Zynergy Wire
Friday, 13 November 2026 · Private edition
粵語 0:00
Archive · Friday, 13 November 2026 Back to Today All dates

Archive

Friday, 13 November 2026

AI 10 items

Google’s official DeepMind post dated November 12, 2026, introduces Gemini 4 Hydrogen, built for “speed and efficiency for everyday coding and agentic workflows,” and distinct from frontier Gemini 4 Carbon and Gemini 4 Argon, whose official introduction was October 28. Docs model id: gemini-4-hydrogen. Paid API pricing, confirmed on the Gemini Developer API pricing page, is $1.50 per 1M input tokens and $7.50 per 1M output tokens (output includes thinking tokens), flat across the stated 1 million-token context, with implicit cache reads at $0.075 per 1M; free-tier input and output are listed as free of charge. “With Hydrogen, we’re also bringing Gemini 4 to everyone — for all tiers of Google AI Studio, the Gemini API, Gemini for business, the Gemini app, Google Search, and paid Google AI Pro and Ultra subscribers in Google Antigravity.” Vendor benches, not independently reproduced here: Hydrogen “surpasses Argon on the Artificial Analysis Intelligence Index at its default high thinking level (51 vs. 50, and reaching 54 at its highest xhigh thinking setting)”; versus Gemini 3.8 Flash, it “finish[es] tasks 2.5 times faster at its high default setting while more than tripling Terminal-Bench 4.1 scores (57% vs. 18.2%).” Four thinking levels: low, medium, high (default), and a new xhigh tier. Orchestrator in Google Antigravity dynamically switches between models such as Carbon and Hydrogen in one session; the Hydrogen post attributes a near-Carbon task-success comparison — on average using 24% less quota and finishing 19% faster — to Terminal-Bench 4.0, while the Flash comparison uses Terminal-Bench 4.1; those versions are not the same. In the Gemini app, free users get Hydrogen via auto model selection when it is the appropriate model, not a claimed full free-tier picker. This is distinct from Gemini Omni 1.2 Flash.

Why it matters

Same-morning first-party Gemini 4 model ship with published API pricing and a broad consumer and developer rollout.

Google / DeepMind (first-party blog) and Gemini Developer API pricing docs — vendor-confirmed benches; not independently reproduced
More ▾ Less ▴
Confirmed
What happened

Koray Kavukcuoglu’s first-party Google / DeepMind post, dated November 12, 2026, introduced Gemini 4 Hydrogen as an everyday coding and agentic model alongside the already-shipped Gemini 4 Carbon and Gemini 4 Argon frontiers. 9to5Google stamped the post at 10:01 a.m. PT November 12, or about 02:01 HKT Friday, November 13, inside this window. The Gemini Developer API pricing page lists model id gemini-4-hydrogen at $1.50 per 1M input tokens and $7.50 per 1M output tokens, with output including thinking tokens, implicit caching at $0.075 per 1M, and a free tier listed as free of charge. The same post says Hydrogen is rolling out across Google AI Studio, the Gemini API, Gemini for business, the Gemini app, Google Search, and paid Google AI Pro and Ultra subscribers in Google Antigravity. Free Gemini-app users get it through auto model selection when it is the appropriate model, not through a claimed full free-tier picker. Hydrogen has four thinking levels — low, medium, high as the default, and a new xhigh tier. Vendor benchmarks in the post are not independently reproduced here. The Hydrogen post attributes Orchestrator’s near-Carbon task-success comparison, with 24% less quota and 19% faster completion on average, to Terminal-Bench 4.0, and attributes the 57% versus 18.2% comparison with Gemini 3.8 Flash to Terminal-Bench 4.1. Those versions should not be flattened. This is not a recut of Carbon’s October 28 introduction or of Gemini Omni 1.2 Flash.

Watch next
  • A resolved Orchestrator Terminal-Bench citation (4.0 versus 4.1) on a later first-party page
  • Any Antigravity Orchestrator plan expansion beyond the paid Pro and Ultra Hydrogen rollout described here
  • Independent reproductions of the Hydrogen model card’s vendor benches
Limitation

Pricing was confirmed on first-party docs. Benchmark and Orchestrator efficiency figures are Google-reported. Publication stamp for clocking the post into this window is a 9to5Google dateline, not a time printed on the Google blog itself.

An official Nvidia newsroom release dated November 12, 2026, and a Form 8-K (earliest event November 11, 2026; filed November 12, 2026) announce that the Board accepted the resignation of Jensen Huang (8-K legal name: Jen-Hsun Huang) as President, Chief Executive Officer, and a member of the Board, effective immediately. Colette Kress, Executive Vice President and Chief Financial Officer since September 2013 (8-K: Colette M. Kress, age 59), is appointed Interim Chief Executive Officer and will continue as CFO until a permanent successor is named. The Board “is initiating a search for a permanent Chief Executive Officer and will consider internal and external candidates.” The Board also appointed Harvey C. Jones as its first-ever Chair of the Board (8-K: on the Board since 1993; age 73). Jones, quoted in the release, said the Board has “full confidence in Colette and the leadership team, and we expect the company’s operations, customer commitments and product roadmap to continue without interruption.” Kress said, “NVIDIA’s strength has always come from its people, and together with our leadership team, I am committed to building on the extraordinary foundation Jensen created over the last 33 years.” Third-quarter fiscal 2027 results remain scheduled for Tuesday, November 17, at approximately 1:20 p.m. PT, with the earnings call at 2 p.m. PT. The official release and 8-K do not state a reason and do not mention the already-Wired independent probe or related reporting — after, not because. This is a new first-party leadership transition, distinct from the Tuesday probe statements. Share prices stay off this card.

Why it matters

First-party leadership change at the central AI-compute supplier, and a new event distinct from the already-Wired board probe.

More ▾ Less ▴
Confirmed
What happened

Nvidia’s newsroom, dated November 12, 2026, and a Form 8-K whose earliest event is November 11 and whose filing date is November 12, say the Board accepted Jensen Huang’s resignation as President, Chief Executive Officer, and a director, effective immediately. The 8-K uses his legal name, Jen-Hsun Huang. Colette M. Kress, 59, Executive Vice President and Chief Financial Officer since 2013, was appointed interim CEO and will keep the CFO role while the Board searches for a permanent successor, considering internal and external candidates. The Board also appointed Harvey C. Jones, 73, a director since 1993, as its first-ever Chair of the Board. Jones said the Board has “full confidence in Colette and the leadership team” and expects operations, customer commitments, and the product roadmap to continue without interruption. Kress said she is committed to building on “the extraordinary foundation Jensen created over the last 33 years.” Fiscal third-quarter 2027 results remain scheduled for Tuesday, November 17, at about 1:20 p.m. PT, with the earnings call at 2 p.m. PT. Neither the newsroom release nor the 8-K states a reason for the resignation, and neither mentions the already-Wired November 8 board statement or the November 10 Wachtell / Joon H. Kim retention. This card is the first-party transition, not a second probe card, and not a claim that those reports caused the resignation. A CNBC staff-email write-through at 11:59 a.m. EST November 12, or about 00:59 HKT Friday, landed after Thursday’s close. Share prices stay off this card.

Watch next
  • Any later first-party statement on the CEO search or a permanent appointment
  • The November 17 earnings call, including any leadership remarks
  • Keep Powell / affair color on the already-Wired Tuesday probe card, not on this transition card
Limitation

Confirmed from Nvidia’s newsroom and the filed 8-K. Official materials are silent on cause. After is not because. No NVDA tape print is used on this AI card.

An official Anthropic article, “Introducing Claude Meetings,” says Claude can now join video calls as a visible participant, follow the conversation, and start work during the meeting. “Claude Meetings is now in beta.” Teams can ask Claude questions out loud or in chat; by default Claude answers in writing and can use connected tools with permission. It can build docs, whiteboards, or other artifacts live, and it writes the notes and action items. “Claude Meetings is starting to roll out in beta on Team and Enterprise plans. Support for other video conferencing platforms is coming soon.” Support is currently available for Google Meet. On Team plans, inviting Claude is available by default and an admin can turn it off; on Enterprise plans, an admin turns it on for all or part of the organization. While it is in beta, it isn’t available to some organizations, including those with zero data retention (ZDR), HIPAA-ready configuration, or customer-managed encryption keys (CMEK). During the beta, notes and transcripts are free on Team and Enterprise plans; answers in calls are free unless an answer uses a connected tool, which counts toward plan usage. Docs, whiteboards, and other artifacts Claude makes in a call are free through February 10, 2027. Claude adds a separate page for each meeting in claude.ai; by default it stays out of calls when someone from outside the organization is invited. It saves a transcript, summary, and screen descriptions, not audio. This is the increment from Thursday’s already-Wired office-tools card, which still had Meetings as coming soon, and it is distinct from already-Wired Grok Bot 1.0.1 meeting join and from Whiteboards, Sheets, and Data.

Why it matters

Same-morning first-party move of Claude Meetings from coming soon to a Team and Enterprise beta, starting with Google Meet.

An official Google XR post dated November 12, 2026, by Shahram Izadi, vice president and general manager of XR, says, “Today, pre-orders begin for the new Intelligent Eyewear collections, made in collaboration with Gentle Monster, Warby Parker, and Samsung,” with Gemini for hands-free help. Warby Parker styles start at $395; Gentle Monster starts at $449 on the Google post. “Intelligent eyewear will be available to purchase in stores and online on December 10.” Pre-orders are live online and in-store at Gentle Monster and Warby Parker, with hands-on demos at select Gentle Monster, Warby Parker, and Google Store locations. Google says the frames weigh around 41 grams, the battery lasts up to nine hours, a charging case can recharge the frames up to seven times, and the glasses include a 12-megapixel ultra-wide camera, Gemini Guided Vision, and Maps and Translate. “Intelligent eyewear is only available for use with Google accounts from the United States and South Korea” and requires a compatible 18+ Google Account. Samsung’s November 12 Global Newsroom post confirms that pre-orders begin today and that the release in the United States and South Korea is December 10, attributes a Snapdragon AR1 Gen 1 platform, and prints U.S. prices as Warby Parker from $395 and Gentle Monster from $449 to $489, depending on the model. Samsung also says both collections launch in the United States, while Gentle Monster will also be available in Korea — that market split is Samsung’s wording. This is a price and availability lock, not a claim that units are already in general retail this week, and it is distinct from already-Wired Muse Charm and Muse Ember. Frame colorways are omitted.

Why it matters

Same-window first-party personal-AI-hardware lock: pre-orders are open, U.S. starting prices are published, and retail is set for December 10.

The official World Labs product page, listed on the lab’s Research & Insights index as November 12, 2026, under the title “Atlas: Releasing our next generation world model,” presents Atlas as a “next generation world model” and invites users to “Generate and reconstruct worlds and bring creations into your own workflows with our tools and developer APIs.” Tools launching with Atlas are Viewfinder (reconstruct a scene from photos or generated images, explore new viewpoints, and render camera paths), Chisel (controllable scene building and grayboxing), Splats to Mesh, Relight, and a Developer API and integrations, including Blender. Docs quickstart confirms the HTTP API root https://api.atlas.worldlabs.ai/api/v2 and a CLI (@worldlabsai/cli). Billing docs say Atlas Developer API usage is billed in credits, “the current conversion is 1,000 credits per US dollar,” and “the task rate card can change,” so a complete public task price list is not locked here. This is a public tools and developer-API launch, distinct from the September 1 Atlas model introduction and from the already-Wired Frameboy preview. The fetched product page does not say “GA”; an image filename is not a quote. X / @drfeifei was discovery only; the card is from the first-party page.

Why it matters

Same-window first-party world-model tools and developer-API launch, distinct from the September 1 introduction and the already-Wired Frameboy preview.

An official Google DeepMind developer post dated November 11, 2026, by Min Choi and Felix Yu, introduces the Gemini Decisions API, “powered by our new Gemini Decision 1 model (gemini-decision-1),” in the Gemini API and Google AI Studio. It is a closed-set decision endpoint, not a chat model: one shared multimodal input (text, images, audio, video, and PDFs, up to 256K tokens) plus up to 1,000 independent questions; the model “scores your candidate options in a single forward pass, returning probabilities and confidence scores.” Three question types: noul (yes/no), choice (2 to 255 options), and score (ordered rubric). Vendor latency: 150–250 ms end-to-end (median 50–100 ms of model compute) on contexts up to 16K tokens, “remaining nearly flat from 1 to 100 questions per request.” Pricing, confirmed on the same pricing docs as Hydrogen, is $0.10 per 1M input tokens and $0 for output tokens (the model generates no output tokens); shared input is billed once per request. An API no-cost tier is listed for prototyping in Google AI Studio. This is distinct from same-morning Hydrogen and from Perplexity Decider v1.3. The first-party post is dated November 11, inside Thursday’s window and missed at that close. Cloud and Vertex support is “coming soon” on the fetched post; no GA-on-Vertex date is given.

Why it matters

Missed-Thursday first-party API and model ship for low-latency, non-generative routing and policy checks, distinct from Hydrogen.

Perplexity’s docs changelog for November 2026 releases Decider v1.3 and Decider Flash v1.3 as the successors to earlier Decider decision models. Both are decision models — typed probabilistic answers, not chat generation — and the Hugging Face model cards, plus the first-party docs, place them on Hugging Face, OpenRouter, and the Perplexity Decisions API. Docs lock API ids and size/price: pplx-decider-v1.3 (27B, $0.02 per million input tokens) and pplx-decider-flash-v1.3 (4B, $0.01 per million input tokens); output tokens stay free; same prices as v1.2. Both accept text and images; v1.3 adds a new locate question type (span extraction; text only; not with web grounding). Aliases: pplx-decider-latest now selects v1.3, and pplx-decider-flash-latest selects Flash v1.3; v1.2 ids remain available. The Hugging Face API lists perplexity-ai/pplx-decider-v1.3 and perplexity-ai/pplx-decider-flash-v1.3 as public and ungated, with license apache-2.0 and weight files present, even though a line on the fetched README still calls the repository private. This is an open-weight decision-model successor, not a frontier LLM, and is distinct from Google Decision 1.

Why it matters

Same-window open-weight decision-model drop with public API ids, prices, and Hugging Face slugs, kept separate from Google Decision 1.

A first-party SpaceX Updates post dated November 12, 2026, says the company is applying its manufacturing experience “to constructing supercomputers” and is advancing “a major new supercompute facility at Starbase in the Rio Grande Valley in Texas, and a second campus in Cochise County, Arizona. Together, they represent more than $74 billion in planned American investment.” In the Rio Grande Valley, SpaceX “is planning to invest more than $24 billion in new supercomputing infrastructure,” building on more than $15 billion already invested in Starship and related infrastructure at Starbase, and says the facility “will generate its own electricity without drawing power from the public grid.” In Cochise County, it is “pursuing a roughly $50 billion supercompute campus.” The same post says SpaceX has “also recently announced the $100B Terafab advanced chip manufacturing project in Grimes County, Texas” — a pointer to a prior announcement, not a new Terafab filing on this card. The dollar figures are planned or pursued investment, not a reported capital-expenditure spend and not an earnings print, and the page does not give a nameplate compute capacity in gigawatts for these campuses. This is distinct from already-Wired SpaceX Super Intelligence quarterly results and from the Treasury SI dialogue on this edition.

Why it matters

First-party AI-infrastructure plan with a published combined $74 billion planned investment at two U.S. supercompute campuses.

First-party Treasury prepared remarks, page dated November 11, 2026, marked “As prepared for delivery,” for Secretary Scott Bessent’s speech at the National Committee on U.S.-China Relations Gala. The Treasury page does not print a gala calendar date. Quote Treasury: the United States is “establishing a new dialogue with China on SI, designed to guard against shared risks and prevent a misunderstanding between our two systems”; expected exchange of “best practices for evaluating and testing advanced models before their release, and for keeping human judgment firmly in command of this technology”; “a direct channel to ensure that either nation can promptly alert the other to serious incidents involving its SI models”; and “we should harbor no illusions about the differences between the United States and China.” The remarks use “superintelligence,” then “SI” for the dialogue, which is distinct from SpaceX Super Intelligence. Lutnick / Megaspeed claims stay off this card. “Harbor no illusions” is not a commitment to limit AI advancement. Shenzhen and “before a fourth encounter in Miami” are in the same remarks; no fourth-summit date is given. This is a US–China SI dialogue and incident-alert channel, as prepared, not a binding treaty and not an export-control action. Missed at Thursday close (posted 11 November, after Wednesday ingest).

Why it matters

First-party U.S.–China policy ship: a named bilateral SI dialogue and incident-alert channel, missed at Thursday close.

On-record Meta confirmation to TechCrunch (12:12 p.m. PST, November 11, 2026): “Santosh Janardhan, the company’s longtime head of infrastructure and co-head of engineering, will be stepping down, Meta told TechCrunch. He will be replaced by Dave Brown, who joined the company as vice president of infrastructure earlier this year, the company added.” A Meta representative, verbatim: “Santosh will stay on for transition and he is moving into an advisory role on Meta Compute.” “He is expected to leave next year.” No Meta newsroom or IR release was found this close. CNBC, citing an internal memo, uses “former AWS senior vice president David Brown, who joined Meta in July from AWS,” and says Janardhan “will help support Brown through the first quarter of 2027” — attribute the AWS SVP title and Q1 2027 support window to CNBC, not to Meta’s on-record TechCrunch comments. Meta’s on-record line is “advisory role on Meta Compute,” not a certified “CTO of Meta Compute” title. This is an on-record infra-leadership succession, not an immediate departure, not a first-party newsroom post, and not a capex recut. Published inside the Thursday close window and missed at close.

Why it matters

On-record hyperscaler personnel change at Meta infrastructure / Meta Compute, missed at Thursday close.

Markets 5 charts
US majors (Thu) 2026-11-12T16:00:00-05:00 · US
Asia indexes (Thu) 2026-11-12T16:10:00+08:00 · mixed
Korea memory (Thu) 2026-11-12T15:30:00+09:00 · KR
Oil (Thu settles) 2026-11-12T14:30:00-05:00 · US
US Treasury (Thu close) 2026-11-12T16:00:00-05:00 · US
Finance 8 items

Dow Jones Market Data via Morningstar (12 November) said front-month ICE Brent crude for January delivery gained $3.06, or 3.30%, to settle at $95.82; the largest one-day dollar and percentage gain since 5 November 2026; and up for two sessions, +$4.33 or +4.73% over those two days, all as DJ-attributed. Reuters via Yahoo: U.S. WTI closed at $88.74, up $2.21 or 2.6%, matching the same Brent settle; both benchmarks were still down about 2.5% for the week. Reuters: prices jumped after attacks on Saudi energy infrastructure and after shipping disruptions through the Strait of Hormuz; Yemen’s Iran-aligned Houthis said they had attacked a pumping station on the Saudi East-West oil pipeline, an oil facility in the Red Sea city of Yanbu, and an Aramco building in Riyadh; the Saudi-led coalition said it had intercepted and destroyed two drones, one in the Riyadh region and another in the Eastern Province. Iran’s military said four oil tankers were hit by explosions in the Strait of Hormuz after ignoring warnings; Reuters could not immediately verify that account. Trade was volatile, with crude selling off early and then rebounding after the Houthi claims. Reuters also noted a Trump social-media post that Washington did not need a deal with Iran — session color only; after or coincided with the attacks and that comment, not because of a confirmed new Hormuz closure or a signed deal. This print is distinct from already-Wired Wednesday settles of $92.76 / $86.53, and from earlier talks and strikes cards.

Why it matters

A $3.06 jump that takes Brent back through $95, after already-Wired unwinds, is the live crude print — and the session coincided with new Saudi-site attacks and a U.S. no-deal comment.

AP how-indexes (12 November): the S&P 500 fell 26.98 points, or 0.4%, to 7,592.40; the Dow Jones Industrial Average fell 189.40 points, or 0.4%, to 52,297.09; the Nasdaq composite fell 100.10 points, or 0.4%, to 25,366.09; the Russell 2000 fell 47.20 points, or 1.7%, to 2,799.06. Exact percentage moves on those same levels, also printed on the Reuters-via-Yahoo wrap, are −0.35%, −0.36%, −0.39%, and −1.66%. Official Treasury CSV: the 10-year par yield closed at 5.10%, the 2-year at 4.86%, and the 30-year at 5.43%, versus Tuesday’s last official close of 4.98%, 4.73%, and 5.35% (Treasuries were shut Wednesday for Veterans Day). Yahoo Finance regular-session closes: Nvidia $166.19 (−4.34%), after opening at $151.01 and trading as low as $148.80; AMD $585.30 (+5.86%); Broadcom $364.36 (+5.54%); CoreWeave $56.60 (−6.86%). The Huang leadership story remains AI-desk owned; Finance notes only the tape weight. The session coincided with the oil jump and with Nvidia’s swing — after or coincided with, not because of a confirmed change in December FOMC odds. This session is distinct from already-Wired Wednesday closes of 7,619.38 / 52,486.49 / 25,466.19. Thursday’s −6.86% is cash follow-through on already-Wired CoreWeave Q3 results, not a new earnings print.

Why it matters

A modest index dip with the official 10-year back at 5.10%, and a sharp Nvidia swing that rivals offset, is the live U.S. starting tape for Friday Asia.

SoftBank Group’s 10 November IR consolidated financial report: for the three months ended 30 September, net sales were ¥2,215,446 million; net loss attributable to owners of the parent was ¥1,115,337 million, or ¥196.54 a share, versus net income of ¥2,502,247 million a year earlier. The figure used in the basic-EPS calculation was a loss of ¥1,120,111 million after ¥4,774 million not attributable to ordinary shareholders — the ¥1.12 trillion print used in wires. For the six months ended 30 September, net loss attributable to owners of the parent was ¥768,007 million; other loss included ¥597,248 million (US$3.73 billion) from a realignment of procurement plans for power generation-related equipment for U.S. AI data-center projects, including the PORTS-Pike Technology Campus in Pike County, Ohio, recorded at subsidiary Silver Bands 7 LLC. That six-month charge comprised ¥457,030 million (US$2.85 billion) of non-refundable advance payments and ¥140,218 million (US$0.88 billion) of contract modifications and settlements; SB7 had made cumulative procurement payments of ¥1,500,886 million (US$9.48 billion) through 30 September, and became entitled to rebates of up to US$2.1 billion over 10 years that were not recognized as an asset. A separate six-month impairment of ¥72,543 million was recognized on Lordstown, Ohio property and buildings acquired for a data-center-equipment manufacturing facility, separate from the US$3.73 billion realignment. On 30 September SoftBank completed the US$3.1 billion acquisition of DigitalBridge Group. Yahoo Finance: SoftBank Group (9984.T) closed at ¥3,738 on Thursday, down ¥117 or 3.0%. After or coincided with, not because. This is not a recut of SoftBank Astra / GPT-6 coverage. A reported GE Vernova comment was not verified against a GE Vernova primary and is omitted.

Why it matters

A named quarterly loss at the parent, driven in large part by a US$3.73 billion U.S. power-equipment realignment charge, is the window’s live test of AI-infrastructure procurement risk at a major listed backer.

More ▾ Less ▴
Confirmed
What happened

SoftBank Group’s English consolidated financial report, dated 10 November 2026, shows a second-quarter net loss attributable to owners of ¥1,115,337 million, or ¥196.54 a share, on net sales of ¥2,215,446 million, compared with net income of ¥2,502,247 million a year earlier. The basic-EPS numerator was a loss of ¥1,120,111 million after ¥4,774 million not attributable to ordinary shareholders, which is the ¥1.12 trillion figure used in wires. Those two yen prints are not the same line. For the six months ended 30 September, other loss included ¥597,248 million, or US$3.73 billion, from a realignment of procurement plans for power-generation equipment for U.S. AI data-center projects, including PORTS-Pike in Pike County, Ohio, at subsidiary Silver Bands 7 LLC. That charge split into ¥457,030 million (US$2.85 billion) of non-refundable advances and ¥140,218 million (US$0.88 billion) of contract modifications. Cumulative procurement payments were ¥1,500,886 million (US$9.48 billion) through 30 September, and rebates of up to US$2.1 billion over 10 years were not recognized as an asset. A separate six-month Lordstown impairment of ¥72,543 million must not be merged into the US$3.73 billion realignment. The same report records the 30 September close of the US$3.1 billion DigitalBridge acquisition. Yahoo Finance put the 9984.T Thursday close at ¥3,738, down 3.0%. That tape print is after or coincided with the results, not because of them. A GE Vernova comment was omitted pending a GE Vernova primary. This is the first Finance card of the 10 November print, not a recut of AI-desk SoftBank Astra or GPT-6 coverage.

Watch next
  • Any SoftBank or supplier follow-up on the procurement realignment or rebate recognition
  • Friday Tokyo follow-through, only if a later Asia-open amend is asked after about 09:30 HKT
Limitation

Figures are from the SoftBank IR primary. The ¥1.12 trillion wire print is the EPS numerator, not the attributable loss. Dollar figures are the company’s conversions where printed. The Tokyo close is Yahoo Finance.

Nebius Group, in a 12 November IR press release and CEO letter, said third-quarter revenue was US$980.8 million, up from US$146.1 million a year earlier (+571%). Operating income was US$50.4 million; net loss from continuing operations was US$37.5 million, or US$0.13 a share, versus a US$119.6 million / US$0.50 loss a year earlier. Adjusted EBITDA was US$465.8 million; adjusted net income was US$90.9 million. The company said some third-party capacity expected this year is now scheduled for the first half of 2027, with the contracts still in place, and updated 2026 guidance: revenue now expected at the low end of the existing US$3.0–$3.4 billion range; adjusted EBITDA margin unchanged at about 40%; capital expenditures approximately US$27 billion, above the prior US$20–$25 billion range; ARR, previously guided at US$7–$9 billion by December, now expected to exceed US$7 billion for the month of February, a shift of about two months, even before New Jersey ramps to full capacity; contracted power still expected at 5 GW by year-end. Initial 2027 outlook: revenue US$12–$13 billion; pre-tax breakeven by the end of 2027; capex US$40–$45 billion, with 75–85% expected to be covered by advanced prepayments and investment-grade asset-backed financing plus the balance sheet; more than 1 GW of new capacity each year. CNBC said shares sank 16%; Yahoo Finance regular-session close was $128.64 (−16.84%). This is distinct from already-Wired CoreWeave Q3 results and the raised outlook.

Why it matters

A 571% revenue print that still points 2026 sales to the low end of the range, while lifting capex to about US$27 billion, is the window’s live neocloud test after already-Wired CoreWeave results.

More ▾ Less ▴
Confirmed
What happened

Nebius’s 12 November 2026 earnings release printed third-quarter revenue of US$980.8 million, up 571% from US$146.1 million a year earlier, with operating income of US$50.4 million and a net loss from continuing operations of US$37.5 million, or US$0.13 a share. Adjusted EBITDA was US$465.8 million and adjusted net income was US$90.9 million. The CEO letter is the primary for the outlook update. Some third-party capacity expected this year is now scheduled for the first half of 2027, with the contracts still in place. 2026 revenue is now expected at the low end of the existing US$3.0–$3.4 billion range, not a secondary cut of that range to US$3.0–$3.2 billion. Adjusted EBITDA margin is still about 40%. Capex is approximately US$27 billion, above the prior US$20–$25 billion range. ARR, previously guided at US$7–$9 billion by December, is now expected to exceed US$7 billion for the month of February. Contracted power is still expected at 5 GW by year-end. The initial 2027 outlook is US$12–$13 billion of revenue, pre-tax breakeven by the end of 2027, and US$40–$45 billion of capex, with 75–85% expected to be covered by prepayments and investment-grade asset-backed financing plus the balance sheet, and more than 1 GW of new capacity each year. Yahoo Finance put the regular-session close at $128.64, or −16.84%; CNBC rounded the move to 16%. This is a distinct neocloud earnings card from already-Wired CoreWeave Q3.

Watch next
  • Formal 6-K or 20-F follow-through
  • Any further delay or financing update, including the first-half 2027 third-party capacity shift
Limitation

Quarterly figures are from the IR press release; 2026/2027 guidance wording is from the CEO letter. Share-price percentages are Yahoo Finance (exact) and CNBC (rounded). Guidance is outlook, not a reported result.

Cisco’s official newsroom release dated November 12, 2026, reported first-quarter fiscal 2027 results for the period ended October 24, 2026. The company printed revenue of $18.5 billion, up 24% year over year (the consolidated statements show $18,517 million), GAAP net income of $4.6 billion or $1.14 a share, and non-GAAP net income of $5.6 billion or $1.40 a share. It described these as record revenue and record GAAP and non-GAAP earnings per share, and “Cisco’s strongest revenue and non-GAAP earnings per share growth rates in over 15 years.” Product revenue was up 32%, with Networking up 43%. Cisco said it raised fiscal 2027 revenue and non-GAAP EPS guidance, now $74.8 billion to $75.8 billion of revenue and $5.47 to $5.53 of non-GAAP EPS, with GAAP EPS guided at $4.38 to $4.44. Second-quarter fiscal 2027 revenue guidance is $18.0 billion to $18.2 billion. The news summary also said Cisco “Raised expectation for AI infrastructure revenue from hyperscalers to $9B in FY27, more than double the total for FY26” — that $9 billion figure is an expectation, not a reported first-quarter segment result. This is distinct from already-Wired CoreWeave results and from this edition’s Nebius card. Share prices are omitted.

Why it matters

A same-window first-party earnings print that raises fiscal 2027 guidance and the company’s hyperscaler AI-infrastructure revenue expectation to $9 billion.

Yonhap (12 November ticker): Kospi closed 5,342.73 (−32.63). MoneyToday: that is −0.61%; foreigners net bought 370.5 billion won, while retail net sold 204 billion won and institutions net sold 691 billion won; the won closed at 1,330.7 per dollar, down 2.9 won. Businesskorea: Samsung Electronics ₩208,000 (+0.97%); SK hynix ₩1,254,000 (−1.80%). Aju Press: Kosdaq 696.5 (−8.6 / −1.2%). Nikkei Indexes: Nikkei 225 60,819.36 (−0.70% / −429.23); Technology contributed −447.96 yen; advancers 153 / decliners 72. RTHK: TOPIX edged up almost eight points, or 0.2%, at 3,937. Xinhua: HSI 24,008.55 (−1.26%); HSCEI 8,056.76 (−1.4%); HSTECH 3,931.93 (−2.36%). RTHK rounded HSI 24,008 (−307 / 1.3%) on HK$202.65 billion turnover, and printed Shanghai Composite 3,655 (−0.67%), Shenzhen Component 11,197 (−1.68%), and ChiNext 2,509 (−2.44%). The regional pullback traded after Wednesday’s already-Wired U.S. rebound and after Nvidia-leadership reports that remain AI-desk owned — after or coincided with, not because. Distinct from already-Wired Wednesday Asia (Kospi +2.60% / Nikkei 61,248.59 / HSI −0.98%).

Why it matters

A one-day Seoul, Tokyo, and Hong Kong pullback, with Samsung up and SK hynix down, is the live Asia cash follow-through after two strong Korea and Japan sessions.

NBC News and The Guardian (12 November), quoting a Truth Social post: President Trump wrote, “We don’t need a deal with Iran. Those days have long since passed. We’ve got their money, Hundreds of Billions of Dollars, and we’ve got their Strait!” NBC: he said he asked Vice President JD Vance to focus not on diplomatic talks with Iran but on alleged “crime and corruption” in “Democrat run Cities and States.” The Guardian: Vance had spearheaded recent efforts to restart negotiations, and the post appeared to shut those talks down and redirect him to domestic issues. This is a distinct, later increment from already-Wired Monday diplomacy and from Tuesday’s Developing minor Iran air-defense strikes. It is distinct from those earlier cards and from Thursday’s already-Wired oil session. After or coincided with Thursday’s crude jump, not because of a confirmed new military operation or a signed Hormuz agreement. “Hundreds of Billions” and “we’ve got their Strait” are the president’s claims, not independently verified figures or a legal status of Hormuz.

Why it matters

An on-record presidential statement that the United States does not need an Iran deal, and that Vance should stand down from that track, is the live official policy signal for the oil and rates tape.

NBC News (12 November, updated 12:55 p.m. ET): Treasury Secretary Scott Bessent moved to a military base following threats from Iran, according to a source familiar with the situation; the move was first reported by The New York Times. The source told NBC the concerns rose after The Wall Street Journal published a photo that allowed Bessent’s Washington home to be geolocated, and that “security concerns began to increase which have now forced the Secretary to relocate to a military base for his safety.” A Treasury spokesperson said Bessent “will not be deterred by the Iranian regime’s threats and remains committed to executing Operation Economic Outcast and maintaining the blockade, which together are collapsing Iran’s economy.” NBC: the department has not revealed the nature of the threats that led to the move. This is distinct from the Trump/Vance post and from already-Wired Iran cards and Thursday oil. After is not because on oil or Treasuries.

Why it matters

A named U.S. Treasury secretary changing residence over Iranian threats, with an on-record department comment, is a live security increment around the same Iran-policy turn as the no-deal post.