[Tier-1 stub] DART filing: 투자설명서(일괄신고)
2026-09-15 17:20 KST: JB금융지주 (175330, KOSPI) appeared on DART todayRSS — 「투자설명서(일괄신고)」. Stub only from RSS title; body not fetched. Direction unclear/mixed until enrich.
Daily brief
mixed
confidence medium · Tue · Mon Hynix -7.1% foreign dump; paper Seoul still NO · AJ overnight fade AI selloff
Tue 15 Sep 2026 ~08:01 HKT — morning report
Research/paper only. No live.
watchlists Korea attractive EMPTY. AJ: no NEW ticker overnight (NVDA Y confirmation/color only — fade AI selloff + anti-pacing). Still Y AVGO / MU / MRVL / NVDA / XPEV / META / AMD + ORCL wait / TSLA caution / SPCX watch; EOSE avoid; NIO caution; EOS avoid.
funding ~08:01 HKT BN lastFR bp: BTC +0.4 / ETH +0.1 / SOL +1.0 · mark BN−HL SKHYUSDT↔xyz:SKHY +2.8 / SKHYNIXUSDT↔xyz:SKHX -2.2. Next BN 2026-09-15 16:00 HKT. Seoul rule NO. No fills/recs.
Hynix Mon cash 000660 1,683,000 (-7.12%) · For -1.22m Inst -266k Ind +906k · shorts UNKNOWN. SKHY Mon $175.63 · prem RTH +40.60%. Tue PREOPEN / NXT ~1697k — not cash. Paper Seoul NO (Leg A −7.12%, need ≤−8%). No NEW 000660 DART after Chongqing clarification 20260909800559.
DART NEW Me2on 201490 mixed 8%+ Kakao Games third-party ₩27.9bn (~30% shares) max-holder flip (after-close); NEW Osung AM 052420 up 8%+ sells affiliate Hwail Pharm for ₩52.93bn (~55% mcap). Also: Satoshi 223310 mixed 3-8%; KM Pharma 225430 mixed 8%+; HVM 295310 up 1-3%; GenCurix 229000 down 8%+ (halted); Gigavis 420770 up 1-3%; Doosan Fuel Cell 336260 up 3-8%; Hyosung Heavy 298040 mixed 1-3%; E8 418620 mixed 8%+; Hanshin 004960 up 1-3%; STX 011810 mixed 3-8% (halted).
HYROX Warsaw still closed (“TICKET SALES START SOON!”); no ping.
DART · material KR
Last scan 2026-09-15 16:01 KST · RSS n=50 · NEW Me2on 201490 mixed 8%+ Kakao Games ₩27.9bn control placement; NEW Osung AM 052420 up 8%+ Hwail Pharm sale ₩52.93bn · DROP SKAI final / Davorlink CB amend / IR·AGM·ELS noise
[Tier-1 stub] DART filing: 신탁계약해지결과보고서
2026-09-15 17:25 KST: 금호석유화학 (011780, KOSPI) appeared on DART todayRSS — 「신탁계약해지결과보고서」. Stub only from RSS title; body not fetched. Direction unclear/mixed until enrich.
[Tier-1 stub] DART filing: [발행조건확정]증권신고서(채무증권)
2026-09-15 17:26 KST: 대한항공 (003490, KOSPI) appeared on DART todayRSS — 「[발행조건확정]증권신고서(채무증권)」. Stub only from RSS title; body not fetched. Direction unclear/mixed until enrich.
[Tier-1 stub] Litigation
2026-09-15 17:10 KST: 코오롱티슈진 (950160, KOSDAQ) appeared on DART todayRSS — 「[기재정정]소송등의제기ㆍ신청(일정금액이상의청구)」. Stub only from RSS title; body not fetched. Direction unclear/mixed until enrich.
[Tier-1 stub] DART filing: 일괄신고추가서류
2026-09-15 17:11 KST: JB금융지주 (175330, KOSPI) appeared on DART todayRSS — 「일괄신고추가서류」. Stub only from RSS title; body not fetched. Direction unclear/mixed until enrich.
[Tier-1 stub] Single sales/supply contract
2026-09-15 16:53 KST: 한전기술 (052690, KOSPI) appeared on DART todayRSS — 「단일판매ㆍ공급계약체결」. Stub only from RSS title; body not fetched. Direction unclear/mixed until enrich.
[Tier-1 stub] DART filing: 투자설명서(일괄신고)
2026-09-15 16:36 KST: NH투자증권 (005940, KOSPI) appeared on DART todayRSS — 「투자설명서(일괄신고)」. Stub only from RSS title; body not fetched. Direction unclear/mixed until enrich.
[Tier-1 stub] Securities issuance results report
2026-09-15 16:44 KST: NH투자증권 (005940, KOSPI) appeared on DART todayRSS — 「증권발행실적보고서」. Stub only from RSS title; body not fetched. Direction unclear/mixed until enrich.
[Tier-1 stub] Share cancellation decision
2026-09-15 16:19 KST: 금호석유화학 (011780, KOSPI) appeared on DART todayRSS — 「주식소각결정」. Stub only from RSS title; body not fetched. Direction unclear/mixed until enrich.
[Tier-1 stub] DART filing: 일괄신고추가서류(파생결합증권-주가연계증권)
2026-09-15 16:26 KST: NH투자증권 (005940, KOSPI) appeared on DART todayRSS — 「일괄신고추가서류(파생결합증권-주가연계증권)」. Stub only from RSS title; body not fetched. Direction unclear/mixed until enrich.
[Tier-1 stub] DART filing: 기업설명회(IR)개최(안내공시)
2026-09-15 16:14 KST: GS (078930, KOSPI) appeared on DART todayRSS — 「기업설명회(IR)개최(안내공시)」. Stub only from RSS title; body not fetched. Direction unclear/mixed until enrich.
Kakao Games third-party placement ₩27.9bn (~30% new shares / ~29% mcap) — max holder flips on Nov close
What: after-close board stamps a Kakao Games-led control package — ₩27.9bn third-party placement (~30% new shares / ~29% of ~₩97bn mcap at market) plus a same-day old-share buy from the founder group that flips max holder to Kakao Games on the Nov 16 close (FTC-contingent). Why mixed: strategic Kakao Games capital and a 1-year lockup are constructive for a KOSDAQ game name, but the float jumps ~30% and control overhang / dilution math is the other side of the trade (Artist Studio / E8 / thin-KOSDAQ placement class). Why 8%+: both the cash face vs mcap and the share-count increase clear high-single-digit catalyst size on a sub-₩1tn ADV name that can auction hard either way; not a quiet 1-3% IR note. Why p≈48%: stamp landed 15:55 after the 15:30 cash close, so regular-session leftover is zero and the product is NXT/after-hours plus overnight→09:00 — real after-close window. Haircut from ~55% because close is weeks out (Nov + FTC), the name already printed −5.0% in cash today on other tape, and mixed control/dilution cards often gap then give back.
Sells affiliate Hwail Pharm stake for ₩52.93bn (~55.5% mcap); ₩14.1bn deposit today, balance Nov 19
What: after-close Osung boards a cash exit of most of its affiliate Hwail Pharm stake for ₩52.93bn (~55.5% of ~₩95bn mcap), with ₩14.1bn (~14.8% mcap) hitting the account today and the rest due 19 Nov. Why up: investment-recovery cash at a large % of mcap is immediate balance-sheet fuel on a thin KOSDAQ materials name — deposit alone clears mid-teens of mcap same day. Why 8%+: face ≫ mcap mid-single digits and near-term cash (deposit) already clears a high-single-digit band; thin ADV (~₩2.2억 today) can gap hard overnight (Sammok/Dual liquidity class). Affiliate/계열사 flag applies to the asset being sold, not to an affiliate supply restatement — still note it is portfolio recovery, not a new growth order. Why p≈45%: stamp 15:54 after cash close → leftover is NXT + overnight→09:00. Haircut because the balance is Nov-dated, today’s cash tape already +1%, and recovery exits sometimes print then fade once the deposit is known.
Early buyback of own CB 12th+14th — cash out ₩1.31bn (~5.2% mcap); KSD cancel; convert OTM vs spot
What: after-close board retires ₩1.3bn face of 12th+14th private CBs for ₩1.31bn cash (~5.2% of ~₩25bn mcap) and plans KSD cancel, funded with recent rights-issue proceeds; convert prices ₩4,225 / ₩5,693 sit ~37% / ~85% above spot and conversion only opens in 2027. Why mixed: orderly overhang cleanup / “CB gone” can print as a squeeze-or-cleanup positive on a thin microcap, but paying ~par to extinguish deep-OTM converts burns ~5% of mcap cash that just came in from the raise (QuantaMatrix class — cash leaving ≠ automatic relief when convert is OTM). Why 3-8%: cash face clears mid-single digits of mcap on an ~₩25bn ADV-thin name that can whip either way overnight; not 8%+ because the near-term dilution option was already soft (OTM + 2027 start). Why p≈40%: stamps landed 15:42–15:43 after the 15:30 cash close, so regular-session leftover is zero and the product is NXT/after-hours plus the overnight→09:00 auction — real after-close window, haircut because the name already printed −5.7% in cash on other tape and cleanup narrative is two-sided.
Files stay injunction on 14 Sep delisting decision; KRX extends halt and postpones cleanup trading until court rules
What: after the 14 Sep mcap-rule delisting print (cleanup was set 16–28 Sep, delist 29 Sep), KM Pharma amends its court stay to target the delisting decision itself, and KRX responds by extending the halt and shelving cleanup trading until the court rules. Why mixed: a granted stay could reopen a path to keep the listing and defer forced cleanup, but the ₩~45억 mcap sits far below the ₩200bn KOSDAQ floor and street already treats exit as the base case — injunction is delay/binary, not a rescue. Why 8%+: halted delisting/cleanup names gap violently on any resume or single-price window (Dawonsys / Willbes / GenCurix listing-risk class); no price limit in cleanup if it eventually runs. Why p≈45%: name is halted (frozen ₩698 / zero volume), so the band is unused in the tape — leftover is the court binary and whatever session exists after the ruling. Haircut from ~60% because the 14 Sep delisting/halt narrative is already public and the stay does not restore the mcap test.
Aerospace forging capex ₩41.57bn (~38.4% equity / ~6.5% mcap) through 2028 — multi-year spend
What: board green-lights ₩41.57bn (~38.4% of equity / ~6.5% of ~₩6427억 mcap) of new aerospace forging-equipment capex, with spend running through end-2028 (EUR 26.75m face). Why up: disclosed growth capex into aerospace forging capacity is a constructive capacity/narrative print on a specialty-metals KOSDAQ name and clears the unusual-capex bar vs equity. Why only 1-3% (not 3-8%): equity % / face-vs-mcap is the KRX disclosure bar, not a same-session price band — cash drips over ~2+ years with no attached order or near-term revenue, so near-term fuel is thin (multi-year capex class). Why p≈12%: enrich ~15 min after the 15:02 stamp and live is already +3.37% (high +5.5%), so the 1-3% band is already in the tape with little cash left — leftover is fade/hold into the close, not unused mid-session alpha.
GenesWell BCT domestic stop ₩5.88bn (~77.9% sales / ~12.8% mcap) + listing-eligibility review halt
What: GenCurix stops domestic GenesWell BCT testing (₩5.88bn / ~77.9% of 2025 sales / ~12.8% of ~₩45.8bn mcap) after NECA research-stage classification, and KRX extends the trading halt until a listing-eligibility substantive-review designation decision. Why down: losing the core domestic franchise (~78% of sales) is a cash-flow and narrative hit on a thin diagnostic microcap; KRX tying the stop to listing-eligibility review + extended halt is Dawonsys / Willbes / listing-risk class, not a routine ops note. A filed stay/lawsuit may eventually reopen service but does not undo today's stop or the review trigger. Why 8%+: main-business stop on ~78% of sales plus listing-review halt historically gaps hard on resume (limit-down class on ~₩46bn names); sales % is huge and equity risk is the listing/franchise package, not a 1-3% ops blip. Why p≈60%: name is halted (frozen at ₩1,957 / zero volume at enrich), so the 8%+ band is not in the tape — leftover is the resume/single-price gap after KRX decides designation. Haircut from ~70% because the inquiry halt already started 14 Sep 18:04 (some narrative overnight) and the company is already seeking a stay to resume service.
JP substrate-maker equip supply ₩13.0bn (~24.7% sales / ~1.0% mcap); 30% advance — unnamed CP
What: third-party overseas equipment order of ₩13.0bn (~24.7% of sales / ~1.0% of ~₩1.30tn mcap) into an unnamed Japanese substrate maker, with a real 30% deposit and delivery through Dec 2027. Why up: clean non-affiliate supply win with near-term advance cash on a semiconductor-equipment name; repeat customer (same-type deals in last 3 years) supports franchise continuity. Why only 1-3% (not 3-8%): sales % is the KRX disclosure bar, not the price band — face is ~1% of mcap (Vinatech / Woori Ro / <2%-mcap class, often drop); unnamed CP + multi-year drip through 2027 cut near-term fuel further. Why p≈8%: enrich ~7 min after the DART stamp and live is already +9.05% (high +10.1%), so the 1-3% band is already overshot — leftover is fade/hold into the close, not unused mid-session alpha.
US affiliate HyAxiom fuel-cell supply ₩322.2bn (~71% sales / ~9.3% mcap); 20% advance — affiliate haircut
What: parent books a ₩322.2bn (~70.8% of sales / ~9.3% mcap) fuel-cell system supply order into US affiliate HyAxiom for datacenter and C&I end-use, with a real 20% advance. Why up: large disclosed backlog plus near-term advance cash is constructive for the fuel-cell order book on a name that can trade on US power / datacenter narrative. Why only 3-8% (not 8%+): affiliate/계열사 supply cuts the band (Doosan Fuel Cell / HyAxiom class in desk rules) — this can be a pipeline restatement through the US sub rather than a clean third-party win, and delivery runs into Oct 2028 so face ≫ near-term cash after the advance. Sales % is the KRX bar, not the price band. Why p≈12%: enrich ~9 min after the DART stamp and live is already +10.5% (high +14% vs prior close ₩47,800), so the 3-8% band is already in the tape — leftover is fade/hold into the close, not unused mid-session alpha.
Twin US HV transformer supply deals via HICO America — combined ₩386.6bn (~6.5% sales / ~1.5% mcap); affiliate, no advance
What: parent books two HV transformer supply contracts into its US sales affiliate that already won the end projects — standard Hyosung US recognition path, not a third-party customer stamp. Why mixed: incremental US power-grid backlog is constructive for the HV franchise narrative, but the counterparty is 계열회사 (affiliate), there is zero advance, and cash dribbles through 2028–2029 on progress-bill. Why 1-3%: sales % (6.5%) clears the KRX disclosure bar but is not a price band — face is only ~1.5% of a ~₩25.7tn liquid KOSPI name (Vinatech/HMM/Samsung Heavy class); affiliate supply cuts the band further (Doosan Fuel Cell / HyAxiom). Why p≈18%: stamp landed at the cash open and live is already −2.35% at enrich, so leftover for an unused up-band is thin — mid-session liquid names usually decide in under a few minutes; treat any bounce as a small headline blip, not unused 3-8%.
Buys 35% of Dacom System for ₩4.44bn (~31% mcap); third-party placement ₩4.04bn closes; control to E8 Value-up No.1 fund
What: same-morning crystallizing of a known Aug placement into control change plus a new 35% bolt-on of a loss-making hardware co sized at ~31% of mcap. Why mixed: strategic cash and a new largest shareholder with lockup vs 30% share-count dilution and a loss-making target — Artist Studio / thin-KOSDAQ placement class, not a clean earnings print. Why 8%+: both the deal face and the float dilution clear mid-single digits on a ~₩142bn management name; these names often auction past 3-8% then fade (Sammok/Dual/Artist). Placement size was public since 7 Aug, so completion alone is not the surprise — control + M&A is. Why p≈10% (recut): into the cash open live already ~₩790 (+10.49%), so the 8%+ band is largely printed — leftover is fade/hold, not unused auction alpha.
KRX extends listing-disclosure committee deadline on delisting appeal — still under review, no decision
What: process extension only — committee vote on whether to delist is delayed for more diligence on the rehab plan, not a rescue or a delisting confirmation. Why mixed: more time can briefly lift 'not dead yet' hopes on a management/halted name, but the July finding that delisting criteria were met still stands and overhang remains. Why 3-8%: when/if single-price or trading resumes around the eventual committee outcome, these names usually gap mid-to-high single digits either way; extension itself is not an 8%+ cash event. Why p≈20%: name is halted today (tradableStatus halt), so next regular cash session may not print at all until KRX lifts the halt — leftover is contingent on a session existing.
Selected as contractor for Hwagok 6-dong Moatown A1 street-housing project — ~₩228.3bn (~19.9% sales / ~167% mcap); contract not signed yet
What: contractor selection (시공자 선정), not a signed EPC yet — face ₩228.3bn is ~19.9% of sales and ~167% of ~₩136bn mcap, but cash will drip over a multi-year housing rebuild with no advance disclosed. Why up: being named builder is incremental backlog for a cheap construction name (PBR ~0.16). Why only 1-3%: unsigned + progress-bill construction class (Samsung E&A / Seohan-style) — face/mcap looks 8%+ but near-term cash does not; expect a headline blip not a melt-up. Why p≈35%: after-close stamp into Tue open is a real leftover window on a thin tape (~14k shares today), but selection news often leaks and the band is small.
Asia news · Yonhap / KED / HKET / SCMP
2026-09-15 08:01 HKT / 2026-09-15 09:01 KST · Tue age-cut Mon+Tue only. 6 items. HKET unavailable (The Standard for HK close). Hynix Mon cash Naver day 1,683,000 (−7.12%) — not Yahoo quirk. Research only.
Z.ai (Zhipu) completes ~US$5bn HK financing for next-gen GLM
On 14 Sep Z.ai announced completion of ~US$5bn combined raise (~US$2bn share placement + ~US$3bn convertible bonds) for next-gen GLM models, self-training, compute and infrastructure; SEDAILY notes MiniMax raised ~$2bn in July and Alibaba/Tencent/ByteDance also tapping capital for AI.
Bullish for China AI spend/compute demand narrative, but dilution + cash-burn optics hit 2513.HK; supports continued HBM/GPU/infra pull from Chinese labs even if US frontier labs talk about slowing.
TechNodeHSI closes higher Mon; Z.AI slumps ~9%, MiniMax ~6.5%
Hang Seng rose ~111 pts / ~0.5% to ~24,917 on Monday with ~HK$192.6bn turnover; Hang Seng Tech muted ~4,317. Z.AI (2513) plunged ~9.1% after a share-and-bond sale plan to raise ~HK$39bn; MiniMax (0100) dropped ~6.5%.
Bearish near-term for HK-listed China model names (2513/0100) on financing overhang + global 'AI slowdown' chatter; HSI divergence shows AI/model names underperformed the broad HK tape.
TheKorean stocks sink as chipmakers, automakers tumble on AI slowdown fears
KED (14 Sep 11:26 KST) reported Seoul stocks tumbling Monday morning led by Samsung Electronics and SK Hynix after top AI executives (Anthropic, OpenAI, SpaceX, Google DeepMind) called to slow development, fueling fears of weaker HBM and advanced-chip demand.
Bearish for 000660/005930 and KOSPI semiconductor beta into Tue open; sentiment shock more than confirmed order cuts—watch whether Tue cash holds Mon foreign selling.
KEDSeoul stocks dip over 3% on Mideast uncertainties, AI warnings
KOSPI closed 6,684.37 (−225.54 / −3.26%) on Mon 14 Sep; Samsung −4.05% to 249,000 won, SK hynix printed −6.35% / 1,697,000 in some wires; desk Mon cash (Naver day) is ₩1,683,000 (−7.12%). Foreigners+institutions net sold ~4.46tn won; Brent stayed >US$108 after Hormuz transit meeting postponement.
Confirms size of Mon KR risk-off: AI pace scare + oil/rates. Tue PREOPEN Naver showed KOSPI indicative still soft (~−1.5%); chips remain the swing factor for KOSPI.
YonhapSamsung and SK Hynix reject KEPCO five-year electricity prepay plan
Industry sources Mon: Samsung and SK hynix rejected KEPCO’s proposal to prepay ~5 years of power bills (Samsung ~20tn won / ~US$14.8bn; SK hynix ~5tn won; combined ~25tn won / ~US$17bn per KED), citing uncertainty AI boom durability; would have funded Yongin/Honam grid buildout.
Mixed/slightly bearish signal on chipmakers’ willingness to lock multi-year power spend; grid/power for new fabs stays a bottleneck story. Capex discipline vs growth optionality for Hynix/Samsung watchers.
YonhapChina/HK traders muted ahead of Xi–Trump meeting; fragile detente
SCMP (14 Sep 14:03 HKT): Xi–Trump sit-down set for 24 Sep in the US is on traders’ radar but expectations for material results are slim; CSI 300 −~4% over past month, HSI −~1%, largely on US monetary-tightening fears.
Macro/HK risk event on calendar; low bar for positives, but any flare-up could add volatility to HSI/tech and KR risk assets already soft on AI/oil.
SCMPSK Hynix map
Korea line
000660 / SKHYNIX / xyz:SKHX
KR PREOPEN. Mon cash ₩1,683,000 (-7.12% vs Fri 1,812,000; For −1.22m Inst -266k Ind +906k — Naver). NXT PRE_MARKET ₩1,697,000 as of ~08:01 HKT — indication only (not Seoul cash). Samsung Mon ₩248,500 (-4.24%); NXT ₩248,000 (-0.40%). KOSPI Mon 6,684.37 (-3.26%). KIND/DART: no new clean 000660 keep (last Chongqing rumor reply 20260909800559). data.krx UNKNOWN. Mon tape: AI-slowdown narrative + foreign dump. BN SKHYNIXUSDT 1258.42 vs HL xyz:SKHX 1258.7 (~-2.2 bp). Paper.
ADR line
SKHY / xyz:SKHY
SKHY Mon RTH $175.63 (Fri $190.07, -7.60%). FX Yonhap Mon 15:30 1347.3. ADS=1/10 → parity theo ~$124.92; ADR prem ~+40.60% vs Mon KRW cash. BN SKHYUSDT ~177.35 vs HL xyz:SKHY 177.3 (~+2.8 bp). Still structurally rich; Mon ADR and KRX both sold off. Do not treat NXT as an ADR-implied gap. Paper.
Day’s best PAPER arb
PAPER: Prefer 09:00 KRX auction vs soft Mon cash (−7.12%) after AI-slowdown + foreign dump — BN↔HL SKHX venue gap ~-2.2 bp / SKHY ~+2.8 bp (tight). Secondary: ADR ~+40.6% rich to parity is multi-day research, not a clean same-session sniper. Seoul rule NO (Mon cash −7.12%, need ≤−8%; NXT ignored). CSOPSKHYNIX2L is 2× HK ETF — not the stock. Paper only. No live.
BN vs HL (Korea pair)
Charts refreshed 15 Sep 2026 08:28 HKT / 15 Sep 2026 09:28 KST (rolling 7d BN↔HL). SKHYNIX↔SKHX mark ~+17.6 bp · SKHY↔SKHY mark ~-2.2 bp · ADR/KRX prem ~+41.42% Research/paper only.
Live marks 2026-09-15 08:01 HKT / 2026-09-15 09:01 KST. SKHYNIX↔SKHX mark ~-2.2 bp (BN 1258.42 / HL 1258.7) · SKHY↔SKHY mark ~+2.8 bp (BN 177.35 / HL 177.3) · ADR/KRX prem Mon ~+40.6%. BN lastFR SKHYNIX +1.70 bp / SKHY +0.00 bp; next BN 2026-09-15 08:00 HKT. HL hourly funding SKHX +0.64 bp / SKHY -0.03 bp. Charts last on disk (rolling 7d) — marks above are live. Research/paper only.
Bottom strip is ADR vs KRX cash premium (SKHY×10×USDKRW/000660−1), not the old backtest equity curve. Paper.
Majors BN lastFR: BTC +0.36 bp / ETH +0.13 bp / SOL +1.00 bp (HL majors not in this snap). Seoul paper rule NO (Mon cash −7.12%, not ≤−8%). NXT not a Seoul-rule input. Excluded CSOPSKHYNIX2LUSDT. Paper only.
US / AJ
| Ticker | Follow | Note |
|---|---|---|
| AVGO | Y | Live. Post-Q3 beat + raised guidance; stock sold off. AJ memo + X teaser. No new overnight AJ idea. Body paywalled. |
| NIO | Y caution | China concentration. Unchanged. |
| EOS | Y avoid | Separate from EOSE. |
| NVDA | Y | Still live. Overnight AJ: fades AI selloff (usage/agents + compute-intensity if pacing); Trump ‘golden goose’ / anti-pacing color. Confirmation only — not a new add. |
| MRVL | Y | Follow-up Y. Unchanged. |
| MU | Y | Cheapest PEG (22 Aug AJ). Overlay: Jukan MU HBM path; Zephyr 12Hi→8Hi mix pushback; Mon KR dump + MU Taiwan labor still on desk. |
| XPEV | Y | AJ $12.20 / IRON. Unchanged. |
| META | booked | Tactical; no new long. |
| AMD | Y | Y confirmation: AJ buy flag $453 (2 Sep) → Fri $516. Unchanged overnight. |
| ORCL | Y | Wait sleeve (AJ). Beat+guide; HC models enter-vs-wait — still wait, don’t chase. No overnight change. |
| PLTR / AAPL | N / event | PLTR expensive. AAPL event sleeve — not AJ idea. |
| EOSE | avoid | AJ red flags |
| TSLA | caution | AJ robotaxi race-to-bottom; overnight soft reinforce via Elon pacing / talking-book color. Still caution not a long. |
| SPCX | watch | Unchanged. |
| RIVN | N | Skip. |
2026-09-15 08:01 HKT / 2026-09-15 09:01 KST. US cash overnight after Mon AI-selloff tape into FOMC 15–16 Sep. AJ overnight: no new ticker — NVDA Y confirmation/color only (fade selloff + anti-pacing). Outlook AJ mail not refreshed this pack. Do not invent targets. Paper.
Korea / KRX
KRX PREOPEN. Attractive list EMPTY besides Hynix desk. Mon close: KOSPI 6,684.37 (-3.26%); 000660 1,683,000 (-7.12%); 005930 248,500 (-4.24%). NXT: 000660 1,697,000 (indication); 005930 248,000 (-0.40%). SKHY ADR Mon $175.63; prem ~+40.6% vs Mon parity. No overnight clean 000660 DART keep. data.krx missing. Mon AI-slowdown + foreign dump already printed.
DART leftover into Tue open: E8 418620 mixed/8%+/p≈50% (control + Dacom bolt-on, thin KOSDAQ — auction window); Hanshin 004960 up/1-3%/p≈35% (unsigned Moatown contractor selection — construction class); STX 011810 mixed/3-8%/p≈20% (delisting-committee extension — halted, contingent). Stubs purged.
No new Korea attractive idea beyond Hynix desk + those leftovers. No live trade recs. Paper.
X · priority
2026-09-15 13:38 HKT / 2026-09-15 14:38 KST · Home unavailable. Sentiment mixed, medium. NEW afternoon: Jukan Samsung +7–10% mobile DRAM/NAND (Apple/China OEMs); AJ Trump vs AI-doomerism credit; Zephyr RT SemiAnalysis GPT-6 Astra loop-transformers. PRIOR overnight AJ selloff-fade / anti-pacing + Mon memory/WFE tape. AJ: no new ticker.
Samsung pushing +7–10% mobile DRAM/NAND to Apple + China OEMs (Q3 China / Q4 Apple)
Jukan cites Korean media (Sisa Journal-e): Samsung seeking another round of mobile DRAM and NAND price hikes. Renegotiation with Chinese smartphone makers for Q3 shipments and with Apple for Q4, centered on an additional 7–10% vs existing levels.
Constructive Samsung memory ASP / mobile DRAM+NAND (005930) and KR memory pricing tape into the US session. Soft peer ASP spillover; not a clean 000660 HBM-only sniper.
tweet Sisa Journal-eAJ credits Trump for standing vs AI doomerism / ‘safety’ as moral high ground (IPO cough)
AJ: nearly everyone fell in with AI doomerism because ‘safety’ is the easy brownie-point stance. Harder is calling out those who exploit AI fear for personal/political/commercial gain (IPO jab). Thanks Trump for standing firm.
Continues overnight anti-pacing / golden-goose policy tape — constructive for US AI-capex / NVDA near-term. No new ticker; standing Y confirmation only.
tweetRT SemiAnalysis_: GPT-6 Astra loop transformers — deeper without bigger params
Zephyr amplifies SemiAnalysis_: GPT-6 Astra reportedly uses loop transformers (re-pass layers) so depth/compute rises without raising parameter count — a tell labs aren’t scaling param sizes as aggressively.
Mixed AI-infra: still compute-hungry (depth/FLOPs), soft caution on ever-larger-param narrative. Not a KR auction input.
tweetAJ: today’s AI selloff ‘most stupid since DeepSeek’ — usage rises even if development halted
AJ: even if AI development were halted, performance stays at today’s level and usage still rises — operators throw more agents at problems as cheaper tokens unlock. Calls today’s selloff the stupidest in a long time, probably since DeepSeek.
PRIOR morning — constructive for standing Y NVDA/AI sleeve — not a new ticker.
tweetAJ: slowing AI progress increases compute required (less-efficient models burn more tokens)
AJ: advanced systems use compute more efficiently. Slowing the pace leaves the industry stuck longer on less-efficient models that consume more compute for the same task — ICE-engine analogy (stop improving the engine, still drive, burn more fuel per mile).
PRIOR morning — counter to ‘pacing = less GPU’ linear thesis. Standing Y confirmation only.
tweetAJ: best thing about owning Nvidia — CEO doesn’t post stupid sh*t
AJ one-liner praising Jensen/NVDA governance/comms contrast vs other mega-cap CEOs.
PRIOR morning — soft affirmation of standing Y NVDA — color only.
tweetAJ: pause Anthropic/OpenAI IPOs instead of pausing AI — ‘don’t kill the golden goose’
AJ into Trump AI comments: prefer pausing Anthropic/OpenAI IPOs (exit liquidity) over slowing US AI; agrees ‘don’t kill the golden goose.’ Thread also: Trump calling slowdown advocates traitors; predicts Elon/Dario/Sam go quiet before Xi meeting; Elon ‘fallen behind’ / talking-his-book on pacing.
PRIOR morning — policy tape constructive for AI-capex / NVDA. Soft TSLA caution reinforce. No new ticker.
tweet golden goose Trump/traitorsBTB on pacing narrative: if AI growth slows, play the ‘slow for society’ cover
BowTiedBull QT Burry-tracker claim that OAI/Anthropic warning talk can cover slower growth ahead of IPOs — if numbers disappoint, say you slowed for the greater good. Follow-up: Dario needs an angle if growth slips.
PRIOR morning — mixed skepticism on lab pacing sincerity. Not a KR cash input.
tweetRT oneill_c: alignment/safety push may squeeze RL-env vendors toward in-house lab QC
Zephyr amplifies: higher safety/alignment bar could hurt RL-environment companies’ unit economics; poor-quality envs → misalignment, so labs may keep QC in-house.
PRIOR — secondary AI-infra vendor color. Not a KR auction input.
tweetTrendForce: why 4-Hi HBM isn’t enough; Jukan — makers wouldn’t build it anyway
Jukan links TrendForceInsights arguing 4-Hi HBM isn’t enough for datacenter needs; personal add: memory makers wouldn’t produce 4-Hi even if bandwidth were “enough.”
PRIOR Mon anti-de-spec tape vs SemiAnalysis 4-hi. Constructive higher-stack / ASP mix (MU / 000660 / Samsung).
tweetChina AI labs (Zhipu/Moonshot/DeepSeek) ~$35B raised since Jun; path ~$60B by Jan 2027
Zephyr: ~$35B in 4–5 months; path ~$60B by Jan’27 with IPOs. Capital + efficiency enough to stay competitive; expects testing/safety scrutiny, not a near-term hard training-compute cap.
PRIOR — still relevant China AI funding color for morning AI strip.
tweetKED: TeraFab largely done front-end equip orders (ASML/AMAT/LRCX/TEL); pilot YE2027 / ops 2028
Korea Economic Daily via Jukan: TeraFab largely completed Q2 front-end tool orders; 2H staging back-end/packaging. Pilot end-2027, ops 2028.
PRIOR — constructive KR foundry-build / WFE / packaging-equip tape.
tweetSEMCO + Qualcomm 2.1D “organic bridge” alliance (bypass Intel EMIB)
TheElec via Jukan: Samsung Electro-Mechanics + Qualcomm on organic-bridge AI multi-die 2.1D packaging as EMIB alternative.
PRIOR — constructive SEMCO FC-BGA / advanced-packaging pivot after Mon glass delay.
tweetMicron Taiwan Taoyuan union rejects bonus package; strike threat
Jukan QT: Micron Taiwan Taoyuan union rejected NT$1m appreciation bonus and threatened strike; wants 15% profit-sharing when profitable plus downturn safeguards.
PRIOR — soft MU Taiwan labor overhang still live into Tue.
tweetSoftBank −11%
Jukan one-liner with chart: SoftBank −11% (Mon session).
PRIOR — risk-off AI-holdco color still on the board.
tweetX · tape
mixed
confidence medium · Home unavailable · inferred from priority public tape · 2026-09-15 13:38 HKT / 2026-09-15 14:38 KST. NEW afternoon: Samsung memory ASP hike (Jukan) + AJ anti-doomerism credit + SemiAnalysis loop-transformer nuance. Sentiment still mixed (no clean shift vs morning).
Method: api.fxtwitter.com profile statuses + status enrich (afternoon intraday X watch). Home unavailable — no browser login. Cutoff: newer than 2026-09-15 08:01 HKT. Skipped AJ emoji/‘Nailed it’ thin color; Citrini/BTB quiet. Full machine JSON: _pack/20260915-x-feed.json.
Overall: mixed — afternoon adds constructive KR mobile memory ASP (Samsung +7–10%) and continues US AI anti-pacing policy color, but SoftBank −11%, MU Taiwan labor, China AI raise, and GPT-6 ‘deeper not bigger’ scaling nuance keep the tape selective. Not a clean KR auction sniper from X alone.
AJ US watchlist: Y unchanged (AVGO/MU/MRVL/NVDA/XPEV/META/AMD); ORCL wait; TSLA caution (reinforced overnight); SPCX watch; EOSE avoid; NIO caution; EOS avoid. No invented tickers.