Zynergy Wire
Wednesday, 11 November 2026 · Private edition
粵語 0:00
Archive · Wednesday, 11 November 2026 Back to Today All dates

Archive

Wednesday, 11 November 2026

AI 9 items

OpenAI first-party (November 10, 2026) introduced Businesses in ChatGPT in beta. Through Business Hub, eligible businesses can claim and verify a ChatGPT profile, update details, and add self-service product feeds with pricing and availability. Hub also shows aggregated insights (mentions, website clicks, discovery topics) with a privacy floor: OpenAI says it does not share users’ conversations or Business Agent chats with businesses, and insights appear only above unspecified thresholds. After claiming, eligible U.S. ecommerce and online-service businesses can build, test, and launch a Business Agent — a clearly labeled, separate conversation from the user’s original ChatGPT thread and from ChatGPT’s independent answers — and may sponsor that agent as a destination for ChatGPT Ads (placements labeled and separate from responses). Shopify merchants can connect from Business Hub or Shopify admin; eligible Shopify Catalog products are already in ChatGPT. Vendor: “1.2 billion” people use ChatGPT each week. Soft-title: platform / business-presence + agent surface in beta, not a new frontier model and not an Odessia acquisition (no OpenAI first-party for that claim this run).

Why it matters

Same-morning first-party ship of a business claim/insights layer plus a U.S. Business Agent surface inside ChatGPT — top of the platforms/agents ranking lane.

OpenAI (first-party) — vendor-confirmed
More ▾ Less ▴
Confirmed
What happened

OpenAI published “Introducing Businesses in ChatGPT” dated November 10, 2026, introducing Businesses in ChatGPT in beta. Through Business Hub, eligible businesses can claim and verify a ChatGPT profile, update key details, and add self-service product feeds with pricing and availability. The Hub also shows aggregated insights (mentions, website clicks, and discovery topics), with a privacy floor: OpenAI says it does not share users’ conversations or Business Agent chats with businesses, and insights appear only when thresholds are met (those thresholds are unpublished). After claiming, eligible U.S. ecommerce and online-service businesses can build, test, and launch a Business Agent — a clearly labeled conversation, separate from the customer’s original ChatGPT thread and from ChatGPT’s independent answers — and may sponsor that agent as a destination for ChatGPT Ads, with sponsored placements labeled and separate from responses. Shopify merchants can connect from Business Hub or Shopify admin; eligible Shopify Catalog products are already in ChatGPT. Vendor: “1.2 billion” people use ChatGPT each week. Soft title: platform / business-presence plus a U.S. Business Agent surface in beta — not a new frontier model and not an Odessia acquisition (no OpenAI first-party for that claim this run). Stamp: OpenAI dateline November 10, 2026, after Tue LIVE.

Watch next
  • Category and geography expansion beyond eligible U.S. ecommerce and online services
  • Advertiser terms and sponsored-agent economics
  • Any official OpenAI note on the secondary Odessia/acquisition claim
  • Published insight-threshold or verification details
Limitation

OpenAI first-party post Confirmed (related Business Hub page also live). The weekly-user figure and beta scope are vendor-stated. Insight thresholds are unpublished. Business Agents are limited to eligible U.S. categories. No Odessia mention on the primary or related hub page; that claim stays Developing and is not this card. Vendor-confirmed ≠ independently reproduced.

Google DeepMind first-party (November 10, 2026, Carolina Parada) introduced Gemini Robotics 2.5, described as its most capable vision-language-action (VLA) model to date, now rolling out to robotics partners (not a public API). Vendor evaluations across 200+ distinct tasks (tongs, gift-box bows, pipetting): 30% to 100% higher success vs prior baselines depending on task type, and tasks 20% to 30% faster. Embodied-reasoning plus 2.5 actions were used in an Apptronik Apollo 2 proof-of-concept on Mercedes-Benz production lines in Kecskemét, Hungary, and Sindelfingen, Germany, for automotive kitting; vendor: 96% success. Safety: inference-time reasoning guidance; vendor: safe task completion around obstacles +33% and around nearby people +43%. DeepMind is also releasing ASIMOV-Jailbreak, an open benchmark of robotics jailbreak prompts; vendor: Gemini 4 Argon at 99.5% jailbreak resistance vs 82% for Gemini Robotics ER 2. Soft-title: partner-gated robotics VLA + measured factory PoC, not a humanoid hardware first-ship and not a Gemini 4 Carbon recard (Carbon official intro is Oct 28). Distinct from Nvidia’s CoRL/GR00T and Alpamayo cards.

Why it matters

Same-morning robotics-model ship with a named on-line factory PoC and a new open jailbreak bench — measured-deployment bar for the robotics lane.

More ▾ Less ▴
Confirmed
What happened

Google DeepMind, in a November 10, 2026 post by Carolina Parada, introduced Gemini Robotics 2.5, described as its most capable vision-language-action (VLA) model to date, now rolling out to robotics partners — not a public API or a factory-partner GA. Vendor evaluations across 200+ distinct tasks (tongs, gift-box bows, pipetting) report 30% to 100% higher success versus prior baselines depending on task type, and tasks 20% to 30% faster. Embodied-reasoning plus 2.5 actions were used in an Apptronik Apollo 2 proof-of-concept / pilot on Mercedes-Benz production lines in Kecskemét, Hungary, and Sindelfingen, Germany, for automotive kitting; vendor: 96% success. Safety: inference-time reasoning guidance; vendor: safe task completion around obstacles +33% and around nearby people +43%. DeepMind is also releasing ASIMOV-Jailbreak, an open benchmark of robotics jailbreak prompts; vendor: Gemini 4 Argon at 99.5% jailbreak resistance versus 82% for Gemini Robotics ER 2. Soft title: partner-gated robotics VLA plus a measured factory PoC — not a humanoid hardware first-ship and not a Gemini 4 Carbon recard (Carbon’s official intro is October 28). Distinct from Nvidia’s CoRL GR00T and Alpamayo cards. Stamp: DeepMind dateline November 10, 2026, after Tue LIVE.

Watch next
  • Public API or weights
  • Broader site metrics beyond the two Mercedes PoC locations
  • Third-party ASIMOV-Jailbreak runs
  • Keep Fei-Fei BEHAVIOR/CoRL as cap-held color at most, not a second card
Limitation

Google DeepMind first-party Confirmed. Success, speed, safety, and the 96% Mercedes figure are vendor-reported; the 96% result is a named partner PoC/pilot on two sites, not a fleet SLA, and is not independently reproduced here. Access is partner-only; no public API was announced. Title tightened to “rolls out … to partners” to match the page (robotics partners, not a factory-partner GA).

Nvidia Technical Blog (November 10, 2026, Marco Pavone and Boris Ivanovic) introduced Alpamayo 2 Edge, a 2B-parameter adaptive-reasoning vision-language-action model for real-time onboard driving on DRIVE AGX Thor. Built on a 2B Nemotron-based reasoning backbone from Cosmos 3 Edge, with a diffusion action expert; it uses surround cameras, temporal context, and navigation inputs and generates Chain-of-Causation reasoning only when needed. Vendor: closed-loop RL in AlpaGym improved closed-loop driving 48.9% over the best checkpoint without that RL; pixel-to-trajectory latency 160–190 ms on a documented Thor-X / DRIVE OS recipe with TensorRT and DFlash2; in AlpaSim it achieves about 60× the distance between at-fault incidents vs Qwen-Drive-1.0 (4B) on Nvidia’s chart. Weights default BF16 on Hugging Face (nvidia/Alpamayo2-Edge); AlpaJudge (reference-based and reference-free) is the companion reasoning evaluator. Soft-title: open/edge AV software model, not a robotaxi fleet deployment and distinct from Confirmed #5 (CoRL GR00T / Isaac Lab) and from the already-LIVE Nvidia board probe.

Why it matters

Same-morning first-party edge AV model + weights + published onboard latency recipe — a second Nvidia product lane, not a governance recard.

Google Keyword blog (November 9, 2026, Amin Vahdat) expanded the eighth-generation TPU portfolio with two additional systems: TPU 8tx and TPU 8i Max — printed names exactly; this is not a new-chip recut of April’s TPU 8t / TPU 8i and not Ironwood (still 7th-gen on the product page). 8tx uses the same silicon as TPU 8t, paired with 5th Generation AMD EPYC CPU hosts, for hardware-level confidential inference aimed at regulated enterprises and governments; vendor: up to 2× lower latency vs TPU 8t with a “near-zero performance penalty,” also useful for small-scale training / private fine-tune and RL on one system. 8i Max puts two TPU 8i chips on one package, doubling compute, SRAM, and HBM bandwidth vs 8i for massive context windows and rack density. Soft-title: AI infra / system configs, not a training-cluster first-rack or a priced GA SKU. Google did not give availability dates or list prices in the announcement.

Why it matters

First-party hyperscaler system expand for confidential inference and long-context density — the Expand B ship in this window.

Nvidia Developer Blog (November 9, 2026, Sasa Docca) at CoRL in Austin introduced GR00T 2, Isaac Lab 3.0, and new open industrial benchmarks. GR00T 2 is an open frontier world action model built on NVIDIA Cosmos, trained including 100,000 hours of open egocentric data; vendor: completed 40 of 50 evaluated tasks at least once with no task-specific demos; 71% average success with one demo per task, 95% with five; API for agent skill sequencing and recovery. Isaac Lab 3.0 lets agents set up training environments, switch PhysX ↔ Newton (deformables such as cables/cloth), and apply a new RL post-training library; vendor: up to 2.5× training throughput vs Isaac Lab 2.3 + PhysX on a vision dexterity bench (docs path v3.0.0 updated Nov 9). New open benches: DexBench (RLWRLD + Nvidia; 15 tasks / four scenes via Isaac Lab-Arena) and VABAR (UC Berkeley + Nvidia; six industrial tasks); Isaac Lab-Arena 0.4 developer preview. Soft-title: robotics software / platform / benchmarks, not a humanoid hardware first-ship; distinct from the already-LIVE Nvidia board probe and from Confirmed #3 (Alpamayo AV). Do not treat an older GTC Isaac Lab 3.0 early-access mention as this CoRL ship.

Why it matters

First-party world-action model + named sim major-version + published industrial benches — the CoRL product card, not a second board-probe story.

Fireworks AI first-party bulletin (last updated November 9, 2026) updated its October 2026 security incident. An earlier Oct 9 bulletin had limited known access to Fireworks-stored secrets used in evaluation jobs. The company now says it also found impact to account and billing information, and for a limited set of accounts, to secrets, datasets, and incomplete models. A third-party cloud provider that hosted Fireworks R&D machines had a security incident on October 23; on October 24 the unauthorized party used that path to reach Fireworks development machines and run a workload that caused additional customer-data impact. Fireworks says it detected the same day, and as of October 25 had not detected further successful unauthorized access on the customer platform. Data categories named: resource metadata; billing/shipping addresses, invoice and payment history, account balance, and credit usage (payment card numbers are not stored and were not impacted); and for some customers a subset of incomplete model weights and datasets. Fireworks attributes the activity to a financially motivated group aimed at selling LLM access tokens (“LLMJacking”), citing a Nov 2 Google TAG/Mandiant post, and says Inference, Training SDK, and Nexus (ZDR by default) showed no detected impact. Unit 42 retained; self-serve dataset upload/download disabled during the investigation. Soft-title: AI-infra provider Security Watch, not a model-lab safety disclosure.

Why it matters

First-party widening of an AI inference/training-platform breach to billing data and incomplete weights — the qualifying Security Watch item this overnight.

Official Prime Minister of Australia media release (Tuesday 10 November 2026, Albanese, Marles, Burke, Ayres) says the government will enshrine mandatory AI safety and security protections in Australian AI Standards legislation, applying to all frontier AI companies that wish to operate in Australia. A new targeted AI safety regulator, working with the Australian AI Safety Institute, would enforce the rules and could impose “significant penalties.” Six proposed company obligations: (1) security controls across development and deployment; (2) identify and manage serious model risks; (3) allow independent evaluation; (4) mandatory rapid incident notification to the Australian Government; (5) maintain and publish safety/security frameworks; (6) respond to updated standards as technology changes. The release says the measures are aligned with interim recommendations of the Joint Select Committee on Artificial Intelligence and would help provide earlier warning of risks from “rogue AI agents.” Soft-title: proposed legislation and a planned regulator, not an already-enacted statute and not a completed enforcement action.

Why it matters

Same-window official Expand A move: mandatory frontier-lab obligations plus a new regulator, with operational incident-reporting bite once legislated.

Google Keyword and page metadata dated November 10, 2026 (Thomas Kurian); the Cloud blog’s visible byline also shows November 11. Google announced Gemini for healthcare, a purpose-built industry solution extending the universal Gemini agent (introduced last month at Gemini at Work), available in preview in the U.S. today. Working across clinical, financial, and administrative systems, it orchestrates multi-step tasks via skills, connectors, a partner ecosystem, and a governed platform with model choice; built to support HIPAA compliance. Named connector examples include Epic and Oracle Health EHRs as read-only, with no write-back to the chart. Named organisations on the Cloud blog include HCA Healthcare, Northwell Health, Blue Cross Blue Shield of Massachusetts, Optum, and ASCO. Vendor: clinicians spend “nearly 28 hours a week” reconciling information — Google’s figure. Soft-title: vertical enterprise agent preview, not a new frontier model and not a clinician-autonomous diagnosis ship.

Why it matters

Same-window first-party vertical agent product with EHR connectors and a U.S. preview gate.

Official GitHub release v2026.11.3 (847611d, github-actions) published 10 Nov 18:53 UTC = Wed 2026-11-11 02:53 HKT, after already-LIVE v2026.11.2. Release notes state “No new configuration options or features.” Highlights are reliability and boundary fixes: private subagent completions kept out of streamed chat; parent handoff after cancelling a child; unanswered tool failures reported instead of a silent success; stalled provider authentication stopped at the run deadline; required-sandbox image previews kept inside allowed folders; saved node certificate pins preserved and workspace-transfer certificates verified before credentials/data move; Gateway startup/responsiveness fixes. Fixes also cover credential isolation across plugins, search/memory key scoping, credential redaction, child-process env isolation, and safer update recovery/rollback. npm package openclaw@2026.11.3 is listed. A same-day v2026.9.37 extended-stable tag is a maintenance-line backport, not a second current-line card (its own notes still pointed at 11.2 as latest when written). Soft-title: security/reliability patch, not Multiplayer or an 11.2 recard. The openclaw.ai blog still tops at the Nov 6 model-catalog post; Multiplayer is Nov 5.

Why it matters

Standing OpenClaw lane has a new stable tag overnight; material as a current-line ship, but explicitly not a feature release.

Markets 5 charts
US majors (Tue) 2026-11-10T16:00:00-05:00 · US
Asia indexes (Tue) 2026-11-10T16:10:00+08:00 · mixed
Korea memory (Tue) 2026-11-10T15:30:00+09:00 · KR
Oil (Tue settles) 2026-11-10T14:30:00-05:00 · US
US Treasury (Tue) 2026-11-10T16:00:00-05:00 · US
Finance 8 items

Bureau of Labor Statistics (10 Nov, 8:30 a.m. ET): the CPI-U increased 0.4% seasonally adjusted in October after +0.6% in September. Core CPI (all items less food and energy) rose 0.3%, the same as in September and August. Shelter rose 0.3% and accounted for over 30% of the monthly all-items increase; food +0.5% (food at home +0.4%, food away from home +0.5%); energy +0.7% after September’s +5.2%; gasoline +1.0% MoM (and +37.4% over 12 months, one of the few official yearly item prints BLS could publish); electricity +0.9%; utility piped gas −0.8%. Used cars and trucks −1.2%; new vehicles +0.2%; transportation services +0.8%; medical care commodities +0.5%; apparel +0.1%. Not seasonally adjusted, the CPI-U index level was 337.511 (+0.2% on the month). BLS did not publish 12-month percent changes for most series, including headline and core CPI, because it could not collect October 2025 survey data during last year’s lapse in appropriations. CNBC: economists surveyed by Dow Jones had expected +0.3% headline and +0.2% core — poll color, not the BLS lead. Fed Governor Christopher Waller, speaking in London before the release (Fed Board text): even if October improved, he would still support a 25 bp hike at the December meeting; but if core CPI came in “around 0.3% or even higher,” a 50 bp hike “should be considered,” and hiking only 25 bp would “put our credibility at risk.” The 0.3% core print matched that pre-set threshold — after/coincided with, not because Waller caused the number, and not an FOMC decision.

Why it matters

A third straight 0.3% core month, with no official yearly rate, is the live inflation input into December FOMC pricing after Waller’s pre-release 50 bp threshold.

More ▾ Less ▴
Confirmed
What happened

The Bureau of Labor Statistics released the October 2026 CPI at 8:30 a.m. ET on November 10, 2026 (21:30 HKT). The CPI-U increased 0.4% seasonally adjusted after +0.6% in September. Core CPI (all items less food and energy) rose 0.3%, the same as in September and August. Shelter rose 0.3% and accounted for over 30% of the monthly all-items increase; food +0.5% (food at home +0.4%, food away from home +0.5%); energy +0.7% after September’s +5.2%; gasoline +1.0% month-over-month and +37.4% over 12 months, one of the few official yearly item prints BLS could publish. Not seasonally adjusted, the CPI-U index level was 337.511 (+0.2% on the month). BLS did not publish 12-month percent changes for most series, including headline and core CPI, because it could not collect October 2025 survey data during last year’s lapse in appropriations. CNBC: economists surveyed by Dow Jones had expected +0.3% headline and +0.2% core — poll color, not the BLS lead. Fed Governor Christopher Waller, speaking in London before the release (Federal Reserve Board text), said that even if October improved, he would still support a 25 bp hike at the December meeting; but if core CPI came in “around 0.3% or even higher,” a 50 bp hike “should be considered,” and hiking only 25 bp would “put our credibility at risk.” The 0.3% core print matched that pre-set threshold — after/coincided with, not because Waller caused the number, and not an FOMC decision. Soft title: October monthly CPI with no official headline or core yearly rate. Do not treat private 3.9–4.0% reconstructions as a BLS print. Stamp: BLS 8:30 a.m. ET November 10, 2026.

Watch next
  • December FOMC (8–9 Dec)
  • November CPI due 10 Dec, 8:30 a.m. ET
  • Next PCE
  • Any official BLS restatement of withheld yearly rates
Limitation

BLS primary Confirmed for the monthly prints and the withheld yearly-rate explanation. There is no official October headline or core YoY rate; wire or economist reconstructions (some outlets cited 3.9–4.0%) are not a BLS figure and are not used as one. Waller spoke before the release; his 50 bp comment is a conditional preference, not an FOMC decision — after≠because. CNBC’s Dow Jones survey is poll color, not the Confirmed lead. Airline-fares and other item moves not named in the BLS summary text are omitted.

AP (10 Nov): U.S. stocks dipped after the inflation report came in worse than economists expected; the S&P 500 slipped 0.4% and gave back some of Monday’s already-Wired +2.3% rally; the Dow dropped 157 points, or 0.3%; the Nasdaq sank 0.7%. Exact closes as printed by TradingKey (cross-checked with RTT/Finanznachrichten): S&P 7,520.93 (−29.01 / −0.38%); Dow 52,231.89 (−157.86 / −0.30%); Nasdaq 25,065.87 (−187.26 / −0.74%). Treasury par curve (10 Nov CSV): 10y 4.98% (from Monday 4.88%), 2y 4.73% (from 4.56%), 30y 5.35% (from 5.26%). AP narrative printed the 10-year at 4.99%; use the official 4.98% close. AP: the single heaviest weight on the S&P was Nvidia, which fell 5.5% — Finance notes only the tape weight; the Huang/board investigation remains AI-desk owned. AP also: interest-rate-sensitive homebuilders fell, including Lennar −5.4%, D.R. Horton −3.8%, and PulteGroup −3.9%. The session coincided with the hotter-than-expected CPI print and the jump in yields — after/coincided with, not because of a confirmed December hike. Distinct from Tue LIVE Monday best-day-since-April rebound.

Why it matters

Tuesday’s pullback and a 10 bp jump in the 10-year are the live U.S. starting tape for Wednesday Asia after Monday’s already-Wired rebound.

Anthropic (10 Nov company announcement): the lab launched the roadshow for its proposed IPO of Class A common stock. The proposed offering is 825,000,000 shares — 801,351,354 offered by Anthropic and 23,648,646 by certain existing stockholders; the company will not receive proceeds from the selling-stockholder shares. It also expects to grant underwriters a 30-day option to buy up to an additional 123,750,000 shares. The IPO price is expected to be between $70.00 and $78.00 per share. Anthropic has applied to list on the Nasdaq Global Select Market under the symbol “ANT.” Copies of the preliminary prospectus are available from joint book-runners Goldman Sachs and Morgan Stanley. A registration statement has been filed with the SEC but has not yet become effective; the announcement is not an offer to sell, and ANT is not already trading. CNBC, citing the Tuesday amended filing: Anthropic has agreed that AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC will purchase $26 billion of Class N common stock in a concurrent private placement at the IPO price; those shares will not be registered in the offering and will not be subject to underwriter lock-ups. CNBC also calculated that, at the top of the range, the 825 million-share offering would total about $64.35 billion and the company could be worth up to about $1.405 trillion on a fully diluted basis — wire math, not a public trading price. Never attach a public share price to this still-unlisted name. CNBC, citing anonymous people familiar: the deal is planning to price 18 Nov and trade 19 Nov, with about 20% retail allocation — not in the company announcement, so not the Confirmed lead. Distinct from already-wired earlier S-1 / listing-aim color; this is the first official price range and roadshow launch.

Why it matters

A named $70–$78 range and formal roadshow are the window’s qualify IPO event — offer terms, not a traded share price — ahead of a potential Nasdaq debut under ANT.

More ▾ Less ▴
Confirmed
What happened

Anthropic, in a November 10, 2026 company announcement, launched the roadshow for its proposed IPO of Class A common stock. The proposed offering is 825,000,000 shares — 801,351,354 offered by Anthropic and 23,648,646 by certain existing stockholders; the company will not receive proceeds from the selling-stockholder shares. It also expects to grant underwriters a 30-day option to buy up to an additional 123,750,000 shares. The IPO price is expected to be between $70.00 and $78.00 per share. Anthropic has applied to list on the Nasdaq Global Select Market under the symbol “ANT.” Copies of the preliminary prospectus are available from joint book-runners Goldman Sachs and Morgan Stanley. A registration statement has been filed with the SEC but has not yet become effective; the announcement is not an offer to sell, and ANT is not already trading. CNBC, citing the Tuesday amended filing, reported that Anthropic has agreed that AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC will purchase $26 billion of Class N common stock in a concurrent private placement at the IPO price; those shares will not be registered in the offering and will not be subject to underwriter lock-ups. CNBC also calculated that, at the top of the range, the 825 million-share offering would total about $64.35 billion and the company could be worth up to about $1.405 trillion on a fully diluted basis — wire math, not a public trading price. Never attach a public share price to this still-unlisted name. CNBC, citing anonymous people familiar, said the deal is planning to price 18 Nov and trade 19 Nov, with about 20% retail allocation — not in the company announcement, so not the Confirmed lead. Soft title: official expected offer range and roadshow launch, distinct from the already-wired October 26 S-1. Stamp: Anthropic announcement November 10, 2026.

Watch next
  • Effective registration / priced deal / first trade if and when confirmed on the calendar
  • Any further S-1 amendment
  • Whether the concurrent private placement appears in company-primary terms
  • Confirmation of any pricing/trading timetable from the company or the exchange
Limitation

The $70–$78 expected offer range, 825 million-share split, greenshoe, ANT application, and GS/MS role are company-primary Confirmed. The $26 billion private placement and CNBC’s $64.35 billion / $1.405 trillion figures were not independently re-read from the S-1/A in this close and stay CNBC-attributed. Pricing 18 Nov / trading 19 Nov and the 20% retail figure are anonymous-source CNBC, not the company statement. The registration statement is not yet effective; size and timing can still change. Do not describe ANT as already trading, and do not attach a public or traded share price to unlisted Anthropic. This is not a recard of the October 26 S-1.

General Administration of Customs English USD table (10 Nov; unit USD 100 million): China’s October exports were US$412.33 billion (+35.2% YoY, −7.8% MoM); imports US$307.05 billion (+41.6% YoY, −2.2% MoM); the October surplus was US$105.28 billion. January–October exports US$3,782.76 billion (+22.7% YoY); imports US$2,740.17 billion (+29.0% YoY); the goods surplus reached US$1,042.59 billion. January–October total trade was US$6,522.93 billion (+25.3% YoY). RTHK and CNBC, citing GAC: the surplus topped US$1 trillion for a second straight year and was the fastest to that mark; CNBC said October exports beat a Reuters-poll +33.5% forecast and imports beat +32%, and called the import jump the strongest since May 2021 — poll/comparison color, not the GAC table lead. RTHK, citing GAC: October exports to the United States rose 32.9% YoY. The USD “20%” Jan–Oct growth figure in some CNY-based wrap-ups is the GAC CNY table (+20.0%), not the USD +25.3% print — do not mix the two. Tape: RTHK had the Shanghai Composite +0.08% to 3,663 the same day — after/coincided with, not because.

Why it matters

Crossing US$1.04 trillion in 10 months, with imports also accelerating, is the live China external-balance print ahead of next week’s APEC leaders’ meetings.

Yonhap (10 Nov): Kospi closed 5,239.13 (+4.39% / +220.28), snapping three down sessions, after a high of 5,300.92 and a buy-side sidecar at 9:06 a.m.; winners beat losers 714 to 174; foreigners net bought 365.9 billion won and institutions 489 billion won, while individuals net sold 1.3 trillion won; the won was 1,337.5 per dollar, up 9.6 won. Korea Times (Yonhap-captioned prices): Samsung Electronics ₩198,900 (+4.74%); SK hynix ₩1,212,000 (+6.69%); Kosdaq 667.61 (+4.48%). Nikkei Indexes: Nikkei 225 60,459.58 (+2.87% / +1,684.08), back above 60,000 on a closing basis; Technology contributed +991.37 yen; advancers 161 / decliners 64. RTHK: Topix 3,893 (+1.98% / +75); HSI 24,557 (+0.4% / +108) on HK$178.17 billion turnover; HSTECH +1.1% to 4,084; HSCEI +0.1% to 8,265; Shanghai Composite 3,663 (+0.08%); Shenzhen Component +0.86% to 11,241. RTHK color: mainland chip IPOs Suzhou Denglin Technology closed HK$213 (−7.79%) and Jiangsu Xinde HK$11.40 (+28%) after raising a combined HK$2.36 billion — debut color, not a standalone. The Asia bounce traded off Monday’s already-Wired Wall Street rebound and lower oil — after/coincided with, not because of Tuesday’s U.S. CPI, which printed after the Asia close. Distinct from Tue LIVE Monday Asia split (Nikkei bounce / Kospi still down). Yonhap’s Nvidia-probe session color is AI-desk owned; Finance uses it only as named tape context.

Why it matters

Seoul’s 4.4% rebound and Tokyo’s return through 60,000, led by memory and other tech, are the live Asia cash follow-through after last week’s already-Wired washout.

CNBC (10 Nov): Brent settled $91.49 a barrel (−1.17%); WTI settled $86.04 (−0.93%). Brent’s session low was $89.16, briefly below $90 for the first time since August. Dow Jones Market Data (via Morningstar): front-month ICE Brent for January delivery lost $1.08, or 1.17%, to $91.49; lowest settlement since 31 Aug 2026; down two sessions and −$6.83 / −6.95% over those two days. AP also rounded Brent’s drop to 1.2% at $91.49. CNBC session narrative: prices fell after Trump said the Iran war could end “fairly soon” and, following a White House meeting with Qatar’s prime minister, that “Iran militarily is effectively over” and Tehran was “talking very politely” — color on already-Wired diplomacy, not a third Iran card and not a signed deal. Session coincided with those comments and with recovering regional flows in analyst notes — after/coincided with, not because of a confirmed Hormuz or nuclear agreement. Distinct from Tue LIVE Monday settles of $92.57 / $86.85.

Why it matters

A second-day drop to the lowest settlement since August, including a brief break of $90, extends last week’s already-Wired crude unwind into Tuesday’s close.

Federal Reserve Bank of New York (9 Nov): the October Survey of Consumer Expectations, fielded 1–31 Oct, showed median inflation expectations unchanged at the one-, three-, and five-year horizons at 3.9%, 3.3%, and 3.0%. Mean unemployment expectations — the mean probability that the U.S. unemployment rate will be higher one year from now — increased 1.4 percentage points to 45.3%, the highest since April 2020, driven by respondents without a college degree. The mean perceived probability of job loss in the next 12 months rose 0.4 point to 13.9%; the mean probability of finding a job if the current one were lost rose 0.4 point to 46.5%, the highest since November 2025. Median year-ahead expected price changes fell 0.6 point to 4.2% for gas, were unchanged at 5.5% for food, and rose to 9.3% for medical care, 7.6% for college, and 7.7% for rent (+0.9 point). Median home-price growth expectations were unchanged at 3.0%. Median expected household-income growth eased to 3.0%; spending-growth expectations fell to 5.3%. The average perceived probability of missing a minimum debt payment over the next three months rose to 13.8%, the highest since December 2025. Released Monday after Tuesday’s wire close, so it is new for this book; it is an expectations survey, distinct from Tuesday’s BLS CPI print.

Why it matters

Stable 3.9% short-term inflation expectations beside the highest unemployment-rise reading since April 2020 is the official household-outlook companion to Tuesday’s CPI.

Eli Lilly (9 Nov press release): ATTAIN-OSA, a Phase 3 program presented at the American Heart Association Scientific Sessions 2026 and published in JAMA, met its primary endpoint in adults with moderate-to-severe obstructive sleep apnea and obesity or overweight. In participants not using PAP therapy, Foundayo (orforglipron) reduced AHI events by 21.1 per hour, or 43.8%, at 52 weeks from a baseline of 50.7 events per hour; 31.5% achieved remission or mild disease. In participants using PAP, AHI fell by 18.3 events per hour, or 27.5%, from a baseline of 57.6; 19.6% achieved remission or mild disease. Sleep apnea-specific hypoxic burden fell 55.9% without PAP and 45.8% with PAP. Mean weight loss was 28.0 lbs (11.5%) without PAP and 28.2 lbs (11.2%) with PAP (efficacy estimand). The safety profile was generally consistent with earlier ACHIEVE and ATTAIN trials; GI events were the most common adverse events, and discontinuations due to adverse events were 12.2% vs 3.8% placebo (no PAP) and 6.8% vs 0.6% (with PAP). Lilly plans to submit Foundayo to the FDA for moderate-to-severe OSA in adults with obesity or overweight by the end of 2026. Foundayo is already FDA-approved for adult obesity; it is not yet approved for OSA. Figures are the company’s efficacy estimand.

Why it matters

A named Phase 3 OSA win and a year-end FDA-submission plan extend Lilly’s oral GLP-1 franchise beyond the already-approved obesity indication.