Silicon Valley Daily
At 3:44 p.m. ET on Friday the FAA halted every departure in the US, lifting the stop only at 8:15 p.m., the third nationwide ground stop in its history. Transportation Secretary Sean Duffy blamed old routers overloaded with data, CBS sources pointed to an update that crashed the air traffic communications network, and the FBI does not believe it was a cyberattack. Nvidia fell 2% to close below a $4 trillion market value, capping its longest weekly losing streak since 2012, while its board said it has not yet picked a search firm or engaged any candidates. Bloomberg traced banned Nvidia chips to a Shanghai data center tied to Alice Huang, and Semafor detailed how Nvidia's general counsel invoked Trump to speed China export licenses. Elsewhere, OpenAI launched ChatGPT Wallet, Anthropic's IPO shut out mainland China and Hong Kong institutions, and Starship reached orbit on its 15th flight.
FAA halts every US departure for over four hours after air traffic network failure
The FAA ordered a nationwide ground stop at 3:44 p.m. ET on Friday after communication problems hit multiple Air Route Traffic Control Centers, and lifted it at 8:15 p.m. It was only the third nationwide stop in the agency's history, after 9/11 and a 2023 NOTAM outage; FlightAware counted more than 6,300 cancellations and 9,500 delays. Transportation Secretary Sean Duffy blamed old switches and routers overloaded with data, Homeland Security Secretary Markwayne Mullin said patches made while installing new equipment overwhelmed the older system, and CBS sources said an update to the L3Harris-run communications network caused it to reboot and crash. Bloomberg reports the FBI is probing whether a software update was the cause and does not believe it was a cyberattack.
Bloomberg: banned Nvidia chips reached a Shanghai data center tied to Alice Huang
Bloomberg reports that a company Alice Huang founded, now majority-owned by her sister, listed 52 Super Micro units with Nvidia H100 or H200 chips, which cannot be sold to China without U.S. approval. A Singapore Nvidia cloud partner separately bought 22 Dell servers with H800 chips, and the hardware went to a Shanghai facility run by her husband, Kang Wei. China on Friday warned Washington against imposing new AI chip export controls.
Nvidia board says CEO search has barely begun after report it approached Arm's Haas
The Financial Times reported that Nvidia's board sounded out Arm CEO Rene Haas, who spent seven years at Nvidia before joining Arm in 2013, about succeeding Jensen Huang. In a Friday statement to CNBC, the board said it is 'in the process of selecting a search firm,' has 'not formally engaged any potential candidates,' and called speculation about individual candidates premature. Colette Kress held her first meeting as interim CEO with senior executives on Thursday.
WSJ: Huang called Kress a 'deer in the headlights' over chip guarantee pushback
The Wall Street Journal reports that in an August finance meeting, Huang described Colette Kress, now interim CEO, as 'a deer in the headlights' and said she could not help steer strategy. The meeting concerned billions in Nvidia guarantees for data center operators that could not pay for chips up front, including deals involving Huang's son-in-law Nico Caprez. Kress had questioned how to value those guarantees and earlier cut Nvidia's credit commitment to OpenAI's Ohio project from a discussed $250 billion to $105 billion.
OpenAI launches ChatGPT Wallet so its dots agents can make purchases
OpenAI launched ChatGPT Wallet, letting its dots agents complete purchases for a limited group of U.S. ChatGPT Pro subscribers aged 18 and up. Built with American Express, Mastercard and Visa, it charges cards through transaction-specific payment tokens, so neither the agent nor the merchant sees card details, and each purchase is approved with Face ID or a fingerprint. OpenAI says it is not taking a cut of transactions at launch and plans to add spending limits for more autonomous purchases later.
Anthropic IPO shuts out mainland China and Hong Kong institutions; pricing set for Nov. 18
Bloomberg reports that banks are rejecting or deprioritizing orders from institutional investors in mainland China and Hong Kong for Anthropic's IPO, with some citing U.S. ITAR export rules, as was the case for SpaceX. Goldman Sachs and Morgan Stanley lead the deal, which is expected to price on Nov. 18 and trade on Nasdaq as ANT. The Information reports Anthropic is targeting a $1.25 trillion to $1.4 trillion valuation.
Microsoft apologizes for Microsoft 365 outage, blames network capacity upgrade
Microsoft apologized on Friday for an outage that took down Outlook, Teams, Copilot and other Microsoft 365 services for nearly 24 hours across more than 70 countries. The company said the incident was introduced during an earlier network capacity upgrade, surfaced after a recent network change, and was not related to a cyberattack or AI agents. It extended email timeouts to 48 hours to preserve delayed mail and promised a full post-incident report.
Nvidia posts sixth straight weekly loss, longest streak since 2012
Nvidia fell 2.01% on Friday and closed below a $4 trillion market value for the first time since July 2025. The stock lost 12.75% this week, its sixth straight weekly decline and longest losing streak since 2012; over those six weeks it has shed about 30%, or roughly $1.7 trillion, and fallen behind Apple. Nvidia reports quarterly results on Tuesday, where analysts expect revenue of $108.9 billion, up 91% from a year earlier, according to LSEG.
Commerce readies rule on overseas data centers as Bessent says Treasury wanted Megaspeed blacklisted
The Information reports that the Commerce Department is preparing to close a loophole that lets Chinese AI firms reach restricted Nvidia chips through overseas data centers. Treasury Secretary Scott Bessent said Treasury, Defense and State had for months favored adding Megaspeed to the entity list, but that Commerce is in charge. Al Jazeera reports China has warned the U.S. against further AI chip controls.
Semafor: Nvidia's top lawyer invoked Trump to speed China chip licenses
Semafor reports that in January Nvidia general counsel Tim Teter told more than a dozen officials, 'I was with the president when he and Jensen discussed this... He wants this to go fast,' while pushing back on conditions for H200 export licenses to China. Teter called a requirement to know cloud customers a 'poison pill' and pushed for know-your-customer checks to start at tens of thousands of chips; a former official described a 'scorched earth, adversarial attitude.' The Bureau of Industry and Security said such consultations were 'completely routine.' Separately, Reuters reports, citing sources and messages, that Huang misled several Nvidia leaders, including Teter, about the nature of his relationship with Alice Huang.
Huang's exit unsettles Nvidia's $500B financing vehicle and startup portfolio
The New York Times reports that Nvidia's planned $500 billion AI lending consortium is in flux and at least one lender may exit, while Bloomberg says Nvidia was already under pressure to raise its 25% residual value backstop before Huang left. The Information reports Nvidia plans to invest $10 billion in Anthropic's IPO.
CNBC: Cantor cut its Nvidia holdings by half as Lutnick faced smuggling pressure
A CNBC analysis of filings found Cantor Fitzgerald, the firm Commerce Secretary Howard Lutnick formerly chaired and his sons now run, cut its reported Nvidia share count by nearly half between September 2025 and March 2026. That period overlaps with Axios reporting that officials pressed Lutnick to act on Megaspeed, which he resisted. CNBC found no evidence the firm's interests influenced his decisions; Cantor said its proprietary Nvidia positions were immaterial.
Amodei calls compute planning Anthropic's 'deepest business dilemma'; Lilly CEO doubts cure timeline
Bloomberg reports Dario Amodei said deciding how much computing power Anthropic will need in a year or two is its deepest business dilemma because revenue is hard to predict. His IPO-linked prediction that AI could cure cancer, Alzheimer's and other diseases within a decade drew pushback from Eli Lilly CEO Dave Ricks, who wrote that it would be amazing but is not likely, and that progress will likely be steady and incremental.
Duffy raises air traffic modernization ask to $50B; AI tool maker ASI denies role
Before the outage, Duffy had asked Congress for $30.1 billion to upgrade air traffic control; after it he said, 'I want $50 billion.' The failed FTI network is midway through a fiber upgrade by L3Harris, and a September GAO report warned the FAA may be prioritizing speed over cost and disruption. Air Space Intelligence, whose AI traffic-flow tool SMART went nationwide on Nov. 10 under an $875 million contract, said the ground stop 'is not based on or a result of' its system; Mullin said officials were looking into it.
Nscale filing shows $5.5B of debt and a bigger Nvidia financing role
Nscale's amended S-1 shows its debt rose to $5.5 billion in September from $137.4 million in June, and Nvidia committed $2 billion of a convertible financing of more than $3 billion. The data center developer said it needs more funding for 2027 commitments after ordering Nvidia chips for a $45 billion Anthropic deal, a disclosure that raises questions about its planned $35 billion IPO.
VCs brace for smaller Nvidia checks after Huang's exit
Nvidia joined 105 funding rounds this year, often as the largest investor, according to PitchBook data cited by TechCrunch, and held $48 billion of stakes in 229 private companies at the end of July, per the New York Times. Investors told TechCrunch that interim CEO Colette Kress is 'much more conservative' and expect smaller checks and lower valuations, particularly for neoclouds that raised half their GPU budgets from Nvidia. 'Nvidia is the central bank of A.I. and all of a sudden, the Fed chairman is out,' Theory Ventures' Tomasz Tunguz told the Times.
Reuters: Amazon plans a 'World Product Graph' of nearly every product sold online
An internal Amazon document seen by Reuters describes a project to gather data on nearly every product sold anywhere, including items Amazon does not sell, possibly by crawling rivals' sites. The aim is to keep shoppers and AI shopping agents relying on Amazon's index rather than Reddit or Google, and to fix its assistant, which the document says gives answers that are 'factually wrong or incomplete.' The document lists an Oct. 1, 2027 release date; Amazon said brainstorming documents do not necessarily reflect current plans.
Investors weigh Nvidia without Huang; Databricks' Ghodsi calls it 'a blow to everything'
At Newcomer's Cerebral Valley event, Databricks CEO Ali Ghodsi said the crisis is 'a blow to everything, every player, except their direct competitors,' while General Catalyst CEO Hemant Taneja praised the board for upholding its fiduciary duty. Huang biographer Stephen Witt wrote in the New York Times that Nvidia had a culture of deference that let Huang 'be his worst self.' In Israel, where Nvidia employs 6,000 people, startups are watching whether the next CEO keeps its commitment there.
Trump voices sympathy for Huang and again floats Tim Cook for Nvidia
Forbes reports President Trump expressed sympathy for Huang after his resignation and again promoted Apple CEO Tim Cook as a possible Nvidia chief. NBC News reports some commentators in China are questioning the timing of the reports about Huang and Alice Huang, with Trump due to meet Xi Jinping in Shenzhen next week.
LangChain raises $300M Series C at a $3B valuation
LangChain raised a $300 million Series C at a $3 billion valuation led by CapitalG, with GIC, DFJ Growth, IVP and Sapphire Ventures participating. The valuation is more than double the $1.25 billion it reached about a year ago. LangChain says its agent management platform LangSmith has 7,000 customers, including Harvey, Rippling, Nvidia, Toyota and Salesforce.
OpenAI disbands its standalone robotics team
OpenAI has disbanded its standalone robotics team and moved its researchers into research and Applied units, Fortune reports. The company said general models such as GPT-6 Astra can be applied directly to robotics.
SpaceX's Starship reaches orbit on flight 15 and deploys 26 Starlink satellites
Starship lifted off from Starbase, Texas, on Friday, reached orbit for the second time and released 26 Starlink V3 satellites about 40 minutes into the flight. The Super Heavy booster was diverted into the Gulf over an unspecified issue instead of returning to be caught by the tower. SpaceX may attempt its first-ever catch of the upper stage on Saturday if the vehicle passes health checks.
SpaceX cuts most of the Grok Imagine team as Grok Bot learns to join meetings
The Information reports SpaceX's AI group laid off about 60 Grok Imagine staff in the past week, leaving a handful to maintain the image and video tool as it prioritizes coding tools and Grok Bot. On Friday Grok Bot gained the ability to join meetings, answer questions and track decisions; its Mac app can now use Messages, Mail, Notes, Calendar and Contacts.
Google lets Gemini read Google Wallet spending data for budgeting
Google added a Google Wallet connector to the Gemini app in the U.S., letting users ask about spending, recurring subscriptions and budgets, with charts generated in the chat. Accounts are linked through Plaid, and the integration is read-only: Gemini cannot move, manage or invest money. The launch came the same day OpenAI let its agents make purchases.
Claude helps decode Gödel's notebooks, revealing hundreds of unknown theorems
The Guardian reports that Helsinki logician Jan von Plato used Claude to digest 1,009 pages of Kurt Gödel's 1940s workbooks, much of it in an obsolete German shorthand. Manual progress had run about a page a week; within a couple of days Claude produced a digest showing Gödel had proved hundreds of theorems no one knew about. Von Plato said the notebooks contain 14 themes in logic that have never been pursued.
United Airlines restricts third-party AI agents from booking flights
United Airlines said it will restrict unauthorized third-party AI agents from booking and managing reservations, citing cybersecurity risks and federal rules that require a human to complete steps such as check-in declarations. The airline warned that sharing credentials with AI assistants can lead to unintended purchases or use of MileagePlus miles. Bank of America last week urged customers not to share passwords with AI assistants.
WSJ: Moonshot AI dissolves Cayman structure to prepare for an IPO
The Wall Street Journal reports, citing sources, that Chinese AI startup Moonshot AI is deactivating its Cayman Islands holding company and moving investors' stakes into its main operating entity in China. The restructuring is meant to pave the way for an initial public offering.
Dwarkesh Patel now expects AI to match top AI researchers within a couple of years
In a new essay, podcaster Dwarkesh Patel says he now believes AI is only a couple of years, possibly a couple of months, from matching the best humans at AI research, and expects AI at least as good as top experts in every field by 2030. He cites METR's finding that Anthropic's progress rate on well-scoped tasks has doubled in a year, and Anthropic's report that Claude 'leads' more than 26% of its AI R&D work, up from under 1% in February. His argument rests on how fast AI advanced in mathematics, another domain with verifiable answers.
Uber says Fireworks AI breach exposed drivers' license data
Uber is notifying drivers that personal and driver's license data was stolen in an October breach at a vendor, which TechCrunch identified and Uber confirmed as Fireworks AI. Notices filed with state attorneys general show 2,567 affected individuals in Massachusetts alone. Notion has also confirmed its data was taken in the Fireworks hack.
Wiz CTO Ami Luttwak: AI cut attack timelines from months to about 20 minutes
On the a16z podcast, Wiz co-founder and CTO Ami Luttwak described 'Black October' as a wave in which models became good enough to find and exploit complex vulnerabilities, running attacks that took the best human attackers weeks or months in about 20 minutes. He said a PHP flaw touched almost all Wiz customers and roughly half patched within two to three days, and urged new companies to build systems that can patch automatically.
Researcher drops 'SignDrop' Windows zero-day granting System rights
A researcher using the name Nightmare Eclipse disclosed SignDrop, a Windows zero-day in the Common Log File System that lets a local user gain System rights on some Windows versions. The exploit tricks Windows Defender into opening a crafted CLFS container, abusing a mitigation Microsoft added in October 2025.
OpenAI Foundation board member Paul Christiano sees 'meaningful risk' of AI takeover within a year
Paul Christiano, an OpenAI Foundation board member and head of the Alignment Research Center, told a Toronto audience that for most people in the room the most likely cause of death is an AI accident. He said that on the fastest trajectory humanity faces 'something like a coin flip' of AI takeover, including a meaningful risk in the next six to 12 months, while conceding there is no quantitative model behind the estimate.
Mike Volpi: AI build-out differs from the fiber bubble on asset life and demand
On the a16z podcast, Mike Volpi, who helped shape Cisco strategy during the internet build-out, compared the AI build-out with the 1990s bandwidth boom. He said fiber stayed usable for 20 to 30 years and outlived the companies that went bankrupt laying it, while the useful life of GPUs is the key unknown now. He argued AI's practical applicability is higher than bandwidth's was, citing the revenue ramps at Anthropic, OpenAI and open-model hosts.
WSJ: Meta staff are testing Takes, a prediction-market app using points
Meta employees have begun testing a prediction-market app currently called Takes, the Wall Street Journal reports. Users wager on the outcome of questions and events with 'cred' rather than real money.
Nvidia releases GR00T 2 open robotics model
Nvidia introduced GR00T 2, which it calls its most advanced open System One model for robotics research. The model learns skills from human demonstrations, predicts how its actions will change the physical world, and can carry skills across different robot bodies.
ElevenLabs previews ElevenAI, a team note-taker and dictation app
ElevenLabs, valued at $22 billion, previewed ElevenAI, an app for teams that combines meeting notes, cross-meeting search, dictation in more than 90 languages and follow-up actions through its ElevenAgent platform. Access is through an early-access waitlist.
Claude services hit by login errors for about two hours
Anthropic reported elevated errors affecting sign-in to claude.ai, Claude Code, the Claude Console and Claude Cowork on Friday. The incident opened at 09:58 UTC, a fix was deployed at 11:05 UTC, and it was marked resolved at 11:58 UTC.
Anthropic publishes 10 recommendations for safety evaluations
Anthropic published 10 recommendations on its engineering blog for designing, running and reporting safety evaluations. The company says such evaluations determine whether a model is ready to deploy, which safeguards to build and where to focus training and research.
$11.4 million of Nvidia hardware vanishes on a truck from Oakland to Texas
Six pallets of Nvidia hardware reportedly worth $11.4 million disappeared after a truck bearing a food transport company's logo collected them in Oakland on Aug. 29 for delivery to Fort Worth, FreightWaves reports. Investigators lost the truck in San Bernardino County; Oakland police confirmed a grand theft investigation but have not confirmed the value or the cargo.
Paytm wins tentative approval to revive its digital wallet
Bloomberg reports India's central bank gave Paytm tentative approval to operate a digital wallet. The business had effectively ended after the regulator canceled Paytm Payments Bank's license in April.
Trending on X
What X was arguing about today that the news hasn't covered. Views are the posters' own.
IFP's Tim Fist says his team partly reproduced Will DePue's Blackwell fingerprint findings
Former OpenAI researcher Will DePue said he searched more than six trillion chip-and-seed combinations and matched six Chinese models, including DeepSeek-V4-Pro and Kimi K3, to Blackwell hardware; he said a B300 reproduced DeepSeek-V4-Flash's starting weights bit for bit. Tim Fist, president of the Institute for Progress, said his team reproduced the patterns for three models on B200 and H200 chips. Nvidia researcher Jean-François Puget replied that the finding does not show the GPUs were in China, since Chinese firms rent compute abroad.
Tim Sweeney says WSJ spy coverage came from a rival AI lobbyist, not officials
A WSJ opinion piece said Huang's resignation was necessary but harmful to Nvidia and America's AI industry, adding that this may have been China's plan. Epic Games CEO Tim Sweeney replied that the paper had created the impression senior officials supplied the spying information anonymously, when it came from a lobbyist at a competing AI company. Sweeney did not name the company.
Uploader of abliterated models says a push to remove 6,500 from Hugging Face would change nothing
Transformer News reported that Hugging Face, which Nvidia agreed to acquire, hosts thousands of uncensored models that could aid cyberattacks. An uploader of abliterated Qwen and GLM models said one of his got about 358,000 downloads in 30 days, that abliteration only removes the refusal direction, and that base weights are already on millions of drives. Brian Roemmele called it 'the great culling' of open models; Venice.ai founder Erik Voorhees mocked the story by comparing it to Best Buy selling 'uncensored laptops.'
John Crickett predicts CS fundamentals will outvalue framework knowledge in 3 to 5 years
Engineering educator John Crickett predicted that over the next three to five years, algorithms, networking, databases, architecture and security will matter more than knowing the latest frameworks, because a coding agent can build a solution but engineers must judge trade-offs and spot when it is wrong. Boot.dev founder Lane Wagner agreed, saying framework minutiae are exactly what LLMs handle well; one reply said Claude already knows all of this.