Daily Brief Nov 13, 2026 · How we checked →

Huang resigns as NVIDIA CEO; filing doesn't say if forced

Today's front-page cartoon: one single picture that brings together today's stories.

At a glance

1. NVIDIA's CFO Colette Kress becomes acting CEO, and the November 17 earnings report stays on schedule.

2. Google launches the low-cost model Gemini 4 Hydrogen, which scores 57% on the Terminal-Bench 4.1 terminal-task benchmark in Google's own test, against 18.2% for the previous Gemini 3.8 Flash.

3. Perplexity releases the classification model Decider, which it says costs about 1/12 as much as GLM-5.3-Flash run through an API, not counting local hardware.

Also today: Nebius, a cloud provider that rents out GPUs, steps up capital spending again next year (main-line item 4).

Correction: the November 11 issue said TSMC's subscription for Anthropic shares differed from AMD's practice of trading equity for orders. It said AMD issues its own equity, while TSMC pays cash for shares. But the amended Anthropic IPO registration statement that issue cited states that AMD, in the same round, is also paying cash for shares. That money offsets AMD's share-purchase commitment under a hardware supply agreement and is drawn down in installments as AMD chips are delivered and deployed. For TSMC's subscription, the document describes no link to any supply agreement. That issue also said TSMC chairman C.C. Wei's July answer had been transcribed by only one reporter; The Motley Fool also transcribed the same session, and the question was whether TSMC would invest in its customers' customers. The archived page on the site will be corrected; the conclusion that TSMC's filing states no order terms is unchanged.

Today's main line

1. NVIDIA's CFO Colette Kress becomes acting CEO, and the November 17 earnings report stays on schedule

Concept illustration in line-drawing style of a room: on the left, three people sit in pale chairs along a long table, the leftmost holding an open document drawn with lines of text and the other two each with a dark folder lying flat on the table, while an empty high-backed red office chair stands at the right end of the table. Past a vertical wall post on the right is a closed door bearing two signs lettered "CFO" and "INTERIM CEO", and on the wall beside it is a small sign with a green emblem and the lettering "NVIDIA". Printed on the image: NVIDIA says the board has begun searching for a permanent CEO. — In a filing with the US Securities and Exchange Commission, NVIDIA said the board accepted Jensen Huang's resignation as president, CEO and director on November 11, effective immediately, and made Colette Kress, CFO since 2013, acting CEO. The filing doesn't say whether a removal process came first. NVIDIA's press release says third-quarter results will come out on November 17 as originally scheduled.

Our inference: November 17's report runs only to October 25, so it still covers a quarter Huang ran. The new leadership's approach to investments, guarantees and payment terms will show first in the report's notes on events after November 11 and in management's outlook.

In the full story: how the permanent CEO will be found, when the third quarter ended, and which terms the new leadership can change

2. Google launches the low-cost model Gemini 4 Hydrogen, which scores 57% on the Terminal-Bench 4.1 terminal-task benchmark in Google's own test, against 18.2% for the previous Gemini 3.8 Flash

Concept illustration: a large black-framed board stands on two legs, and along a line across its lower part sit, from left to right, a black rounded block lettered FLASH, a short stack of a few cream sheets with grey lines on the top one, a second black block lettered HYDROGEN, and a much taller stack of red sheets whose top sheet also has grey lines. A small white sign at the board's lower right shows a multicoloured letter G next to the lettering GOOGLE, and a pale pink rectangle fills the background behind the right side. Printed on the image: 57% Hydrogen's Terminal-Bench 4.1 score vs 18.2% for previous Gemini 3.8 Flash; Google's own test. — Google DeepMind released Gemini 4 Hydrogen, the fast, low-cost model in the Gemini 4 family, on November 12, and Google's coding tools split work automatically between Hydrogen for execution and the pricier Gemini 4 Carbon for deep reasoning. Most comparisons with the previous generation are Google's own tests. Google says Hydrogen leads same-class models but does not name which companies' models or which versions were in the comparison.

In the full story: which products it launched in, who measured the scores against the previous model, and how Orchestrator splits the work

3. Perplexity releases the classification model Decider, which it says costs about 1/12 as much as GLM-5.3-Flash run through an API, not counting local hardware

Concept illustration: a large board topped by a black band with a teal geometric shape and the lettering PERPLEXITY is split by a vertical line into two halves. On the left a black laptop lettered DECIDER sits beside a short stack of red discs and a tag reading COST, and on the right a black laptop lettered GLM-5.3-FLASH sits beside a much taller stack of grey discs and another COST tag; in the lower right corner a dark-haired person in dark clothing, seen from behind, faces the board. Printed on the image: About 1/12 Decider cost vs GLM-5.3-Flash via API, same requests, per Perplexity; excludes local hardware. — On November 12 Perplexity released Decider v1.3, a classification model that returns the probability of each answer option for every question in one pass without generating an answer word by word. The cost comparison between Decider and the small LLM GLM-5.3-Flash run through an API is self-reported by Perplexity and excludes the local hardware Decider ran on. In Perplexity's test on live traffic from before the launch, Decider v1.3's answers agreed with a frontier large model's answers 93.98% of the time, against 93.68% for the previous v1.2.

In the full story: how the costs were compared, who supplied the reference answers, and whether the weights can be downloaded

4. AI cloud provider Nebius forecasts 2027 capital spending of US$40 billion to US$45 billion, while 2026 revenue lands at the low end of guidance

Concept illustration: on a slightly tilted light-grey board, two cylindrical stacks of discs stand on a black baseline, a shorter beige one on the left and a taller brick-red one on the right, and a line runs from each stack to a box lettered CAPEX: the left stack's box is to its left and the right stack's box is to its right. Four grey horizontal lines are drawn on each side of the board, and in the top-right corner a blue-and-cyan angular mark sits next to a box lettered NEBIUS, from which a line runs right to a black dot. Printed on the image: US$40B–45B Nebius's self-reported 2027 capital spending forecast, vs about US$27 billion in 2026. — Nebius, a cloud provider that specializes in renting out GPU compute, issued a 2027 capital spending forecast on November 12 and says 2026 revenue will land at the low end of the US$3 billion to US$3.4 billion guidance. Nebius also writes that prepayments on new third-quarter contracts cover more than 100% of the capital spending tied to those contracts, against 50% to 60% in the second quarter, a ratio that counts only newly signed contracts. All of the figures are self-reported by Nebius, and the shareholder letter isn't audited.

In the full story: how the revenue target changed, how much spending new-contract prepayments cover, and how contracted power differs from online capacity

Chips & semiconductors

1. Applied Materials says revenue at its semiconductor equipment segment will grow more than 40% over a one-year period the company defined, running from the second quarter of fiscal 2026 through the first quarter of fiscal 2027.

Concept illustration: a large beige machine cabinet with vent slats on both side panels and a black top band showing a blue geometric mark at the left and the lettering APPLIED MATERIALS; in its central window, on one shelf line, a stack of seven flat beige rounded slabs sits on the left, and on the right, in front of a pale pink patch, a taller stack has seven beige slabs with three brick-red slabs on top. A person in black clothing with a white lanyard badge stands to the left of the cabinet with their face turned toward it, and a drawer panel with a small handle runs along the cabinet's bottom. Printed on the image: Over 40% Applied Materials says equipment revenue grows, fiscal 2026 second quarter–2027 first; base unstated. — Applied Materials CEO Gary Dickerson said in the November 12 earnings press release that semiconductor equipment segment revenue will grow over a one-year period the company defined, and that Applied Materials will grow faster than the wafer equipment market over the next few years. The growth covers one company's segment revenue, not the size of the global wafer equipment market, and the press release doesn't say which period the growth is compared with.

2. TSMC filed with the US SEC that data on some current and former employees appeared without authorization on an outside website and matches data in a cloud service provider's systems.

Concept illustration: two hands hold open a scroll from the left and right sides. The scroll's black top band carries large white letters "TSMC" and a white box with a round grid emblem lettered "tsmc" in red. Below it, on the left, a computer monitor headed "WEBSITE" shows three grey text lines beside a stack of red cards, each with a black person silhouette. On the right, a panel headed "CLOUD PROVIDER" shows a similar stack of red person cards between two black rack-like units marked with grey lines. Printed on the image: Data on an outside website matches data in a cloud service provider's systems, TSMC filed. — TSMC filed with the US SEC that data on some current and former employees appeared without authorization, and the filing doesn't give the number of people affected. TSMC writes that TSMC has found no evidence so far that TSMC's systems were breached.

3. Epoch AI, an independent research group that studies AI compute, estimates from satellite imagery that Huawei's AI campus in Horinger, Inner Mongolia, holds about 81,000 H100 equivalents of compute, about 7% of the compute at Colossus 2, an AI data-center campus in Tennessee.

Concept illustration: a wide pull-down screen has a sign lettered EPOCH AI at the centre of its top bar. On the screen's left is a column of three red-brown chip icons above a label lettered HUAWEI, and on its right are three rows of 13 dark grey chip icons above a long label lettered COLOSSUS 2. To the right of the screen stands a long-haired figure in a black top and light wide trousers. Under one arm the figure holds a partly rolled sheet printed with four boxes of small dark squares, and with the other hand the figure holds the cord hanging from the screen's lower right corner. Printed on the image: About 7% Huawei Horinger AI campus compute as a share of Colossus 2's, Epoch AI satellite-imagery estimate. — Epoch AI, an independent research group that studies AI compute, estimates the compute at Huawei's AI campus in Horinger, Inner Mongolia, from satellite imagery with a model that assumes the campus holds no export-controlled NVIDIA chips. In the same article, Epoch also gives the compute figure for Colossus 2, an AI data-center campus in Tennessee.

Named commentary

1. Chip research firm SemiAnalysis argues that NVIDIA is a company built around Jensen Huang with a thin succession bench, and that the two internal technical leaders being talked about would struggle to take over Huang's command of the supply chain.

Concept illustration: a large white board on legs stands against a mustard-yellow background. At its top left is a framed sign with a green emblem and the lettering NVIDIA, with a big black opening quotation mark below it, and a hand in a black sleeve reaches in from the top right, pointing down at a thick red twisted rope. A figure on the left in a white shirt and black trousers holds the rope in both hands; the rope runs right above two smaller figures standing side by side, then splits into four thin red lines that reach the chimneys of four small factory buildings on a stand lettered SUPPLY CHAIN. Printed on the image: SemiAnalysis argues Ian Buck, Jonah Alben would struggle to take over Huang's supply-chain command. — Chip research firm SemiAnalysis, known for chip supply-chain and data-center models, argued on November 12 that NVIDIA is a company built around Jensen Huang with a thin succession bench. SemiAnalysis attaches no data to the judgment, and SemiAnalysis sells data to chipmakers and investment firms.

2. Eli Lilly CEO Dave Ricks, responding to Anthropic CEO Dario Amodei's claim that major diseases will be a thing of the past within ten years, says that would be great but is unlikely.

Concept illustration on a pink background: a brick-red path runs from the bottom edge to a vanishing point at the centre, with a long row of hospital beds on each side receding into the distance, patients lying in them and capsules on some of the small bedside tables, and a sign lettered HUMAN TRIALS at the front of the left row. At left, a desk holds a laptop whose screen shows a speech-bubble icon, a rack of three test tubes and a tray of dark pills, above a sign lettered LILLY and, in red script, Lilly; at right, a person seen from behind stands at a lectern facing a screen with a play button, with a sign in front lettered ANTHROPIC and an angular mark beside the word; a large thought bubble at the top holds an empty hospital bed and bedside table, and its trail of small circles leads down to the right toward that screen; a large black quotation mark sits in the top-right corner; a hand in a dark sleeve reaches in from the lower left with its index finger pointing up and to the right toward the path; and a single capsule lies on the path near the bottom. Printed on the image: Ricks cites randomized human trials as the slowest step. — Eli Lilly CEO Dave Ricks says Anthropic CEO Dario Amodei's claim that major diseases will be a thing of the past within ten years would be great but is unlikely, and offers that view as a personal judgment. Amodei's remark comes from a video connected with Anthropic's IPO, reposted by a third-party account.

Research & models

1. A preprint finds that a single fake news article in an AI forecasting system's search results can reverse 74% to 82% of the forecasts that originally opposed the article's position, while average accuracy barely changes.

Concept illustration: a frame topped by a black bar lettered ARXIV holds two side-by-side scenes on one shelf line; on the left, a stack of beige pages with grey text lines, propped by a black bookend, stands beside an open laptop with a blank screen whose speech bubble shows a black thumbs-up. On the right, a similar stack of beige sheets with a black bookend has a red sheet lettered FAKE NEWS with a white thumbs-down in front, beside an identical laptop whose speech bubble shows a black thumbs-down. Printed on the image: 74%–82% Share of AI forecasts opposing a fake article that the one article reversed, per preprint authors. — A preprint by Zeyu Zhang and Bradly C. Stadie finds that average accuracy barely changes when one fake news article enters the search results of AI forecasting systems, which search the web before making a call, against an attacker who knows only the forecasting question. The paper doesn't name the two forecasting systems tested, and the results are the authors' self-reported findings that no one else has repeated yet.

2. The preprint Lazarus plants a backdoor in open-weight agent models by rewriting just 7 neurons, without retraining, a result the authors self-report.(arXiv 2611.15118, 2026-11-11)

Product moves

1. Google tested its tool Agent Anomaly Detection, which flags agents overstepping after the fact, on agent-safety benchmarks written by 5 outside teams. The tool flagged 87% of harmful instructions that were executed, but caught only 4 of 18 poisoned instructions written into long-term memory.

Concept illustration: a dark grey computer monitor has a blue header bar lettered MEMORY across the top of its screen; on the white screen below, a small grey rectangle at the lower left holds a black bug icon and is outlined by a red frame, while a larger, taller grey rectangle to its right holds a bigger black bug icon with no frame. To the right of the monitor stands a dark sign lettered with a multicoloured G and GOOGLE, and below that AGENT ANOMALY DETECTION, all against a pale grey-blue background. Printed on the image: 4 of 18 Poisoned instructions in long-term memory caught by Google's after-the-fact tool, Google-run test. — Google tested Agent Anomaly Detection, a Google tool that reads agents' execution logs and flags sessions that overstep, on agent-safety benchmarks written by 5 outside teams, using only one older model. Agent Anomaly Detection only reports after the fact and doesn't block actions.

2. Databricks' data agent Genie is now generally available in Slack, querying data with the asker's own permissions and requiring user approval before changing outside documents, with no usage or pricing figures published.(Databricks, 2026-11-12)

Also happened

1. Not verified by us yet: According to a Reuters report relayed by reporter Dan Nystedt, House Speaker Mike Johnson says a consensus is forming in Congress on restricting chip exports to China (Dan Nystedt, 2026-11-12).

2. Not verified by us yet: According to a report in The Information relayed by analyst Paul Triolo, the Commerce Department's remote-use rules now being finalized extend the countries requiring a license to Brazil, Indonesia, Norway and South Africa (Paul Triolo, 2026-11-12).

3. Not verified by us yet: Analyst Ming-Chi Kuo wrote on X that with NVIDIA's succession uncertain, Google's TPU may be the biggest beneficiary, with the effect showing in 2028 at the earliest (Ming-Chi Kuo, 2026-11-12).

4. Not verified by us yet: According to commentator Zvi Mowshowitz, citing another source, Anthropic's IPO price range is US$70 to US$78 a share, a fully diluted market value of US$1.27 trillion to US$1.41 trillion (Zvi Mowshowitz, 2026-11).

Sections with nothing new this issue

From the archive

SecondSource isn't a news digest: AI reads primary sources across fields, every judgment comes with its original source, and mistakes are corrected in public. The focus is what has changed and which claims still lack evidence, not what happened today.

Written from the same research and judgments as the Traditional Chinese edition. Sources are linked; we distinguish original documents from reporting and mark what we could not verify.