Zynergy Wire
Thursday, 12 November 2026 · Private edition
粵語 0:00
Archive · Thursday, 12 November 2026 Back to Today All dates

Archive

Thursday, 12 November 2026

AI 8 items

Official Claude product pages dated November 10, 2026, say: “Today we’re launching Claude Whiteboards and Claude Sheets in beta, and Claude Dashboards becomes generally available as Claude Data… These are available on all paid plans.” Whiteboards is a shared board for real-time brainstorming with Claude; Sheets lets Claude build and edit spreadsheets (tabs, formulas, formatting; export to .xlsx or CSV). Claude Data adds certified queries and live sharing (both in beta), data stories / narrated videos, design systems, caching, and tracked changes. Claude Meetings is coming soon on the same first-party pages — not a launched beta; the product docs beat TechCrunch’s “launching in beta / Google Meet” claim. On Team plans all three are on by default; on Enterprise, Whiteboards and Sheets are off by default, and Data turns on by default on November 19, 2026. Interactive connectors, which let partners embed apps in Claude, shipped in the same cluster and are not a second card. This is a workplace-artifacts ship (Whiteboards and Sheets in beta, Data generally available), not a Claude Meetings launch and not a recut of Fable or Mythos. Anthropic’s newsroom had no office-tools post as of this close.

Why it matters

First-party workplace-artifact ship missed at Wednesday close — the top platforms/agents item for this book.

Anthropic / Claude (first-party product pages + product-updates) — vendor-confirmed; newsroom silent
More ▾ Less ▴
Confirmed
What happened

Anthropic’s official Claude product pages, dated November 10, 2026, launched Claude Whiteboards and Claude Sheets in beta and made Claude Dashboards generally available as Claude Data, available on all paid plans. Whiteboards is a shared board for real-time brainstorming with Claude. Sheets lets Claude build and edit spreadsheets — including tabs, formulas, and formatting — and export to .xlsx or CSV. Claude Data adds certified queries and live sharing, both themselves in beta, plus data stories and narrated videos, design systems, caching, and tracked changes. Anthropic’s newsroom still had no office-tools post as of this close; the Confirmed card is from the product pages. Claude Meetings is listed as coming soon on the same first-party pages, not as a launched beta. Product documentation is the lead over TechCrunch’s “launching in beta / Google Meet” wording. Interactive connectors, which let partners embed apps in Claude, shipped in the same cluster and are not a second card. Admin defaults: on Team plans, Whiteboards, Sheets, and Data are on by default. On Enterprise, Whiteboards and Sheets are off by default, and Data turns on by default on November 19, 2026. This is a workplace-artifacts ship, not a Claude Meetings launch and not a recut of Fable or Mythos. Stamp: Claude product-updates dated November 10, 2026; TechCrunch 9:30 a.m. PST November 10 ≈ 01:30 HKT November 11, inside the Wednesday window and missed at that close.

Watch next
  • A first-party Meetings ship page, or a newsroom write-through
  • Enterprise Data default-on on November 19, 2026
  • Any published HIPAA/CMEK/ZDR support claim for these tools
  • Enterprise admin-default uptake after the November 19 Data change
Limitation

Confirmed from Anthropic/Claude first-party product pages dated November 10, 2026; the newsroom was silent. Beta limits on Sheets and Whiteboards, and beta status for certified queries and live sharing, are as stated on those pages. HIPAA/CMEK/ZDR support was not claimed as ready on the fetched pages. Meetings remains coming soon; TechCrunch’s beta/Meet wording is secondary and is not the lead. Vendor-confirmed ≠ independently reproduced.

Official Grok Bot changelog (Latest v1.0.1 · November 11, 2026; the same entry also includes 1.0.0), titled “Meetings, Microsoft Teams, and shared connectors.” This card is from that changelog, not from @Bot or Musk posts. On Team, Enterprise, and Ultra plans, a Bot can join a Google Meet, Zoom, or Microsoft Teams call from a link, answer in the meeting chat, and send a summary afterward. Team Bots can join Microsoft Teams once a team admin turns it on, an owner or Manager connects the Bot, and a Teams admin approves it. Owners and Managers can share their own app login as a shared connector after a team admin turns on Allow Shared Logins. What teammates teach a Team Bot waits under Review learnings until the owner or a Manager approves it; team admins can limit who shares Bots. The Voice menu adds a Speed slider from 0.5x to 2x. An expired order card offers Check price again. This is an official Bot capability ship — meeting join, Teams, and shared connectors — not a new Grok frontier model and not a recut of the November 2 “Introducing Grok Bot 1.0” news post. It is also distinct from Claude Meetings, which remains coming soon.

Why it matters

Official release note for a material Bot capability cluster — meeting join, Teams, and shared logins — the standing Grok Bot lane’s first Confirmed ship in this window.

More ▾ Less ▴
Confirmed
What happened

The official Grok Bot changelog, dated November 11, 2026 and marked Latest v1.0.1, also includes 1.0.0 in the same entry, titled “Meetings, Microsoft Teams, and shared connectors.” The Confirmed card is from https://x.ai/changelog/bot, not from @Bot or Musk posts. On Team, Enterprise, and Ultra plans, a Bot can join a Google Meet, Zoom, or Microsoft Teams call from a link, answer in the meeting chat, and send a summary afterward. Team Bots can join Microsoft Teams once a team admin turns the integration on, an owner or Manager connects the Bot, and a Teams admin approves it. Owners and Managers can share their own app login as a shared connector after a team admin turns on Allow Shared Logins. Learnings that teammates teach a Team Bot wait under Review learnings until the owner or a Manager approves them; team admins can limit who shares Bots. The Voice menu adds a Speed slider from 0.5x to 2x, and an expired order card offers Check price again. This is an official Bot capability ship, not a new Grok frontier model and not a recut of the November 2 “Introducing Grok Bot 1.0” news post. It is distinct from Claude Meetings, which remains coming soon. Stamp: November 11, 2026 changelog dateline; no clock time on the fetched page.

Watch next
  • Official help or docs on meeting-join limits and plan gates
  • Any later changelog split that isolates 1.0.0 from 1.0.1
  • First-party notes on Teams admin approval or shared-connector scope
  • Treat @Bot or Musk capability claims as watch-only unless they appear on the changelog
Limitation

Confirmed from the official Grok Bot changelog (Latest v1.0.1 · November 11, 2026; the same entry also includes 1.0.0). Meeting join is gated to Team, Enterprise, and Ultra; Teams and shared-login features are Team-admin gated. The changelog does not print a clock time or a separate 1.0.0-only URL. This is not sourced from @Bot or Musk posts and is not a recut of the November 2 “Introducing Grok Bot 1.0” news post. Vendor-confirmed ≠ independently reproduced.

Extra sources

Google DeepMind first-party (November 11, 2026, Science; Adam Gayoso, Natasha Latysheva, Pushmeet Kohli, and Žiga Avsec) announced AlphaGenome 2, a next-generation DNA sequence model that predicts how DNA sequences and variants affect specific human cell types and states, rather than broad tissue averages. It is a conditional model (tissue, cell type, and state as input alongside a million-letter DNA sequence) and can generalize to new single-cell experiments without retraining via a cell encoder. Trained on bulk and single-cell human and mouse data, including over 107 million individual cell profiles across 68 tissues, it also predicts DNA methylation. On the TenK10K single-cell map of 28 primary immune cell types, DeepMind reports it identified nearly twice as many causal variants (30% vs. 16%) and directional gene effects (43% vs. 23%) as the original AlphaGenome fine-tuned on single-cell data and other specialized tools. It is available today for non-commercial research via the AlphaGenome API, and in private preview for commercial research. This is a research DNA and cell model with a research API, not a clinical or diagnostic product — DeepMind says it is “research use only” and has “not been validated or approved for clinical or diagnostic purposes.” Antigravity Skills are described as coming soon, not as part of this ship.

Why it matters

Same-morning first-party science-model ship with a public research API — a material model-lane item, not a consumer chatbot launch.

Official ERNIE product update (November 11, 2026) released ERNIE 5.5 Flash Preview, “the first post-trained model built on the ERNIE 5.5 Flash Base we released earlier.” First-party specs: 280B total parameters with 14B active and a 1M-token context, positioned for coding, 3D modeling, knowledge work, and content creation. “The API is now available on the Qianfan platform, and the weights will be open-sourced under Apache 2.0 soon.” The post stresses this is an intermediate post-training checkpoint — “training is not finished,” some capabilities “may still fall short,” and an official Flash release will follow. Vendor demo color in an agent harness includes a long-horizon Unreal Engine 5.8 game-coding task and Blender/CAD work; those demos are not independently scored benches. This is a preview intermediate checkpoint with the API available now and weights later, not a finished official Flash general availability and not already-open weights. No November 25 open-source date appears on the fetched first-party page.

Why it matters

Same-morning first-party MoE preview with published active/total parameter counts and a stated Apache-2.0 weights path — the open/reproducible watch starts when weights actually land.

Runway first-party news (November 11, 2026) says Praxis-1 “is now available to everyone with open weights under the Apache 2.0 license.” The September research post was a partner-gated announce (“will release it publicly in the coming months”); today’s post is the public ship. Vendor: a 27B-parameter mixture-of-experts diffusion transformer that predicts video and actions together. On NVIDIA’s RoboLab-120 leaderboard, Runway reports 46.2% success across 1,200 episodes, versus 42.9% for FLUX 3 Action, 36.8% for Cosmos3 Nano, and 28.0% for π0.5 (data as of November 2026). Weights, inference, and fine-tuning code (full and LoRA), including commercial use, are released under Apache 2.0; the post says the model is on GitHub, Hugging Face (native LeRobot support), and vLLM-Omni, with hosted inference on Reactor, and runs from H100/B200 to RTX 5090 / RTX PRO 6000 to Jetson Thor. Launch partners already running it on their own hardware are named as Noble Machines, Standard Bots, and Ultra. This is an open-weight world-action / policy model public release, not a humanoid hardware first-ship, and is distinct from the already-Wired Nvidia GR00T 2 / Isaac Lab 3.0 and from Gemini Robotics 2.5. No Hugging Face or GitHub slug is printed on the fetched news page, and none is invented here.

Why it matters

Same-morning open-weight robotics policy ship with a published RoboLab-120 comparison — the open/reproducible + robotics lane, not a GR00T recut.

MiniMax Research first-party (2026-11-11) released MiniMax H3-Image, an open-weight image generation model built on H3, supporting text-to-image, reference-based generation, and image editing. Vendor: it can generate a 4K (4096 × 4096) image in under 10 seconds, with support for up to 16 reference images. Weights are on Hugging Face (MiniMaxAI/MiniMax-H3-Image); the model is also on the MiniMax API, MiniMax Code, and Hailuo AI. Pricing and “50–70% of comparable mainstream offerings” / “5× the speed” claims are vendor-stated and should be read as such; launch-week 50% off is a promotion, not a lasting list price. This is an open-weight image-generation model, not a new frontier text or agent model. A parameter count is not printed on the fetched blog and is not invented here.

Why it matters

Same-window open-weight 4K image model with a public weight drop — qualifies on the open/reproducible lane, behind the larger model/platform ships.

OpenAI first-party Safety post dated November 9, 2026, on the news listing (“Control Room: A new way to track and stop AI workloads”; article at /index/control-room — an introducing-control-room slug returns 404). Control Room is “a central system to track model workloads across our research infrastructure and stop them quickly when safety or security concerns arise.” It “works together with our misalignment monitoring systems, and halts model output within 35 seconds of a decision to intervene to stop or pause a job.” The 35-second figure is the maximum time between Control Room receiving a stop instruction and further progress no longer being possible, from an authorization cache; OpenAI says it is working to drive the number to zero. It is already deployed on research infrastructure and is not a ChatGPT or API production kill-switch (the post distinguishes the external model-serving stack). Fail-closed: if Control Room goes down or is disabled, including by tampering, “all inference requests will be denied” until it resumes; fail-closed is rolled out now to all frontier model research workloads and will finish across remaining research workloads “over the next few days.” Every research workload that does agentic inference must be associated with an approved safety case; engines check Control Room by job_id before fulfilling a request. The September 20 DNS incident is cited only as an example that a monitor caught it — this is not a recut of that incident or of the October 30 misalignment-monitors card, and this post does not say Control Room halted that job in 35 seconds. This is a research-infrastructure control plane, already deployed; 35 seconds is the maximum lag after a stop instruction is received.

Why it matters

A first-party lab-safety operations ship: a fail-closed research-inference halt plane with a stated maximum stop lag.

More ▾ Less ▴
Confirmed
What happened

OpenAI’s Safety post dated November 9, 2026, titled on the news listing “Control Room: A new way to track and stop AI workloads,” is published at https://openai.com/index/control-room (an introducing-control-room slug returns 404). Control Room is “a central system to track model workloads across our research infrastructure and stop them quickly when safety or security concerns arise.” It “works together with our misalignment monitoring systems, and halts model output within 35 seconds of a decision to intervene to stop or pause a job.” The 35-second figure is the maximum time between Control Room receiving a stop instruction and further progress no longer being possible, drawn from an authorization cache; OpenAI says it is working to drive that lag to zero. It is already deployed on research infrastructure and is not a ChatGPT or API production kill-switch — the post distinguishes the external model-serving stack. If Control Room goes down or is disabled, including by tampering, “all inference requests will be denied” until it resumes. That fail-closed behavior is rolled out now to all frontier model research workloads and will finish across remaining research workloads “over the next few days.” Every research workload that does agentic inference must be associated with an approved safety case, and engines check Control Room by job_id before fulfilling a request. The September 20 DNS incident is cited only as an example that a monitor caught it. This card does not recut that incident or the October 30 misalignment-monitors post, and it does not say Control Room halted that job in 35 seconds. Stamp: November 9, 2026 Safety listing; no clock time on the fetched article — inside the Wednesday window and missed at that close.

Watch next
  • Any follow-up on driving the 35-second authorization-cache lag to zero
  • Completion of fail-closed rollout across remaining research workloads
  • Any later official note extending Control Room beyond research infrastructure
  • Keep this distinct from the October 30 monitors post and the September 20 DNS incident
Limitation

OpenAI first-party Safety post Confirmed (November 9, 2026). Deployment status, fail-closed scope, and the 35-second figure are vendor-described; 35 seconds is maximum lag after a stop instruction is received, not detection latency, and is not a claim that Control Room halted the September 20 DNS incident in 35 seconds. This is research infrastructure, not a ChatGPT or API kill-switch. Fail-closed was still finishing on non-frontier research workloads “over the next few days” as of the post. Vendor-confirmed ≠ independently reproduced.

OpenAI Alignment Research Blog (November 10, 2026; Jeffrey G. Wang and co-authors, including Jakub Pachocki as equal advising) published “What affects chain-of-thought monitorability across the model lifecycle?” The post is distinct from the December 18, 2025 “Evaluating chain-of-thought monitorability” paper and from Control Room. It is motivated by observations around GPT-6 Astra (greater no-CoT capability and CoT-monitor evasion under adversarial testing versus prior models) and is not a recut of a GPT-6 product card. Four first-party findings: (1) no-CoT capability is a smooth function of decreasing base-model perplexity, with a weaker link to higher CoT controllability; (2) architecture can steepen no-CoT gains on some planning and serial-computation evaluations at a given perplexity; (3) CoT controllability increases during RL even with no process supervision other than length constraints and final-answer grading (tool supervision adds a slight further increase); (4) CoT controllability can be transmitted to the next generation via SFT on model-generated data, resembling subliminal learning. The authors call the results “sobering” and say that “further progress in pretraining [will] require new approaches to monitorability”; they also warn that the controllability evaluation is a noisy, imperfect proxy and does not always track monitor-evasion success. This is a lab-safety research disclosure on monitorability trends, not a new model ship, not a Control Room recut, and not a GPT-6 product recut.

Why it matters

Same-window lab-safety disclosure that chain-of-thought monitoring, a load-bearing oversight method in OpenAI’s safety cases, is getting harder as pretraining and RL scale.

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

TSMC (10 Nov board resolutions, company IR): the board approved capital appropriations of approximately US$40,926.70 million (about US$40.93 billion) to meet long-term capacity plans, mainly for (1) installation and upgrade of advanced technology capacity; (2) installation and upgrade of advanced packaging, mature and/or specialty technology capacity; (3) fab construction and fab facility systems; and (4) 2027 R&D capital investments, sustaining capital expenditures, and capitalized leased assets. In separate resolutions, the board approved an investment of no more than US$500 million to establish an investment company for strategic investments in the semiconductor industry and related ecosystem; an investment in Anthropic, PBC, at its per-share IPO price in an amount not to exceed US$500 million to acquire Anthropic common shares; and an investment in OpenAI Group PBC in an amount not to exceed US$500 million to acquire OpenAI preferred stock. The three ceilings are distinct: they are board approvals and not-to-exceed limits, not closed purchases, and they are not a public share price for either unlisted name. The same meeting approved the 2026 third-quarter business report and financial statements — consolidated revenue NT$1,494.24 billion, net income NT$763.06 billion, diluted EPS NT$29.43 — and a NT$9.0 per share Q3 cash dividend (record date 17 Mar 2027, ex-dividend 11 Mar 2027 for common shares and for ADSs, payable 8 Apr 2027). The Q3 figures are the formal board approval of results TSMC already reported in mid-October, not a new earnings print; the new matter is the appropriations package, the three investment ceilings, and the dividend timetable. This is distinct from the already-Wired Anthropic $70–$78 / ANT roadshow, and no assumed IPO midpoint or implied valuation is printed as a traded price.

Why it matters

A named US$40.93 billion appropriations print plus three separate US$500 million ceilings is the live listed-foundry capacity and strategic-investment decision for Thursday’s book.

More ▾ Less ▴
Confirmed
What happened

TSMC, in 10 November 2026 board resolutions published on company IR, approved capital appropriations of approximately US$40,926.70 million (about US$40.93 billion) to meet long-term capacity plans, mainly for installation and upgrade of advanced technology capacity; installation and upgrade of advanced packaging, mature and/or specialty technology capacity; fab construction and fab facility systems; and 2027 R&D capital investments, sustaining capital expenditures, and capitalized leased assets. In separate resolutions, the board approved three distinct not-to-exceed ceilings of US$500 million each: an investment company for strategic investments in the semiconductor industry and related ecosystem; an investment in Anthropic, PBC, at its per-share IPO price to acquire Anthropic common shares; and an investment in OpenAI Group PBC to acquire OpenAI preferred stock. These are board approvals and not-to-exceed limits, not closed purchases, and they are not a public share price for either unlisted name. This card is distinct from the already-Wired Anthropic $70–$78 / ANT roadshow, and no assumed IPO midpoint is printed as a traded price. The same meeting formally approved the already-reported 2026 third-quarter business report and financial statements — consolidated revenue NT$1,494.24 billion, net income NT$763.06 billion, diluted EPS NT$29.43 — and a NT$9.0 per share Q3 cash dividend, with record date 17 March 2027, ex-dividend 11 March 2027 for common shares and for ADSs, and payable 8 April 2027. The Q3 figures are formal board approval of mid-October results, not a new earnings print. The new matter is the appropriations package, the three investment ceilings, and the dividend timetable. Stamp: TSMC Board of Directors Meeting Resolutions, 10 November 2026.

Watch next
  • Any closed purchase or further disclosure on the three investment vehicles
  • 2027 capex guidance when TSMC provides it
  • Anthropic pricing on the already-Wired roadshow, without recutting the $70–$78 range
  • Formal payment of the NT$9.0 Q3 dividend on the stated timetable
Limitation

TSMC IR primary Confirmed. Appropriations are spending authority, not a restated 2026 capex budget or a closed cash outlay. The three US$500 million items are not-to-exceed ceilings, not completed purchases, and not a public share price for unlisted Anthropic or OpenAI. Mid-October Q3 results were already reported; this is formal board approval plus the dividend timetable, not a fresh earnings surprise. No $74 midpoint or implied valuation is used as a traded price.

AP (11 Nov): a rebound for stocks in the artificial-intelligence industry lifted the U.S. market; the S&P 500 rose 1.3% and got close to recovering all of last week’s already-Wired sell-off, pulling within 2.6% of its all-time high set last month; the Nasdaq composite rallied 1.6%; the Dow Jones Industrial Average added 254 points, or 0.5%. Exact closes as printed by HK01 (via Longbridge), consistent with Kiplinger’s rounded levels: S&P 7,619.38 (+98.45 / +1.31%); Dow 52,486.49 (+254.60 / +0.49%); Nasdaq 25,466.19 (+400.31 / +1.60%). AP: Micron Technology rallied 11.9% and Broadcom rose 4%. Kiplinger: Nvidia climbed 2.9%; CoreWeave jumped 24.7%, its biggest one-day percentage gain since 3 Jun 2025. The Huang/board investigation remains an AI-desk story; this card notes only the tape weight. Kiplinger also: Costco rose 2.0% after the board approved a $30-per-share special dividend and a $4 billion buyback — color only, not a standalone card. U.S. stocks were open on Veterans Day; the Treasury market was closed, so there is no official Wednesday yield close (last official 10-year remains Tuesday’s 4.98%). The session coincided with CoreWeave’s already-released results and the AI-share rebound — after or coincided with, not because of a confirmed change in December FOMC odds. This is distinct from the already-Wired Tuesday pullback to 7,520.93 / 52,231.89 / 25,065.87.

Why it matters

Wednesday’s 1.3% S&P rebound, with Treasuries shut, is the live U.S. starting tape for Thursday Asia after Tuesday’s already-Wired CPI pullback.

CoreWeave (10 Nov IR, for the quarter ended 30 Sep): third-quarter revenue was US$3,661 million, up from US$1,365 million a year earlier (+168%). Operating income was US$207 million; interest expense, net, US$846 million; net loss US$709 million, or US$1.27 a share, versus a US$110 million / US$0.22 loss a year earlier. Adjusted operating income was US$379 million; adjusted EBITDA US$2,256 million; adjusted net loss US$529 million. Revenue backlog was approximately US$132.5 billion as of 30 Sep. CEO Michael Intrator said the company expects approximately US$9 billion in customer prepayments and deposits in the next three months alone. Outlook: full-year 2026 revenue guidance raised to US$13.0–$13.3 billion; adjusted operating income guidance raised to US$1.275–$1.375 billion, a 26% increase at the midpoint versus prior guidance; capex now US$37–$39 billion, raising the low end of the prior US$35–$39 billion range because infrastructure is expected to be delivered sooner; annualized run-rate revenue guidance raised to US$20–$21 billion to exit 2026; still more than 1.85 gigawatts of active power at year-end. CNBC, versus LSEG consensus: revenue beat US$3.55 billion expected; adjusted loss per share 95 cents versus US$1.16 expected — poll color, not the IR lead. CNBC, citing the statement, said CoreWeave owed almost US$47 billion in debt at quarter-end; CFO Nitin Agrawal, on the call as reported by CNBC, said the company does not intend to issue additional parent-level unsecured debt through 2027 at current yields and projects under US$5 billion in parent-level debt financing through 2027. Wednesday cash: Kiplinger and Reuters wraps had the stock +24.7%. The Huang investigation is an AI-desk story. A Reuters wrap’s “about ten times” demand and first-half-2028 adjusted-profitability comments are not in the IR release fetched for this close and are omitted.

Why it matters

A 168% revenue print, a raised 2026 outlook and a named prepayment figure are the window’s live neocloud test of AI-infrastructure demand — and the catalyst that coincided with Wednesday’s tape rebound.

More ▾ Less ▴
Confirmed
What happened

CoreWeave, in a 10 November 2026 IR release for the quarter ended 30 September, reported third-quarter revenue of US$3,661 million, up from US$1,365 million a year earlier (+168%). Operating income was US$207 million; interest expense, net, US$846 million; net loss US$709 million, or US$1.27 a share, versus a US$110 million / US$0.22 loss a year earlier. Adjusted operating income was US$379 million; adjusted EBITDA US$2,256 million; adjusted net loss US$529 million. Revenue backlog was approximately US$132.5 billion as of 30 September. CEO Michael Intrator said the company expects approximately US$9 billion in customer prepayments and deposits in the next three months alone. Outlook: full-year 2026 revenue guidance was raised to US$13.0–$13.3 billion; adjusted operating income guidance was raised to US$1.275–$1.375 billion, a 26% increase at the midpoint versus prior guidance; capex is now US$37–$39 billion, raising the low end of the prior US$35–$39 billion range because infrastructure is expected to be delivered sooner; annualized run-rate revenue guidance was raised to US$20–$21 billion to exit 2026, with still more than 1.85 gigawatts of active power at year-end. CNBC, versus LSEG consensus, said revenue beat a US$3.55 billion estimate and the adjusted loss was 95 cents a share versus US$1.16 expected — poll color, not the IR lead. CNBC, citing the statement, said CoreWeave owed almost US$47 billion in debt at quarter-end; CFO Nitin Agrawal, on the call as reported by CNBC, said the company does not intend to issue additional parent-level unsecured debt through 2027 at current yields and projects under US$5 billion in parent-level debt financing through 2027. Wednesday cash: Kiplinger and Reuters wraps had the stock +24.7%. A Reuters wrap’s “about ten times” demand and first-half-2028 adjusted-profitability comments were not in the IR release and are omitted. Stamp: CoreWeave IR, 10 November 2026, quarter ended 30 September.

Watch next
  • Nebius earnings scheduled Thursday
  • Any follow-through on financing or 2027 capex
  • Formal 10-Q filing
  • Delivery against the US$20–$21 billion exit ARR and 1.85 GW year-end power guidance
Limitation

Revenue, loss, backlog, guidance, capex, ARR and the US$9 billion prepayment/deposit figure are company-primary Confirmed. LSEG comparisons and the “almost US$47 billion” debt total are CNBC-attributed. The no-unsecured-debt / under-US$5 billion financing comments are CNBC’s report of the call, not the IR text. The Reuters “about ten times” demand and first-half-2028 profitability comments were not in the IR release and are omitted. +24.7% is the Wednesday cash move from Kiplinger/Reuters wraps, not an IR figure. Guidance is outlook, not a reported result. Finance owns this neocloud earnings card.

Yonhap (11 Nov ticker): Kospi closed 5,375.36 (+136.23). MoneyToday, citing Korea Exchange and NextTrade: that is +2.60%; Samsung Electronics ₩206,000 (+3.57% / +7,100); SK hynix ₩1,277,000 (+5.36% / +65,000); Kosdaq 705.10 (+5.62% / +37.49); foreigners net bought 776 billion won and other corporates 534.9 billion won, while retail net sold 893 billion won and institutions net sold 421.5 billion won; the won closed at 1,333.6 per dollar, down 3.9 won; a buy-side sidecar was triggered at 11:28 a.m. Nikkei Indexes: Nikkei 225 61,248.59 (+1.31% / +789.01); Technology contributed +816.33 yen; advancers 140 / decliners 85. RTHK: HSI 24,315 (−0.98% / −241); HSCEI 8,171 (−1.14%); HSTECH −1.41%; Shanghai Composite +0.5%; Shenzhen Component +1.32%; ChiNext +3.1%. The Korea/Japan advance traded after Tuesday’s already-Wired U.S. CPI/yields session and after CoreWeave’s Tuesday-evening results — after or coincided with, not because. Hong Kong closed lower the same day, a split regional tape. This is distinct from the already-Wired Tuesday Asia session (Kospi +4.39% / Nikkei through 60,000 / HSI +0.4%).

Why it matters

A second-day Seoul advance through 5,300, with memory leading and Tokyo through 61,000, is the live Asia cash follow-through — even as Hong Kong slipped.

Dow Jones Market Data via Morningstar (11 Nov): front-month ICE Brent crude for January delivery gained $1.27, or 1.39%, to settle at $92.76; largest one-day dollar and percentage gain since 5 Nov 2026; snaps a two-session losing streak. WTI: Odessa American’s 11 Nov daily oil price listed crude at 86.53 (+0.49); HK01 via Longbridge printed New York crude futures at $86.53 (+$0.49 / +0.57%) and London Brent at $92.76 (+$1.27 / +1.39%), matching the DJ Brent settle. RTHK: oil prices rose Wednesday after falling to two-month lows in the previous session, as the market swung between hopes of de-escalation in the Iran war and concerns over continued attacks on shipping in the Strait of Hormuz — session color only. After or coincided with those mixed signals, not because of a confirmed Hormuz or nuclear agreement. This is distinct from the already-Wired Tuesday settles of $91.49 / $86.04, and is not a recut of Monday’s talks card, Tuesday’s developing minor-strikes card, or Wednesday’s oil card.

Why it matters

A $1.27 bounce that breaks the two-day slide keeps crude in the already-Wired unwind range rather than extending Tuesday’s break of $90.

AP (11 Nov, Abu Dhabi): Russia and Ukraine agreed to a large exchange of prisoners of war, both sides said, as two days of U.S.-mediated talks concluded with no major breakthrough on ending Moscow’s nearly five-year-old full-scale invasion — the first meeting in nine months between the delegations. BBC, confirming the same close: both Moscow and Kyiv said a large-scale prisoner exchange was in the works and could be carried out soon; U.S. negotiators Steve Witkoff and Jared Kushner will travel on Thursday to Moscow to meet Vladimir Putin, then to Dublin to meet Volodymyr Zelenskyy. BBC: Ukraine negotiator Davyd Arakhamia called the round “one of the hardest” but saw “good momentum” and wanted a second round soon; Kremlin negotiator Vladimir Medinsky said a lot depends on Ukraine, the U.S. position, “and on whether Europe comes to its senses.” Kushner, per BBC, said territorial issues remain the hardest and that a peacetime security framework cannot occur until there is agreement on territory and other bilateral issues; Medinsky repeated that “every inch” of land would be “returned to Russia one way or another.” European officials were in Abu Dhabi on the sidelines; the Kremlin said it had not met them. This is a distinct track from already-Wired Iran/Hormuz diplomacy and from Tuesday’s developing minor Iran air-defense strikes, and a POW exchange is not a ceasefire or a peace deal.

Why it matters

The first face-to-face meeting in nine months produced a named POW exchange and a Putin/Zelenskyy follow-up, but no settlement — the live official readout for this track.

Federal Reserve Vice Chair for Supervision Michelle W. Bowman (10 Nov, Council on Foreign Relations; identical remarks later the same day): she supported holding the federal funds rate target range at 3-3/4 to 4 percent at the October FOMC, and said bringing inflation down remains a priority. She said increased investment in and use of AI is boosting economic growth and productivity, and that over time she expects broader AI adoption to raise productivity and put some downward pressure on inflation as lower production costs pass through to consumer prices. She also warned that AI revenues “are not yet sufficient to pay for the expense of maintaining existing and developing new AI models and data centers,” that AI-related debt is on track to exceed US$800 billion this year, and that lease, contract and residual-value obligations have potential payments estimated at more than three times that debt. She said a shortfall in AI-related revenues could bring a sharp correction in AI-firm equities, higher financing costs and a slower buildout. On policy, she said the benefit of additional tightening has to be weighed against costs for interest-rate-sensitive sectors, and that “there is still work to do to bring inflation closer to 2 percent.” She also said “today’s consumer price index suggests that 12-month core PCE inflation likely moved up further to 3.3% in October” — that is Bowman’s inference from the already-Wired CPI, not a Bureau of Economic Analysis print, and it is not an official BLS yearly CPI rate (BLS withheld most October YoY figures). This is not a recut of the October CPI card. Labor-market figures in the speech (including an October unemployment rate of 4.1%) are Bowman-stated and were not independently re-fetched from BLS for this close.

Why it matters

A named Fed vice chair’s on-record view — AI may lift productivity, but financing risks and unfinished inflation work still shape the path into the December FOMC.

Hong Kong Monetary Authority (10 Nov press release): the Northern Metropolis Financial Advisory Taskforce, jointly set up by the HKMA and The Hong Kong Association of Banks, held its third meeting. To strengthen bank financing for Northern Metropolis projects, the HKMA proposes to amend the Banking (Capital) Rules to designate government-related entities — including 11 post-secondary institutions and three park companies — as Domestic Public Sector Entities, a change it said would help lower banks’ capital costs and assist those entities in obtaining bank financing. The HKMA intends to submit the amendments to the Legislative Council in the first quarter of 2027, with a view to their taking effect in the second quarter of the same year. The Education Bureau briefed the taskforce on the Northern Metropolis University Town, and participating banks discussed possible financing solutions such as loans and public-private partnerships. This is a proposed rule change, not yet submitted and not in force.

Why it matters

A named capital-rule proposal with a LegCo timetable is the window’s official Hong Kong financing-policy increment for the Northern Metropolis buildout.