Silicon Valley Daily
Nvidia's board is preparing to oust Jensen Huang, the Wall Street Journal reported Wednesday evening, minutes after the New York Times said Huang had an affair with Kimberly Powell, a vice president who reports to him. Bloomberg says the board is working with an outside firm on replacements, and Semafor says some directors favor CFO Colette Kress as interim CEO. Shares that had rebounded about 5% in the session fell more than 8% after hours. In Washington, lawmakers from both parties called for Huang to resign, Warren and Coons demanded records be preserved, and Axios reported Commerce Secretary Lutnick stalled chip-diversion curbs Trump had approved for a third time. Elsewhere, CoreWeave jumped 24.7%, Bloomberg says Anthropic's IPO is oversubscribed, AWS released an emulator written entirely by AI agents, and Google shipped Gemini Decision 1 and AlphaGenome 2.
WSJ: Nvidia's board is preparing to oust Jensen Huang
The Wall Street Journal reports Nvidia's board is preparing to remove CEO Jensen Huang and could meet as soon as Wednesday to decide his future. Minutes earlier, the New York Times, citing five company sources, reported that Huang had an extramarital relationship with Kimberly Powell, Nvidia's vice president of health care, who reports to him; the two were filmed being affectionate at a San Jose hotel during GTC in March. Huang admitted on Sunday to an affair with Alice Huang, whom U.S. officials suspect of spying for China, and Trump said Monday the FBI is looking into it.
Nvidia board weighs interim CEO; some directors favor CFO Colette Kress
Bloomberg reports Nvidia's board has been working with an outside firm to identify possible replacements for Huang and is weighing a temporary CEO during the search. Semafor reports some directors favor naming CFO Colette Kress interim CEO, and an administration official said the board briefed the White House on its meeting. On Sunday the board had backed Huang to stay on.
Congress escalates: senators urge Huang to resign; Warren and Coons demand records
House Foreign Affairs chair Brian Mast and China committee chair John Moolenaar said Wednesday afternoon they had not heard from Huang and gave him until the end of the day to agree to a closed-door briefing, with a subpoena ready. Republican Sen. Kevin Cramer told CNBC Huang should resign, as did Democratic Rep. Suhas Subramanyam. Sens. Elizabeth Warren and Chris Coons asked the administration to preserve all documents tied to Nvidia's China chip exports and the affair, often a first step toward a formal investigation.
Axios: Trump approved chip-diversion curbs a third time; Lutnick stalled again
Axios, citing six sources, reports that Trump on Nov. 5 authorized a crackdown on companies suspected of diverting Nvidia chips to China for the third time this year, after calling Elon Musk. At a Situation Room meeting that day, Treasury Secretary Scott Bessent confronted Commerce Secretary Howard Lutnick over why Megaspeed had not been added to the entity list; Lutnick appeared to agree, then muttered that he needed to check with the president. A Commerce spokesperson denied the confrontation took place.
Anthropic IPO draws multiple times the shares on offer
Bloomberg reports Anthropic's initial public offering has drawn orders for multiple times the shares available. A source said CEO Dario Amodei and JPMorgan's Jamie Dimon will hold a livestream event on Friday.
Nvidia rose about 5% in the session, then fell more than 8% after hours
Nvidia shares rose close to 5% during Wednesday's session, then fell more than 8% after hours once the Journal and Times reports landed. Axios says the company has lost about $1 trillion in market value since the Times first reported on Oct. 29 that Huang had a close relationship with a suspected spy. Earlier in the day Morgan Stanley removed Nvidia from its top-pick list; after hours CoreWeave fell more than 4% and Supermicro nearly 5%.
Google launches Gemini Decision 1 for scored yes/no and multiple-choice answers
Google launched the Gemini Decisions API, powered by a new Gemini Decision 1 model that scores candidate answers in a single forward pass instead of generating text. It handles up to 1,000 yes/no, multiple-choice or rubric questions over text, images, audio, video and PDFs, returns a 0–1 confidence score for each, and costs $0.10 per million input tokens with no output tokens billed. Google says end-to-end latency is 150–250 ms and the API is nearly 5x cheaper than Gemini 3.5 Flash-Lite on its benchmark suite.
Bloomberg: Megaspeed founder Alice Huang has returned to China
Bloomberg, citing people familiar with the matter, reports that Megaspeed founder Alice Huang, the suspected Chinese spy at the center of the Nvidia scandal, has returned to China. One source said she is living in Beijing with the knowledge of security officials.
White House asks Congress to force a rule on remote AI-chip access
The Information reports the White House asked Congress to require the Commerce Department to write a rule blocking remote access to certain AI chips. Nvidia has lobbied against closing that loophole, which lets overseas customers use restricted chips hosted elsewhere.
DeepMind's AlphaGenome 2 predicts variant effects in specific cell types
Google DeepMind released AlphaGenome 2, trained on more than 100 million single-cell profiles across 68 tissues to predict how DNA variants change gene expression in specific cell types and states. DeepMind says it identifies nearly twice as many expression-changing variants as its previous model. It is available via API for non-commercial research and in private preview for commercial use.
UK chipmaker Fractile raises $720M Series C
Fractile raised a $720 million Series C led by Thrive Capital, Founders Fund, Lightspeed and Redpoint, with backing from Britain's Sovereign AI Fund; the UK government put the valuation at over £5 billion. The company stacks DRAM directly on its processors to ease the memory-bandwidth limit on large-model inference, and CEO Walter Goodwin says it will ship its first chips and systems next year.
Semafor: Huang grew more admiring of China after US chip curbs
Semafor analyzed 42 Jensen Huang interviews since 2022 and found his public rhetoric became more admiring of China, and more critical of export controls, after US chip restrictions took effect. The analysis adds to scrutiny of Huang's influence over chip policy.
DeepMind's AlphaProof Code formally verifies Signal's cryptographic core in four weeks
Google DeepMind used a new framework, AlphaProof Code, to formally verify key security guarantees of the Signal protocol's cryptographic core code in four weeks. Agents built on Gemini Carbon wrote more than six million lines of Lean proofs and flagged about 1% of the codebase for human review; two deliberately planted bugs were caught. The work did not cover the full Signal app.
Applied Intuition in talks to raise $1.5B at about $30B valuation
Bloomberg reports autonomous-vehicle simulation software maker and defense contractor Applied Intuition is in talks to raise at least $1.5 billion led by General Catalyst at a pre-money valuation of about $30 billion.
CoreWeave jumps 24.7% and plans an investment-grade loan of $4B+
CoreWeave shares rose 24.71% on Wednesday, their best day since April 2025, after third-quarter results beat estimates. Bloomberg reports the company plans to raise at least $4 billion through an investment-grade loan as soon as next week; CEO Michael Intrator told CNBC it should price at 7% to 7.5%, about half the yield on its junk bonds. Intrator also said he has paused his 10b5-1 stock sale plan, under which he had sold $794.2 million of shares.
Meta infrastructure chief Janardhan to leave; ex-AWS exec Dave Brown takes over
Santosh Janardhan, Meta's head of infrastructure and co-head of engineering since joining in 2009, will leave in early 2027, according to an internal memo. David Brown, a former AWS senior vice president who joined Meta in July, will lead infrastructure and serve as CTO of the Meta Compute initiative run by Dina Powell McCormick and Daniel Gross.
Nvidia shareholder sues company and Huang over Megaspeed statements
Shareholder Mark Kistenmacher sued Nvidia and Jensen Huang, seeking class status for investors who bought shares between Oct. 9 and Nov. 8. The complaint targets an Oct. 29 company statement that Huang's interactions with Alice Huang were routine customer meetings, which it alleges was untrue. Engadget noted Nvidia was on track to lose more than $470 billion in market value over two days.
ChinaTalk: US data vendors sell about $500M in training data to Chinese labs
A ChinaTalk investigation found at least nine US data firms, including Mercor, Surge AI, AfterQuery and Turing, selling roughly half a billion dollars of training data to Chinese AI labs, including reinforcement-learning environments first built for OpenAI or Anthropic. Buying proven environments lets the labs skip costly development and failed training runs, and the trade falls largely outside export controls; the authors call for controls in areas such as cybersecurity and biology.
Georgia residents file RICO suit against OpenAI over Project Camellia data center
Three Effingham County, Georgia residents sued OpenAI, its affiliates and county officials under civil racketeering law over Project Camellia, a planned 3.2-gigawatt data center expected to cost $20 billion to $30 billion. They allege the defendants used NDAs and misleading statements to keep residents uninformed and conspired to advance the project through closed-door decisions, and say nearby property offers have dropped or been withdrawn.
AWS releases a local emulator it says was written entirely by AI agents
AWS released AWS Emulator, a free tool that simulates the APIs of more than 200 AWS services on a local machine or in CI, without an AWS account. AWS says its engineers did not write or review a single line of the code, which was produced by thousands of autonomous agents. The project is open source on GitHub and publishes per-service fidelity scores with each release.
Torus raises $500M to build container-sized AI data centers
Salt Lake City energy-storage startup Torus raised a $500 million Series C led by Magnetar, Bloomberg reports. The money will fund a shipping-container-sized AI data center product in 1 MW to 20 MW sizes.
Alibaba files to list chip unit T-Head in Hong Kong
Alibaba said T-Head, its data-center chip design unit, confidentially filed for a Hong Kong listing on Sept. 11, and the exchange has cleared the proposed spin-off subject to listing approval. Alibaba expects to keep control after the listing; offering size and structure are still being worked out.
Hawley backs a data center pause; US data center power grows 10x faster than the grid
Politico reports Missouri Republican Sen. Josh Hawley joined progressive Democrats in calling for a pause on data center construction until Congress guarantees taxpayers will not face higher utility bills. Separately, insurance broker Howden found that power delivered to US data campuses has grown 19% a year since 2000 while grid capacity grew 2% a year, lifting data centers to about 5% of grid capacity in 2025.
Musk targets half a gigawatt of usable SpaceX compute next year
Elon Musk said SpaceX has a chance of bringing about half a gigawatt of actually usable compute online next year, adding that it takes his team about two months of round-the-clock work to make a large cluster reliable. The Information reports delayed power permits for its Texas data center projects could leave SpaceX paying for leased chips before facilities are ready. SemiAnalysis says SpaceX is designing its own Kessel rack with 72 Nvidia Rubin GPUs on an 84-layer PCB midplane, for both terrestrial and orbital use.
Agent startups raise: Rox $100M, Imprint $130M, Edelbyte $70M, Plurai $13M
Rox, which builds AI agents for customer research and outreach, raised a $100 million Series B co-led by General Catalyst and March at a $1.3 billion valuation. Co-branded card platform Imprint raised $130 million led by Khosla at about $1.9 billion as it targets loyalty in agentic commerce. Edelbyte emerged from stealth with $70 million for a data analytics platform built for AI agents, and Plurai raised a $13 million seed to test agents against digital twins of corporate IT systems.
EU proposes Digital Fairness Act targeting dark patterns and loot boxes
The European Commission published the proposed Digital Fairness Act, which would ban manipulative dark patterns such as fake scarcity claims, require options to turn off addictive features, and tighten rules on loot boxes and undisclosed influencer marketing. Apps such as Duolingo, Strava and Ticketmaster could be affected, and the Commission estimates savings for EU consumers of up to €13.7 billion a year.
United Airlines will block personal AI assistants from booking tickets
The Wall Street Journal reports United Airlines will prevent personal AI assistants from booking flights, citing data-privacy risks and possible FAA and DOT rule violations. An executive said the airline does not want to create another middleman.
SpaceXSI launches Grok Bot 1.0 as shareable team agents
SpaceXSI's Grok Bot 1.0 lets teams share bots, work with them in Slack and Teams, invite them to meetings and have them learn from how the team works. Elon Musk urged users to hire it as a colleague and give it their hardest tasks.
Alibaba opens Qwen4-Max preview with a 1M-token context
Alibaba opened a preview of Qwen4-Max, the flagship of its next-generation Qwen family, with image and video input, tool calling and up to 1 million tokens of context. Developers can use it through Alibaba's Token Plan subscription until Nov. 25, and it is also available in the Qwen app and Qoder. Alibaba has not published benchmark results.
IBM to link two 520-qubit Nighthawk r3 chips into a 1,040-qubit system in 2027
IBM unveiled Nighthawk r3, a processor targeting 520 programmable qubits and 10,000 quantum operations. It plans to offer a 1,040-qubit machine in 2027 by linking two chips, rather than nine 120-qubit chips as its earlier roadmap described. IBM says it still aims for error-corrected quantum computers before 2030.
Ramp: US corporate AI spending grew 1.65% in October
The Ramp AI Index shows US business AI spending hit a new high in October but grew only 1.65% month over month. The top 1% of spenders paid a median $10,200 per employee per month while the typical company spent $16.32, and Ramp says that 1% drives 80% of revenue for OpenAI and Anthropic. Software development teams accounted for 72% of tracked AI spending.
Benchmark leads $30M round for Tendrils, which wants to redo the CPU
Cambridge-based Tendrils Compute raised $30 million led by Benchmark, with the UK Sovereign AI Fund and Conviction among participants. It is building a processor with thousands of parallel cores based on interaction nets, alongside an open-source language called Vine; it has not yet published hardware benchmarks.
Apple closes its 28-year-old Support Community forum on Nov. 20
Apple put its Support Community forum into read-only mode and said it will close permanently on Nov. 20, 2026, pointing users to the official Apple Support site. The forum had run for 28 years.
a16z's Mike Volpi: old GPUs are outliving their depreciation schedules
On an a16z podcast, Mike Volpi compared the AI buildout with the fiber boom he saw at Cisco, noting that bankrupt fiber owners' assets still proved useful later. The hosts noted H100s once rented for $1.80 to $2 an hour on an assumed five-year life; Volpi said they now rent for $2.50 to $3 because demand is high and chip manufacturing remains the bottleneck. He also advised Nvidia to make enough acquisitions to get a statistically meaningful result.
After midterms, Clegg says tech's Trump embrace backfired; Karp courts Democrats
Former Meta executive Nick Clegg wrote in The Atlantic that Silicon Valley's herd move toward the Trump administration has backfired, with Democrats possibly seeking retribution and anti-Big Tech sentiment growing. Politico reports Palantir CEO Alex Karp has described his philosophy as progressive since the midterms and praised Trump for saying ICE went too far.
Google brings the Gemini app to Apple CarPlay
Google began rolling out a Gemini app for CarPlay to iOS users worldwide, letting drivers hold Gemini Live voice conversations that sync with the Gemini app on their iPhone. Gemini was previously reachable in CarPlay only through Google Maps.
Google open-sources XSE, which serves 6M queries a second without Python
Google open-sourced XSE, the XLA Serving Engine, which runs compiled ML models without a Python interpreter in the serving path. It handles six million queries a second across Search, Ads and YouTube Shorts, and targets ranking, recommendation, embedding and vision models rather than large-scale LLM serving.
Databricks CEO uses revised Genie agent to surface bad news
Databricks revised its Genie agent to run multistep jobs unsupervised, using an ontology built from data customers store on its platform. CEO Ali Ghodsi showed TechCrunch how he uses it to analyze weak spots in an employee survey and a failed sales pitch, information he says a CEO of a 12,000-person company rarely hears directly.
Indie Games Festival jurors reject AI; Big Walk admits minimal LLM use
Nineteen members of the 50-plus Independent Games Festival jury said they will decline to vote for any game that uses generative AI in any capacity and urged an anti-AI policy for future years. Big Walk developer House House then disclosed that LLMs were used sparingly to fix technical problems, in well under 0.1% of the game's code.
England's university regulator demands AI-free assessments
England's Office for Students says AI has made coursework nearly impossible to assess fairly and urged universities to return to exams and in-person presentations, warning it could sanction institutions that do not adopt reliable formats. A survey of more than 300 respondents at 127 institutions found students using AI for essays and dissertations, often undetected.
Appeals court blocks DOJ's sweeping demand for journalists' LinkedIn messages
A unanimous three-judge appeals panel ruled that a Justice Department leak probe cannot sweep in the confidential sources of journalists with no conceivable link to the investigation. The DOJ had asked LinkedIn for 18 months of targeted journalists' communications under a secrecy order. Judge J. Harvie Wilkinson III declined to quash the order entirely but sent it back for the district court to substantially narrow.
Yandex says drone strikes on its data centers destroyed 2.84% of customer data
Russian cloud operator Yandex said it restored more than 97% of customer data after an October drone strike caused a fire and outage in its ru-central1-b zone, losing 2.84% because of hardware damage and a lack of backups on the platform. It will now enable backups by default for new VMs and opened a new availability zone.
Helsing unveils Clarity, an agentic AI system for target recommendations
European defense company Helsing unveiled Clarity, a platform that uses AI agents to distill sensor data into target recommendations for weapons systems under human oversight and European control. Helsing says every decision will carry a human name.
Trending on X
What X was arguing about today that the news hasn't covered. Views are the posters' own.
Balaji Srinivasan and Tim Sweeney urge Nvidia's board not to fire Huang
Balaji Srinivasan (@balajis), replying to an Economist piece arguing Huang's position is untenable, wrote that Bill Clinton did not resign and Huang should not either, and that the board should not fire him in the heat of the moment. In a second post he said that if Nvidia terminates its founder, Huang may land at TSMC, calling it a win for Asia and a loss for America. Epic Games CEO Tim Sweeney (@TimSweeneyEpic) said an ouster would be Apple-firing-Steve-Jobs-level idiocy and a grievous harm to the United States.
Rep. Ted Lieu says Congress heard Huang testify without being told of the affair
Democratic Rep. Ted Lieu (@tedlieu) wrote to the White House that he attended a closed-door House Foreign Affairs hearing on AI export controls last year where Huang testified, and that it would have been helpful to know Huang was having an affair with a Chinese spy who ran a company that needed AI chips. The post he quoted argued Trump's remark that the affair did not make much of an impact on him shows he knew months ago. Commentator AG Hamilton tied the spy case to China-linked groups helping block data-center projects, calling both part of China fighting for the future.
Jukan posts second-hand notes saying Anthropic told analysts the SpaceX deal ends mid-2027
@jukan05 posted second-hand notes from an Anthropic meeting with sell-side analysts, including that the AI market is winner-take-all, that the SpaceX compute contract will not be extended beyond the middle of next year, and that developing Opus 5.5 cost more than $10 billion. Elon Musk replied that SpaceX would be happy to take the compute back but thinks the opposite will happen. @marsrepublica countered that no verified Anthropic source says this, noting the non-cancellable periods end in January and March 2027 and either side can then exit on 90 days' notice.
Dragonfly's Haseeb Qureshi says founders should ship a company.md for investors' AI
Haseeb Qureshi of Dragonfly said founders are starting to include a company.md, a markdown file written for AI consumption, alongside the deck and data room, and called it a very positive signal. He said it lets his agent probe the business before a meeting, but warned he will ask his AI how biased the file is. Ben Lakoff replied that VCs should also run a security check for malicious instructions hidden in founders' .md files.
Alex Lieberman says founders can be great parents; Linear's Karri Saarinen agrees
Alex Lieberman posted a thread arguing it is entirely possible to be a great entrepreneur and an even greater parent or partner, saying the founders he once idolized had records of divorce, infidelity and absent parenting. He said he protects wake-up, a morning walk and bedtime with his family. Karri Saarinen replied that founders control their schedules more than most people, and if only the founder can fix every fire, the company may need to be designed differently.