd, original open-weight H3 4th, ByteDance Seedance 2.0 5th. Chinese models occupy 8 of the top 10. Alibaba owns SCMP.

Weekend product confirmation for China AI listed/parent names on video-gen, not just chat. DeepSeek / Moonshot / SenseTime not named with numbers here.

SCMP

KOSPI Friday −1.79% to 6,788.88; SK hynix −4.45%, Samsung −3.38%

Yonhap: KOSPI −123.49 pts / −1.79% to 6,788.88, snapping a 3-day rise. Wait-and-see into Jackson Hole plus reported US semiconductor-tariff talk. Samsung ₩257,000 (−3.38%); SK hynix ₩1,653,000 (−4.45%). KOSDAQ +0.09% to 838.41. Foreigners net sold ₩1.7586tn on KOSPI; institutions ₩284.2bn; individuals +₩424.4bn. Won ₩1,372.5/USD at 15:30.

Missing Friday KR close vs the prior pack’s Thursday 6,912 print. Monday 09:00 KST opens from 6,788.88. Warsh hawkish Jackson Hole still unpriced in KR cash. PREOPEN is not a live gap.

Yonhap

HSI Friday +0.07% to 25,584.79; week still −1.63% vs 21 Aug

Yahoo ^HSI: Fri close 25,584.79 (+19.09 / +0.07%). Thu 25,565.74. Week vs Fri 21 Aug 26,009.46 is −1.63%. HSTECH Fri −0.33% to 4,605.15 (secondary). HKET www.hket.com 405 and invest.hket.com 500 — no HKET headlines this run.

HSI only flattened after Thursday’s −0.34%. HK AI names underperformed (02513 −6.03%, 00100 −4.51%).

Yahoo HSI

Samsung: zHBM around 2029 (maybe 2028); Taylor Fab 2 within 3 years

KED lede (body behind meter): Samsung’s US semiconductor chief said next-gen zHBM commercialisation around 2029, potentially as early as 2028, and a second advanced foundry fab in Taylor, Texas, online within three years to support US customers. Heat is the key hurdle. No extra capex figures on the public lede.

Near-term HBM leadership still SK hynix’s tape. Samsung is pointing at 2028–29 3D/zHBM and more US foundry. Do not fill paywalled body numbers.

KED Global

July industrial output flat, retail −2.4%, facility investment +7.5%

Yonhap, Sejong 31 Aug: July industrial production unchanged m/m. Mining & manufacturing +0.2% (electronic components); autos −4.5%. Retail −2.4% (durables/cars). Facility investment +7.5% m/m.

First official KR print of the week, into PREOPEN after Friday’s memory dump. Capex bounce is chip-cycle friendly; flat IP and weak retail are not a demand shock.

Yonhap

SKHYUSDT vs SKHYNIX/10 premium + BN funding

Share-equivalent hedge is 10 SKHY per 1 SKHYNIX. Dollar hedge leaves leftover Hynix beta. Indiana is packaging, not a convert-arb print. Best paper observation: BN ORD-scale book ~58 bp rich vs HL xyz:SKHX; predicted BN FR deeply neg into 08:00 HKT. Fade still vetoed. Paper.

Hynix marks 1 Sep ~07:00 HKT (not a full majors pass). SKHYUSDT ~164 vs xyz:SKHY 163.8; SKHYNIXUSDT ~1225 vs xyz:SKHX 1224.6. Two books stay split. Fade still vetoed. Paper.

Seoul paper rule did NOT fire on Mon 31 (+1.27% / +34.55% vs ≤−8% / ≥+40%). Last cash Mon 31. Prior 31 Aug 07:54 HKT majors left as history only — not refreshed this flash. Paper only.

Ticker	Follow	Note
NVDA	Y	Still live. AJ HC 27 Aug deep-dive. No new weekend memo. Public tape has no new numbers this morning.
MRVL	Y	Follow-up Y. @alojoh post-Q2 HC memo after −8%+. X 28 Aug 16:25 HKT. No newer MRVL post over the weekend.
MU	Y	Cheapest PEG in 22 Aug AJ article. Unchanged. CXMT 1H profit bomb is the China-memory overlay from Friday X, not a new AJ memo.
XPEV	Y	AJ $12.20 / IRON humanoid. Unchanged.
META	booked	AJ sold $598 26 Aug, rebought $569 27 Aug, sold $588 28 Aug (+3.4%). Tactical around strength, currently flat-ish. Not a new long.
AMD	Y	Corridor entries after the dump. Unchanged.
ORCL	Y	Wait, don’t chase.
PLTR / AAPL	N	Expensive growth. Unchanged.
EOSE	avoid	AJ red flags
TSLA	caution	Unchanged. Weekend Waymo/robotaxi posts are thematic, not a TSLA memo.
RIVN	N	Skip. @alojoh CFO commentary 30 Aug is personnel chatter, not a buy/sell memo.
SPCX	watch	AJ follow-up unchanged.

AJ US omit this morning: no new @alojoh after RIVN; Outlook still dark. MRVL + NVDA still the last follow-ups. Table is the standing 22–30 Aug list, not a new extract.

Last cash Mon 31 Aug. Attractive list empty besides Hynix desk. 000660 1,674,000 (+1.27% / +21k) V 3,980,661. For −211,935 / Inst −314,294 (Naver labeled KRX). shorts UNKNOWN (data.krx login-walled). SKHY RTH $164.58; AH $163.75. FX Yonhap 15:30 ₩1,368.60. prem vs RTH +34.55% (1:10). KIND/DART: still 28 Aug FTC group-status only. Seoul paper rule did NOT fire.

Monday official print (Yonhap): July industrial output flat, retail −2.4%, facility investment +7.5%. KED lede: Samsung zHBM around 2029 (maybe 2028) and Taylor Fab 2 within 3 years — body paywalled. Warsh Jackson Hole Friday (CNBC): summer inflation readings “do not tell me that underlying trends have meaningfully improved”; 2y UST +~8 bp to 4.31%; Sept hike odds 55.7% — unpriced in KR cash until 09:00 KST.

Thu 27 (not live): KOSPI +1.53% to 6,912.37 on NVDA then fade after BOK +25 bp to 3.00%. Indiana HBM packaging groundbreaking and SK Ulsan 49% DC sale already in Friday’s pack — not rehashed. No live trade recs.

Rubin CPX is back — and it uses HBM4

QT mingchikuo: just as the market had priced Rubin CPX off Nvidia’s roadmap, fresh checks say it is back, with HBM4 attach.

Tape: HBM4 content/demand bid if CPX returns. Supports SKHY/MU/Samsung HBM4 into the Rubin stack; sits on top of the afternoon Samsung-HBM4-lead card.

tweet

CXMT HBM3E breakthrough (The Information) → commodity DRAM tighter

“Bro WTF?” then TI link: China’s CXMT made a breakthrough in advanced memory / HBM3E. Jukan: a meaningful slice of CXMT DRAM wafers will divert to HBM3E, so commodity DRAM supply gets even tighter. Follow: good for ASMPT TCB bonders into CXMT. Zephyr: “Told u.”

Tape: near-term bullish DDR/LPDDR ASP. Medium-term China HBM3E is the SKHY/Samsung HBM overlay, not a 2026 sole-source kill.

tweet Bro WTF The Information

NVIDIA invests $3.5B in MediaTek

One-liner from Jukan. Zephyr QT: “Lol That’s a very interesting play.”

Tape: SoC/edge-AI partnership. Mild 2454.TW overlay; not a 000660 print.

tweet Zephyr

BYD op. margin 5.4% vs Tesla 1.4% in Q2’26

AJ: before government subsidies (industry subs both receive). H1’26 BYD 4.2% (+1.9ppt yoy) vs Tesla weaker. Later FSD/XPENG card at 21:52 is auto-product, not a semi memo.

Tape: AJ auto watchlist context. Not a new AI/semi HC idea.

tweet

CXMT margin is small-lot HBM; trio HBM looks cheap on wafers burned

Zephyr on CXMT post-IPO black ink: higher margin because they sell small amounts of HBM. Memory trio HBM ASP looks low once you count wafers consumed for DDR-equivalent bits.

Tape: softens “CXMT structural winner” read. Still a China-HBM 2027 overlay.

tweet

~112M base dies / ~20k WPM N3 for MU+SKHY 4EB HBM in 2027

Assuming 70% 8-high / 30% 12-high: ~112M base dies → ~20k WPM of N3, nearly 10% of 2027 N3 capacity. “Diversifying to Intel makes a lot of sense.” QT of the morning SKHY/Intel exclusive.

Tape: capacity math behind Intel Foundry optionality. Supports the afternoon exclusive as bottleneck arithmetic, not rumor-only.

tweet

Samsung ~70% capacity through 2031 in LTAs; S5→memory; SKHY Japan JV

Seoul Economic Daily via Jukan: Samsung Memory has allocated ~70% of production capacity through 2031 to LTAs (NVDA, MS, Google). Samsung DS is considering converting Pyeongtaek S5 foundry into memory as early as next year. SK Hynix is exploring further Yongin and Honam expansion plus a potential Japan JV fab; Jukan follow-up says the KR article flags Kioxia as a possible partner. Spot HBM3E ~4–5× LTA; KR DRAM export unit prices +37% even as volume −13%. Zephyr: his 1.5M WPM Samsung forecast looks conservative (1.7–1.8M).

Tape: core bullish ASP/tightness for 000660/005930. Capacity adds are multi-year; near-term is LTA lock + spot scarcity.

tweet Seoul Economic Zephyr WPM

TrendForce: Q3 PC DRAM now +18–23% QoQ (was +15–20%)

August revision vs July forecast. Zephyr eyes on the print.

Tape: commodity DRAM tightness confirmed by the tracker. Supports the SKHY/Samsung earnings path into US hours.

tweet

eRAM roadmap: foundries could bypass memory makers (long-dated)

Jukan on an eRAM roadmap: foundries might eventually take part of what has been the memory makers’ domain. “Still a long way off… no guarantee.” Follow: packaging scale → larger substrates; don’t ignore ABF.

Tape: long-duration narrative, not a live 000660 print. Mild ABF/substrate overlay.

tweet

Exclusive: SK Hynix considering Intel for HBM4E base dies

Herald exclusive (10:01 KST): SK Hynix is looking at Intel Foundry for a slice of 7th-gen HBM4E base dies. Outsourced base dies today are 100% TSMC 12nm-class. HBM4 TSMC base die is estimated 3–4× the cost of SKHY 1b core dies; LTAs make an immediate pass-through hard. Multi-vendor is cost + TSMC-concentration + custom-HBM optionality. Hot Chips: SKHY America also comparing Intel EMIB vs TSMC CoWoS-S/L. Company: cannot confirm the roadmap; some of the report is not accurate.

Tape: optionality, not an award. Mild INTC foundry overlay. Do not treat as a live 000660 print — exclusive hit 10:19 KST, after the 09:00 KST open dump.

tweet Herald

“Is this why the Korean stock market is falling right now?”

Chart overlay of the Herald Intel-base-die exclusive onto Monday KR cash. 000660 opened ~₩1,595,000 (−3.5% vs Fri ₩1,653,000), low ₩1,577,000 (−4.6%), live 14:46 KST ₩1,663,500 (+0.64%) V 2.34m (Yahoo).

Tape: correlation, not cause. The exclusive is 80 min after the cash open. Warsh hawkish Friday remains the cleaner Monday-open overlay. Bounce back through Friday close by mid-afternoon.

tweet

SemiAnalysis: SKHY/MU HBM4 speed miss; Samsung currently best HBM4

SK Hynix and Micron struggled to hit the highest HBM4 speeds. SKHY had to redesign its base die, delaying HBM4 shipments for Nvidia’s Rubin platform. Samsung currently has the best HBM4. Broadcom’s heavy Samsung HBM sourcing, a HBM3E drag, is now an advantage for OpenAI Jalapeño.

Tape: core SKHY HBM4 yield/speed risk. Pairs with Friday’s NVHBM-Samsung exclusive. Negative for the 000660 sole-source Rubin narrative.

tweet

Where Is Memory Headed? Rubin Ultra HBM4 / 8-high rumor

Jukan longform: sources say Rubin Ultra might be downgraded to HBM4 or from 12-high to 8-high. The 16-high to 12-high cut he could understand; this would be a joke-level step down if true.

Tape: HBM content/ASP risk on Ultra if stack height is cut. Weekend; morning missed Jukan.

tweet

CXMT begins mass production of LPDDR6

CXMT LPDDR6 into commercial phones. Ice Universe: first attach on Xiaomi 18 Fold + XRING O3, launch September.

Tape: China mobile-DRAM catch-up is in phones now. Mixed/negative medium-term for SKHY/Samsung LPDDR; HBM remains the SKHY moat.

tweet

Hana: CXMT 16nm yields >90%; 12-high HBM3E MP targeted 2027

Beijing Fab 1 4th-gen ~16nm done in 2Q26. DDR5 5,600MHz now, 6,400MHz planned. Multi-patterning still means more steps vs leaders, so bit/wafer and cost lag. If CXMT can do ~3–4 million 12-high HBM3E stacks in 2027, that could ease a China AI-GPU bottleneck.

Tape: commodity DRAM catch-up continues. 2027 China HBM3E is a medium-term overlay, not a 2026 print.

tweet

Scale-up boom to compensate for less HBM per GPU

Zephyr QT of tphuang SuperNode ramp (Huaqin >RMB10bn H2 supernode + 100 cabinets/month): “Everyone is blowing up the scale up world to compensate for weaker chips with less HBM per GPU.”

Tape: HBM content/GPU can fall while cabinets/switches/CCL rise. Mild SKHY-per-GPU risk. Pairs with the Rubin stack-cut rumor.

tweet

Weekly Market Valuation: S&P 7,712 Friday, +0.5% w/w, +2.8% vs Q2 close

AJ Saturday card: S&P 500 closed 7,712 Fri 28 Aug, +0.5% week-on-week vs 7,674, and +2.8% above the Q2 close — a level first reached 4 Aug. Gated article; the tape numbers are public in the card.

Tape: index at new-ish highs into Monday Asia. No fresh AI/semi datapoint in the public card. Follow-up is the gated valuation piece, not a new ticker memo.

tweet

AJ META round-trip: avg $562, sold $598, rebought $569, sold $588

Trade record: averaged $562 (18 Aug), sold $598 (26 Aug, +6.5%), rebought $569 (27 Aug), sold $588 (28 Aug, +3.4%). Tactical around strength, currently flat-ish META.

Tape: booked row stays booked. Not a new long. Overlay on the 22 Aug META “2nd-best entry” thesis — he is trading the bounce, not adding.

tweet

RIVN CFO commentary — personnel, not a buy/sell memo

Ex-IB, moved from a $16B mcap company to $243B; CFO after 1 year as MD at JPM. CV “opportunistic” / “risk taker.” No numbers, no entry.

Follow-up N. Do not treat as a RIVN idea.

tweet

TrendForce: Vera Rubin ASP ~2× GB300; GB300+VR stack ~$710bn next year

TrendForce via Jukan: Vera Rubin ASP expected at ~2× GB300. Combined production value of GB300, VR200, and VR300 next year projected at ~$710bn.

tweet

CXMT 1H26: RMB150bn rev (+873%), RMB77.6bn profit swing

CXMT 1H26 revenue RMB150.31bn (+873% YoY). Net profit RMB77.61bn vs a RMB2.33bn loss in 1H25 (1Q alone RMB33.01bn). Zephyr QT: revenue still +60–90% in H2 2026.

Tape: China DRAM catch-up is real P&L now. Mixed/negative medium-term for SKHY/Samsung commodity DRAM; HBM remains the SKHY moat. Folded Zephyr 19:58 QT + source reply.

tweet

Galaxy Z Fold 8 ramp hits DDI crunch at UMC 22nm

Korean media via Jukan: Samsung ramping Fold 8 on strong demand, but DDI supply is tight — UMC 22nm for the Fold 8 DDIs is fully utilized; Samsung asked for “super” priority. Also tapping next-year development volume.

Tape: handset demand + foundry/DDI bottleneck. Mild Samsung overlay; not a HBM/000660 print. Folded zdnet source reply.

tweet

AJ MRVL post-Q2 memo out; −8%+; “is now the time to buy?”

AJ published the Marvell updated-outlook deep dive. Share down >8%. Assesses new earnings path, risks, and whether to buy. Notes NVDA and Alphabet invested. HC-only numbers.

Tape: follow-up to the morning −11% / nailed-$0.94 avoid print. Public tape has the question, not the answer — do not treat as a fresh short.

tweet

[Exclusive] Samsung early lead in Nvidia NVHBM after HBM4

Jukan exclusive: Samsung Electronics poised as a key supplier for Nvidia’s customized HBM (“NVHBM”) after HBM4. NVDA pursuing a new AI-memory strategy for next-gen platforms.

Tape: core SKHY risk. Morning NVHBM/NVLink Fusion had AWS/Annapurna first; this puts Samsung in the early lead. Negative for 000660 sole-source narrative — watch share split. Not a DART.

tweet

Japan MOF July MLCC exports: vol +20%, price −1% mom

Zephyr one-word QT of Japan MOF July trade stats: average MLCC export price −1% mom, volume +20%, value +19%. “Murata...”

Tape: volume upcycle with soft ASP. Pairs with 27 Aug SEMCO 4Q26 hike card — volume confirms, price not yet. Mild 009150/Murata overlay.

tweet

Panasonic hikes CCL prices on copper foil / glass fiber

Jukan: Panasonic raising copper-clad laminate prices on rising copper foil and glass fiber costs.

Tape: CCL/PCB cost pass-through. Supportive packaging-materials; pairs with morning Meritz GB200/300 CCL chain.

tweet

DIGITIMES: wire-bonding bottleneck into 2027

DIGITIMES via Jukan: AI server growth is stressing not only advanced packaging but traditional packaging too. Wire-bonding equipment tightening; IC design faces a 2027 bottleneck as demand outstrips TSMC advanced packaging and classic bond tools.

Tape: packaging capacity risk beyond CoWoS. Equipment bid; SKHY-adjacent OSAT/packaging overlay. Folded the second DIGITIMES link tweet.

tweet

Micron Taiwan strike support hits 80% (prelim)

QT dnystedt: support for a Micron Taiwan strike over bonus pay hit 80% in a preliminary survey. Parallel to Samsung May, when a last-minute deal averted a walkout. Jukan: “Taiwanese people, rise up!!”

Tape: MU labor risk in Taiwan. Watch deal vs walkout. Mild memory-supply overlay, not a SKHY DART.

tweet

Anthropic may want a training chip

Jukan one-liner: “It sounds like Anthropic may be interested in developing a training chip.”

Tape: custom-silicon rumor only. Longer-term NVDA attach risk if real; near-term noise.

tweet

Home @levturtle not confirmed. US-hours priority tape flipped to memory ASP/tightness. citrini quiet. BowTied not re-confirmed.

• NEW: Samsung ~70% capacity through 2031 in LTAs; Pyeongtaek S5→memory; SKHY Yongin/Honam + possible Kioxia Japan JV (Seoul Economic via Jukan). Spot HBM ~4–5× LTA.
• NEW: TrendForce Q3 PC DRAM +18–23% QoQ (was +15–20%).
• NEW: CXMT HBM3E breakthrough (The Information) → wafer divert → commodity DRAM tighter; ASMPT TCB aside.
• NEW: Rubin CPX back with HBM4 (mingchikuo via Jukan).
• NEW: Zephyr N3 base-die math (~20k WPM / ~10% of 2027 N3) supports Intel Foundry optionality; Samsung WPM 1.7–1.8M.
• AJ: BYD vs TSLA margin card + FSD/XPENG — auto, not a semi HC memo.
• Afternoon leftover kept: Intel HBM4E base-die exclusive + SemiAnalysis Samsung HBM4 lead.

Overall: bullish, confidence medium (LTA/spot scarcity + DRAM hike + CPX/HBM4 vs multi-year capacity adds and China HBM3E). JSON: x-feed.json