Treasury Secretary Scott Bessent gave Nvidia's board 10 days to say whether it pressed 'seemingly illicit' China sales to 'goose' revenue, and urged it to hire an outside CEO. On the same day, the Financial Times reported that Jensen Huang has discussed a comeback with confidants, and House committee chair John Moolenaar said Huang agreed to a closed-door interview. Bloomberg reported that Anthropic's IPO is about seven times oversubscribed ahead of its expected Nasdaq debut on Thursday. A ULA Vulcan rocket failed 10 minutes into flight, losing 40 Amazon Leo satellites on the first of 38 launches Amazon has booked. The FAA traced Friday's ground stop to two newly installed Juniper routers.
Policy / Regulation2d ago
Bessent gives Nvidia 10 days to answer on China sales, urges outside CEO
On Fox News, Treasury Secretary Scott Bessent said Nvidia needs a 'dramatic shakeup' and should look outside the company for its next CEO, and that he wants answers from the board and outside counsel within 10 days. He asked whether Nvidia pressed 'seemingly illicit sales to goose their revenues and their earnings,' pointing to rising Singapore revenue as China sales fell; Nvidia stopped breaking out Singapore revenue in November 2025. He said that a bank with such compliance failures would see many executives removed, and that no new legislation is needed because 'we have all the rules we need.'
FT: Jensen Huang discusses a comeback as allies form a 'war room'
Jensen Huang has discussed returning as Nvidia CEO with confidants, and an ally is assembling a 'war room' of tech executives to lobby the board and seek White House support, the Financial Times reported, while the board's independent investigation continues. A reinstatement petition has passed 130,000 signatures, according to Tom's Hardware, though Nvidia VP Bryan Catanzaro called a letter attributed to him fake and backed interim CEO Colette Kress. On the All-In podcast, David Friedberg said Huang should return if he is found not to have broken any law. TechCrunch reported that signers of the petition include Elon Musk, Replit CEO Amjad Masad, Epic Games CEO Tim Sweeney and Y Combinator CEO Garry Tan; the petition says Huang 'was removed without meaningful due process.'
Anthropic IPO reportedly 7x oversubscribed ahead of Thursday Nasdaq debut
Anthropic's IPO is about seven times oversubscribed, Bloomberg reported, citing people familiar; the company and some backers are seeking up to $90.4 billion through the IPO and a separate private placement, which Bloomberg says would be the largest IPO ever. Shares are expected to debut on Nasdaq on Thursday at an expected $70 to $78. AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC are collectively investing $26 billion at the IPO price, CNBC reported. The prospectus lists more than $900 billion in spending commitments, including at least $110 billion with Amazon over a decade; Google will own about 13% after the listing.
NYT: Lutnick pushed to narrow chip rules touching his former firm's clients
The New York Times reported that Commerce Secretary Howard Lutnick pushed to narrow chip curbs on Southeast Asian data centers, some of which are clients of Cantor Fitzgerald, the firm he used to run; one client's Malaysia data center may serve Tencent. Sen. Richard Blumenthal said it is 'time for Lutnick to go.' Semafor reported that Lutnick, who has clashed with Bessent over the Nvidia matter, is expected to fly with Trump to China this week; CNBC reported both secretaries are expected on the trip.
Huang agrees to closed-door House interview as Congress weighs chip curbs
Jensen Huang and Nvidia representatives agreed to a voluntary, closed-door interview before three House committees, Select Committee on the CCP chair John Moolenaar told CBS; the committees had invited Huang for December 1 and dates are still being discussed. Senate Banking chair Tim Scott has asked Huang and Lutnick to testify publicly, and House Speaker Mike Johnson cited 'growing consensus in Congress on limiting chip exports to China.' Bloomberg reported that a bipartisan export-control bill could pass within weeks.
WSJ: Megaspeed arranged nearly $12 billion in Nvidia chip orders for China
Megaspeed arranged nearly $12 billion of Nvidia high-end chip orders for Chinese companies between March 2024 and June 2025, covering several hundred thousand chips, according to documents reviewed by the Wall Street Journal. Megaspeed told CNBC that Alice Huang left the company on August 30, 2025, that reports about its business are not supported by verifiable evidence, and that it has hired outside counsel for an internal investigation. Short seller Culper Research estimates China accounted for more than 20% of Nvidia's revenue last fiscal year, versus the 7% Nvidia reports.
ULA's Vulcan fails 10 minutes into flight, losing 40 Amazon Leo satellites
A United Launch Alliance Vulcan rocket carrying 40 Amazon Leo broadband satellites suffered an anomaly 10 minutes after Sunday's liftoff from Cape Canaveral and failed to reach orbit; Ars Technica called it the first total mission failure in ULA's 20-year history. The flight was the first of 38 Vulcan launches Amazon has booked for its Starlink competitor, and Vulcan is also contracted for roughly half of the US military's national security launches in the next few years. ULA said it did not yet have data on the cause; Amazon said it believes ULA will learn from the mission.
FAA traces nationwide ground stop to two new Juniper routers
The FAA found that newly installed HPE Juniper routers at a radio site in El Dorado, Arkansas, and a radar site in Amarillo, Texas, were 'pinging our network at a very high rate, a thousand connections over every four to five seconds,' FAA technology chief Rebecca Guy said, overloading its decades-old copper network. The agency turned off the responsible router feature across the network. An HPE spokesperson said the routers 'did not fail,' while FAA administrator Bryan Bedford said the agency is still asking whether it was a faulty installation, configuration or device.
Nvidia director Mark Stevens sold about $947 million in stock in September
Nvidia board member Mark Stevens sold about 4.2 million shares between August 31 and September 18 for roughly $947 million, without a Rule 10b5-1 trading plan, Tom's Hardware reported, citing Moneywise. The sales came about two months before the Huang scandal broke; Stevens still holds about 27.5 million shares. Selling as a director is not itself illegal, and there is no indication he anticipated the scandal.
Bloomberg: Nvidia lobbied Taiwan after employee was charged with smuggling
Nvidia appealed to Taiwanese officials after prosecutors on the island accused one of its employees of helping smuggle AI chips into China, and sought advice from TSMC, which declined to get involved, Bloomberg reported, citing people familiar with the matter. Seeking Alpha summarized the move as Nvidia defending the employee in the case.
US and Pax Silica allies agree on joint list of AI tech to protect from China
The US and its Pax Silica partners, including Japan, the UK and India, agreed to draw up a joint list of AI technologies and supply chains to protect from China, the Financial Times reported. The pact comes ahead of Trump's meeting with Xi Jinping at the APEC forum in Shenzhen.
US disrupts North Korean 'FaceStealer' campaign that stole over $400 million
The FBI and Justice Department seized cloud infrastructure behind FaceStealer, a North Korean social-engineering campaign that ran for 18 months and stole more than $400 million in cryptocurrency, finding over 900 secretly recorded webcam videos of tech workers. Attackers hijacked Telegram, LinkedIn, Slack and Discord accounts, replayed footage of victims' contacts in fake meetings, then pushed a malicious 'fix'; the FBI says they also compromised a maintainer of a popular JavaScript library. Tom's Hardware links the campaign to the recent Binance heist of more than $5 billion.
Meituan open-sources 1.6-trillion-parameter LongCat-2.5, served on Huawei chips
Meituan released the weights of LongCat-2.5, a mixture-of-experts agent model with 1.6 trillion total and about 48 billion active parameters and a one-million-token context. It scores 88.4 on Terminal Bench v2.1, slightly behind Claude Opus 5, according to Meituan. Its self-hosting guide uses 194 Huawei Ascend accelerators and offers no Nvidia GPU instructions.
Gurman: Apple to unveil smart glasses at WWDC, ship by end of 2027
Apple plans to announce its first smart glasses at WWDC in June and begin shipping before the end of 2027, Bloomberg's Mark Gurman reported. The first model lacks augmented reality and leans on the revamped Siri AI controlling third-party iPhone apps. The N50 prototypes run a pared-down visionOS on a processor derived from Apple Watch chips and have Wi-Fi for some functions away from the phone.
FBI warns attackers are using AI framework ARTEX to automate intrusions
An FBI FLASH advisory says multiple threat actors have used ARTEX, an open-source AI penetration-testing framework written in Go, since at least August 2026, including in a ConnectWise supply-chain breach that hit Microsoft 365 customers. ARTEX plugs into external models including Claude, GPT, DeepSeek, GLM, Grok and Kimi; the FBI says some intrusions went from initial targeting to privileged access in under three hours.
Judge lets DRAM price-fixing suit against Samsung, Micron, SK hynix proceed
Judge Noël Wise denied Micron, Samsung and SK hynix's motion to dismiss a proposed class action alleging they restricted DRAM supply, as prices rose as much as 500% in a year. The 17 plaintiffs cite simultaneous production cuts, a coordinated pivot to HBM and DDR4 wind-downs; the judge wrote that the conduct alleged is 'more akin to hawks flying in formation.' The ruling is not a finding of liability.
OpenAI: GPT-6 Astra led proof for a lattice definition of the Standard Model
In the first guest post on OpenAI's new science blog, UC Santa Barbara physicist Nathaniel Craig, a visiting scientist at OpenAI, described an October preprint proposing a candidate nonperturbative definition of the Standard Model on a lattice. Craig wrote that he had a strategy but could not prove it covered every case, and that GPT-6 Astra turned it into a construction and 'took the lead in developing the detailed proof.' The hard part is the weak force, which treats left- and right-handed particles differently and has resisted lattice formulations.
TSMC plans $500 million fund to back chip startups and suppliers
TSMC plans a $500 million venture fund for chip startups, suppliers and other upstream partners, senior vice president Kevin Zhang told Nikkei Asia. 'We not only support today's Nvidia, but we also want to support the future's Nvidia,' he said.
Apple launches US study of how people use health AI, with OpenEvidence
Apple launched the Apple Health AI Usage Study in the US, a virtual study with its own app built with clinical AI company OpenEvidence. Participants can ask everyday health questions, see peer-reviewed citations behind each answer, and optionally share HealthKit data such as resting heart rate, sleep or lab results. Apple said the study will examine trust, answer quality and comprehension to inform how it and the industry design health AI.
TSMC confirms employee data leak traced to a cloud provider
TSMC told the SEC that data on current and former employees was posted online without authorization and matched data held in a cloud service provider's systems, adding it has found no evidence its own systems were compromised. A seller claimed a 6.9GB set of 46,000 employee records, including national ID numbers and bank details, and opened bidding at 3 BTC.
All-In's Sacks and Calacanis question who leaked the Huang intelligence
In new All-In clips, David Sacks said there were 'two ops': whatever Alice Huang was doing, and a second that began when someone leaked a classified memo's contents to the New York Times, which he noted is a crime. He asked why Huang got no defensive briefing, what national security secrets Huang could have held, and what evidence shows Alice Huang was a spy, citing the incoming Democratic chair of the House Intelligence Committee saying he was not sure. Jason Calacanis said he expects the leak to trace to warring China factions inside the administration.
Grokbot team: one engineer now ships about 4,000 PRs a month
On YC's Lightcone podcast, Grokbot builders Lauren Tan and Roman Ugarte said the product, which gives AI teammates their own identity, phone number and credit card, went from first line of code to launch in about seven weeks. Tan's output has grown from about 2,000 pull requests a month to over 4,000. Ugarte said the bottleneck is shifting from engineering time to team alignment on what to build.
Microsoft 365 outage tied to chlorine gas exposure at German hotel
Fifteen guests at a hotel in Germany suffered chlorine gas poisoning after the sauna's safety system, which relied on Outlook email to alert operators to high chlorine levels, failed during Microsoft's November 12–14 outage, Tom's Hardware reported, citing Notebookcheck. Forty first responders were deployed and four people were taken to hospital.
SK Hynix to spend about $1.05B over five years on domestic chip suppliers
SK Hynix said it plans to spend 1.4 trillion won, about $1.05 billion, over the next five years to support South Korean suppliers of chip materials, parts and equipment, Reuters reported. The company said the spending is meant to strengthen its supply chain.
Claude Fable 5.2 paints images with up to 700,000 simulated brushstrokes
Anthropic's Jake Eaton showed Claude Fable 5.2 producing a near-photographic parking-garage scene from roughly 700,000 simulated brushstrokes, with no external image model. Each piece is a Python program of 1,000 to 2,000 lines driving a simulated paint engine with bristle texture and wet paint, The Decoder reported.
Intel regains x86 share from AMD for first time since 2025
Intel's overall x86 unit share rose to 69.9% in the third quarter versus AMD's 30.1%, the first quarterly gain for Intel since Q2 2025, driven by desktop, according to Mercury Research. AMD set records in mobile and server, where it reached 35.5%. Server CPU unit sales grew 20% year over year while client shipments fell on high component prices.
FT: Samsung IT arm in early talks to host Anthropic and OpenAI
Samsung wants to become a one-stop shop that builds, runs and powers data centers, the Financial Times reported, and its IT arm is in early talks to sign Anthropic and OpenAI as tenants, according to people familiar. The group hopes to repeat its semiconductor success across its other businesses.
Epoch AI: Huawei's largest AI data center has ~7% of Colossus 2's compute
An Epoch AI analysis estimates that Huawei's Horinger data center, among China's largest for AI, has about 7% of the compute of SpaceX's Colossus 2. Epoch says the gap may persist until at least 2028, even as Huawei draws on China's abundant energy to build data centers.
Nidus, which builds AI-driven robots for US defense manufacturers' tasks such as wire harnesses, raised $30 million across pre-seed and seed rounds led by Kleiner Perkins and says it reached seven-figure bookings in four months. Nebuly, which measures the business value of enterprise AI agents, raised a $21 million Series A led by Planven. Indian chip startup Bharat Semi raised a $24 million seed round led by Info Edge Group, South Park Commons and Indus Bridge Ventures to build image sensors that combine capture, processing and AI on one chip; the pre-revenue company says it has design wins with General Atomics and Kaynes Semicon and targets initial commercial production in 2027.
The Information: Modal taps 30-plus small clouds amid server shortage
Server shortages are severe enough that inference platform Modal now draws on more than 30 small cloud providers, many of them young startups without the security expertise of Microsoft, Amazon or Google, The Information reported.
Australia's central bank sees AI hitting young workers first
Analysis from the Reserve Bank of Australia finds that since ChatGPT's launch, job postings and employment for 15- to 24-year-olds in highly exposed fields such as software development and customer service have fallen more sharply than for older workers. Author Mike Major cautioned that business-cycle shifts and remote-work trends could also explain the numbers.
Microsoft raises India cloud prices 15.7% from January, cuts China prices
Microsoft will raise commercial cloud prices in India by 15.7% from January 1, 2027, and some on-prem offerings by up to 20%, The Register reported; New Zealand prices rise 10%. Prices fall by up to 8.5% in China, where 21Vianet operates its cloud, and by 4.5% in Australia, reflecting exchange-rate adjustments.
Meta employees eye February 2027 'Rapture' as big 2023 stock grants finish vesting
February 15, 2027 marks the final payout of a batch of Meta stock awards granted in 2023, when shares were depressed; the stock has risen more than 400% since early 2023, Business Insider reported. Former employee Andre Nader estimates senior engineers' annual stock compensation could fall more than 40% afterward, which some expect to trigger departures and early retirements.
What X was arguing about today that the news hasn't covered. Views are the posters' own.
Paul Graham says founder ability, not capital, is the limiting factor
Y Combinator co-founder Paul Graham wrote that people who solve hard problems for decades are told 'anyone could have done it,' and that as an investor 'the limiting factor isn't investment. It's founder ability.' Elon Musk replied 'Yes.' The post landed amid the debate over whether Nvidia can be run without its founder.
dax says teammates' agents now answer Slack for them
dax, who builds OpenCode, wrote that an unreleased console feature lets agents receive Slack and GitHub events, and many teammates have told their agents to reply to Slack messages, so he can talk to 'a ghost version' of colleagues. In a follow-up he asked readers to consider whether the exchange was worth having a human involved at all.
Chris Barber polls 56 AI insiders: Terminal-Bench 4 is the most trusted frontier benchmark
Chris Barber asked 56 people in AI which benchmarks they would most trust to show an LLM is frontier. Terminal-Bench 4 led with 25 votes, followed by Terminal-Bench Science with 20 and FrontierCode with 13. The poll ranks benchmarks by insider trust rather than by any model's score.
MIRI's Nate Soares says it is more likely than not that we make it to 2028
Nate Soares, president of the Machine Intelligence Research Institute, wrote that he is not as worried about short AI timelines as some others, and thinks it is more likely than not that we make it to 2028. One reply countered that the speed of AI progress has repeatedly surprised even experts.
Ming-Chi Kuo says OpenAI and Anthropic orders extend Google TPU visibility to 2029
Analyst Ming-Chi Kuo wrote that his supply-chain survey puts combined 2028–2029 TPU demand at about 12 million units for OpenAI and 14 million for Anthropic, with OpenAI buying outright and Anthropic mixing leases and purchases. He estimated TPU shipments of 8–9 million, 14–15 million and 16–18 million units in 2027–2029, still growing in 2029 although the market broadly expects AI capex to peak in 2028. He named MediaTek the biggest winner as sole design supplier from v9 on, and said uncertainty over Nvidia's CEO succession helps Google in long-term talks with customers.