Silicon Valley Daily
SpaceX caught a Starship upper stage with its launch-tower arms for the first time on Saturday, after the ship circled Earth 14 times. On the ground, the FAA said the root cause of Friday's nationwide ground stop is still unknown, even as Transportation Secretary Sean Duffy pointed to L3Harris fiber work; more than 11,000 flights were cancelled across two days. At Nvidia, two holders with about $2.1 billion of stock asked the board to consider Jensen Huang's return as a petition passed 115,000 signatures, while CNN and the Washington Post reported that the spy allegation behind his exit rests on contested intelligence. Anthropic reported a $567 million third-quarter loss and nearly $901 billion in commitments days before its IPO, DeepSeek raised at a $92 billion valuation, and Italy fined Meta €805 million.
FAA says ground stop's root cause is still unknown; Duffy points to L3Harris fiber work
Transportation Secretary Sean Duffy said Friday's nationwide ground stop followed copper-to-fiber work by contractor L3Harris at a remote FAA site, after which a new router sent more than 1,000 connection attempts every four to five seconds. At the same Saturday briefing, FAA administrator Bryan Bedford said the root cause has not been identified and that the agency grounded flights because 'we couldn't explain the instability'; L3Harris said it has completed more than 70% of the conversion and is supporting the investigation. Cirium counted more than 8,200 cancellations Friday evening and at least 3,000 more on Saturday.
Two big Nvidia holders ask board to weigh Huang's return; petition passes 115,000
Altimeter's Brad Gerstner and Atreides' Gavin Baker, whose firms held about $2.1 billion of Nvidia stock as of June 30, urged the board to run a 'slow, measured, and deliberate' search that considers 'the potential return of Jensen Huang.' A 'Stand with Jensen' petition had more than 115,000 signatures by Saturday afternoon, according to TechCrunch, including Elon Musk and Replit CEO Amjad Masad; organizers say more than 5,600 current employees signed and that they are removing fake entries.
SpaceX catches Starship's upper stage with its launch tower for the first time
SpaceX caught the upper stage of Starship with the 'chopstick' arms of its Starbase launch tower on Saturday, the first time the maneuver has worked on the ship itself. The booster from Test Flight 15 was not caught; it was diverted to the Gulf of Mexico and exploded before reaching the water, while the ship orbited Earth 14 times and released 26 Starlink v3 satellites. Ars Technica reports SpaceX can now build a Starship upper stage every three weeks, and full reuse is central to its Starlink and NASA lunar plans.
Reports from CNN and the Post cast doubt on the spy claim that ended Huang's tenure
CNN reported Thursday that U.S. intelligence agencies have not established that Alice Huang was spying for China, and that some CIA officials are unconvinced intercepted communications show she knowingly passed information to Beijing. The Washington Post reported that Secretary of State Marco Rubio used a less conclusive CIA report to write his own memo to President Trump in more assertive terms, and that its leak set off Huang's exit. TechCrunch notes Nvidia's longtime security chief David Reber left in August for health reasons; nothing public ties the change to the Megaspeed matter.
Nvidia's handshake deals and neocloud guarantees face review under Kress
Elon Musk wrote that many of Nvidia's supply-chain deals 'were a handshake with Jensen, not a written contract' and that some will be lost. TechSpot, citing a Wall Street Journal profile of interim CEO Colette Kress, reports Nvidia pushed for more customer-financing deals in 2026 to absorb an oversupply of TSMC wafers, then slowed, shrank or paused some, including an OpenAI project that originally carried a $250 billion backstop. The Information says Kress must also decide the fate of guarantees such as nearly $20 billion of Lambda leases.
Bessent: Nvidia must answer for sales to 'pass-through shell companies'
Treasury Secretary Scott Bessent told CNBC the real story is whether Nvidia knew its customers and sold to 'obviously pass-through shell companies,' adding that anyone responsible should be held accountable, while saying it is not known that Huang knew. He criticized general counsel Timothy Teter's contacts with Commerce. Axios reports Republicans are now targeting Commerce Secretary Howard Lutnick over his slow handling of Megaspeed; CNBC says he will testify to the Senate Banking Committee on Dec. 10.
Anthropic posts $567M Q3 operating loss; Amodei calls compute its 'deepest dilemma'
The Information reports Anthropic's $567 million third-quarter operating loss came mostly from short-term computing costs and $2 billion in matched employee donations. At a JPMorgan event Friday, Dario Amodei said 'compute is the deepest business dilemma' the company faces because it must be bought one to two years ahead. Bloomberg puts Anthropic's commitments at nearly $901 billion, with about $71 billion due by September 2027; the company plans to raise up to $64.4 billion in the IPO plus $26 billion in a private placement.
Italy fines Meta €805 million over Facebook's access to Android photos
Italy's data protection authority fined Meta €805 million ($887 million), saying the Facebook Android app accessed photos and videos of about 20 million users for post suggestions without specific consent. Meta has 60 days to switch the feature off by default and seek separate consent. Meta said it 'strongly' disagrees and called the fine disproportionate.
The Information: DeepSeek raises at a $92 billion valuation
The Information reports DeepSeek's funding round at a $92 billion valuation drew Tencent and battery maker CATL. It says the round also included a group of companies pitched as a major source of black-market Nvidia chips.
Cursor confirms breach that let attackers push a fake Claude Code extension
Cursor said attackers used an access key taken from Fireworks AI to copy historical coding-agent traces, some holding unredacted credentials, then used one credential to plant malicious copies of Anthropic's Claude Code extension in its marketplace cache on Oct. 29. Versions 2.1.312 and 2.1.313 stayed up about 11 hours and exposed up to 5,000 developers to malware that installed persistent SSH backdoors. Cursor says fewer than 0.05% of users had traces exposed.
Researchers say weight fingerprints tie DeepSeek and Kimi models to Nvidia Blackwell chips
Independent group UnemploymentLabs says the initial weight distributions of open models including DeepSeek V4, Kimi K3.1, MiniMax-M3 and Ling-3.1-Flash match what Nvidia Blackwell chips produce, because the random initialization depends on each chip's streaming-multiprocessor count. Changing the SM and thread counts broke the correlation. The researchers say the method cannot show where the chips were or how they were accessed, and Tom's Hardware notes renting Blackwell hardware abroad is not currently illegal for Chinese firms.
ByteDance offers up to $6 billion buyback at a 20% higher valuation
The Information reports that with TikTok's U.S. future settled, ByteDance is offering investors up to $6 billion in share buybacks at a valuation 20% above its previous level. The report says the move could help clear the way for a future IPO.
Tesla opens personal Cybercab orders at $29,990 plus $299 a month
Tesla invited select customers in its Robotaxi cities to buy a Cybercab for personal use at $29,990 plus a mandatory $299-a-month Autonomy Subscription. The car has no manual controls, so cancelling the subscription leaves it unusable, and the terms bar commercial use. Electrek reports the agreement limits the car to Tesla geofences that the company can change at any time.
Commerce moves to close remote-access chip loophole; Gavin Baker warns against a total ban
The Information reports the Commerce Department is preparing to close a loophole that lets Chinese AI firms use restricted Nvidia chips through overseas data centers. Atreides' Gavin Baker argued on X that a complete ban on GPU exports and Southeast Asian cloud access could damage U.S. security by forcing Chinese labs onto Huawei hardware. He cited estimates that the U.S.-to-China AI compute ratio grew from 2.8x in 2023 to 8.8x in 2026.
Asian governments react: Korea says Nvidia reaffirmed 260,000-GPU supply
South Korea's Science Ministry said Nvidia reaffirmed plans to supply 260,000 advanced GPUs, and the country's AI strategy council held an emergency meeting. Singapore said its probe into Megaspeed and six other people remains open, and China's foreign ministry said it was unaware of reports linking restricted chips to a Shanghai data center. The Financial Times reports that despite two decades in China, Huang struggled to reach the policymakers behind Beijing's chip self-sufficiency push.
All-In hosts weigh shareholder suits, China gains and Nvidia's successor field
On the All-In podcast, one host said Nvidia lost almost a trillion dollars of value in about nine days and that shareholder lawsuits will be 'ginormous' even if Huang is cleared. David Sacks said China benefits most, arguing Nvidia's edge is taping out a much faster chip every year and that losing Huang will let Huawei 'catch up a lot faster.' David Friedberg went further, saying he worries the new Congress will require a kill switch on every AI chip and that Huang 'had to be taken out' if the state wants control of AI; he offered no evidence.
Lovable apps left passports and 200,000 private messages exposed
Crissyfield and Handelsblatt examined 159,041 Lovable projects and found that 7,403 of 24,664 sites loading Supabase data exposed tables containing email addresses without login. Exposed records included over 1,000 passport entries and more than 200,000 chat messages from a German forum. The researchers say the 30% figure is not a confirmed leak rate and that they could not tell whether the data was misused.
Sanders urges Democrats to pass a data-center moratorium and ban superintelligence
Sen. Bernie Sanders wrote that Democrats, newly empowered after the midterms, must enact a national moratorium on AI data centers, pause advanced AI development and ban superintelligence. He said the moratorium is supported by 63% of Americans. Axios reports voters rejected data centers in 40 of 41 ballot measures this election.
Malaysia plans law tracking data-center power to catch GPU smugglers
Malaysia plans to require data center operators to disclose AI accelerator imports and power consumption to stop restricted chips being re-exported to countries such as China, and to charge for reserved but unused electricity capacity. Trade minister Johari Abdul Ghani acknowledged that some firms rented local data center space to import GPUs legally and then shipped them abroad.
Eric Trump-backed robot maker Foundation misses payroll
CNBC reports Foundation Future Industries, a San Francisco maker of military humanoid robots, missed payroll on Oct. 30 and workers remain unpaid. CEO Sankaet Pathak borrowed $4.6 million from the company last year to buy a Sausalito home, and is seeking funding at a $3 billion valuation. Pathak's previous startup Synapse collapsed in 2024.
Bloomberg: Applied Digital has sought billions in guarantees from Oracle and Meta
Bloomberg reports, citing a source, that data-center developer Applied Digital has asked tenants such as Oracle and Meta for billions of dollars in guarantees since at least last year to secure grid power before finalizing leases. The report says tech companies chasing AI capacity are increasingly on the hook for the cost of electricity infrastructure.
Databento raises $100M Series C at $1.1B valuation
Databento, which sells market data through APIs, raised a $100 million Series C at a $1.1 billion valuation led by Fifth Down Capital, four months after a $97 million Series B. The company says it has contracts with five of the six largest frontier AI labs and that revenue has more than doubled since the last round.
Gemini 4 Carbon tops MathArena at 15 times GPT-6.1 Sol's cost
Google's Gemini 4 Carbon displaced GPT-6.1 Sol at the top of the MathArena math benchmark, with Anthropic's Fable 5.2 second. At the reasoning settings tested, Carbon and Fable 5.2 cost roughly 15 and 40 times as much as Sol. Carbon's most common failure was exceeding the time limit; in one case it ran a script for more than 11 hours without checking progress.
Trump Media faces Nov. 30 deadline on up to $1B of convertible bonds
The Financial Times reports Trump Media must repay up to $1 billion of convertible bonds sold last May to fund a bitcoin treasury by Nov. 30. Its shares are down more than 60% this year, and the paper says the slump in shares and crypto prices could force it to sell assets or take on costly new debt.
AI startup rounds: Operand, Charta Health, Luminal, WhiteMatter
Operand, which builds AI due diligence tools for private equity, raised $31 million across seed and Series A led by Khosla Ventures at a $220 million valuation. Charta Health raised a $55 million Series B led by Insight Partners for medical chart review and coding. Inference-compiler startup Luminal raised a $25 million Series A led by Standard Capital, and WhiteMatter raised a $20 million Series A for guardian agents that oversee other AI agents.
Runway open-sources Praxis-1 robot control model under Apache 2.0
Runway released weights and code for Praxis-1, a model that uses video pretraining to predict scene changes and derive robot movements, under the Apache 2.0 license. It scores 46.2% on Nvidia's RoboLab-120, leading the world-action-model category and ranking fifth overall. Adapting it to a Franka arm still took about 277 hours of DROID robot data.
AI speeds up attackers: lone ransomware crews and Wiz's 'Black October'
Wired reports about 111 ransomware groups claimed attacks between July and September, with claimed incidents up 20% to 40%; ReliaQuest says one new group was run by an inexperienced teenager who made more than $800,000 in under two weeks. On an a16z podcast, Wiz CTO Ami Luttwak called October 'Black October,' saying attacks that took skilled attackers weeks or months now run in about 20 minutes and that Wiz's list of breached AI development platforms grew from nine to 15. He said about half of Wiz customers patched the PHP flaw within two to three days.
OpenAI says robotics team was reorganized, not shut down
OpenAI clarified that its robotics team was reorganized rather than disbanded: researchers move to the main research group under Mark Chen, while applied robotics and hardware staff join Greg Brockman's Applied organization. The company appears to bet that general models like GPT-6 Astra can control robots directly; in an Nvidia study an Astra-powered agent hit 74.8% on simulated tasks versus 51% for the best specialized models.
Ex-Anthropic researcher Coxon cites IPO slide in call for a pause
Jacob Coxon, who trained models at OpenAI and Anthropic, said a compute-efficiency chart in Anthropic's IPO roadshow was a key reason he left after about four months. He wrote that Anthropic 'could already retrain Mythos with a fraction of the compute originally needed' and called for a pause on frontier development next year so competitors can catch up.
AI Futures Project pulls its superhuman-researcher forecast forward to late 2027
The AI Futures Project now expects superhuman AI researchers by late 2027, earlier than its original forecast of 2029 or later, The Information reports. 'We're at the beginning of the takeoff period,' said Daniel Kokotajlo, the former OpenAI researcher who leads the group.
COMESA readies objections over Meta's WhatsApp AI policy
Global Competition Review reports that COMESA's competition watchdog is preparing a statement of objections against Meta over its WhatsApp AI policy. A separate formal antitrust probe into Meta's wider commercial terms in the region is to follow.
The Information: AI inference providers turn to lenders to pay for neocloud contracts
The Information reports AI inference providers have been talking to lenders about borrowing to pay for compute contracts with neoclouds rather than relying on venture capital. Fal has told investors it is interested in using debt to finance hardware purchases.
Anthropic tells Opus 5.5 users to drop 'think hard' and start at Medium
Anthropic made Claude Opus 5.5 the default on Pro, Max, Ultra and Team plans and published guidance against older prompting habits. It says phrases like 'think hard' only slow responses, and that Opus 5.5 at Medium effort matched Opus 5 at High in internal tests. Teams that run Sonnet at Extra High or Max effort pay about as much as Opus 5.5, it says.
The Information: Google explores buying AI ad-creation tools
Google's corporate development team has explored acquiring AI tools for creating ads, and Google is paying startup Hightouch to move advertisers onto its customer-data tools, The Information reports. The moves aim to address advertiser complaints that they cannot see how ads in Google's AI features perform.
GAO: Pentagon cannot account for 26 million F-35 parts worth $31B
A Government Accountability Office report says the Pentagon's inventory of about 26 million F-35 spare parts, worth an estimated $31 billion and spread across 700 facilities, is 'outdated, inaccurate, and incomplete.' It is the GAO's third report on the issue since 2019, and it says the Defense Department has made minimal progress on earlier recommendations.
DeepMind releases the datasets behind WeatherNext 3
Google DeepMind released PARDIG and SatWeave, the foundational datasets behind its WeatherNext 3 forecasting model, for researchers and developers building climate applications. DeepMind says accurate forecasting starts with how clearly the atmosphere can be observed in real time.
Trending on X
What X was arguing about today that the news hasn't covered. Views are the posters' own.
Elon Musk promotes Grok Bot after a user says it rebooked a cancelled flight by phone
Elon Musk (@elonmusk) wrote 'Try Grok @Bot' while quoting a user who said the agent sat through 11 minutes of hold music, booked a 7:40 a.m. flight and obtained a hotel and meals while 187 people waited at the airline desk. The user said the bot announced itself as an AI on the call and asked before paying a $49 seat fee. The official @bot account separately said Grok Bot can now join meetings, take questions and track team decisions.
Steve Yegge tells executives to slow down AI agents and not buy a software factory
Veteran engineer Steve Yegge (@Steve_Yegge) summarized his conference talks: pace AI output to human speed, because his agents have overwhelmed his game's players with features, his accountant with emails and his merge queue with tickets. He said he burned down his previous five software factories because rewriting them became cheaper than fixing them, and advised companies to build rather than buy. He predicted that next year most people will have a personal chief-of-staff agent and bring it to work.
Tom Lee blames crypto's slide partly on investors raising cash for the Anthropic IPO
Fundstrat's Tom Lee (@fundstrat) listed reasons for crypto's weakness, including margin calls among Asian retail traders after the Nvidia selloff and investors selling ahead to fund Anthropic IPO purchases, with a heavy retail allocation via Robinhood. @AshCrypto said nearly $2.35 billion of bitcoin ETFs have been sold this month and compared it with June's SpaceX IPO. Lee still called crypto 'the future of finance.'
Take-Two CEO Strauss Zelnick says OpenAI and Anthropic should protect others' IP too
A widely shared post from @Pirat_Nation quoted Take-Two CEO Strauss Zelnick: 'I think all companies should be protecting intellectual property, their own and everyone else's.' He called it 'ironic' that OpenAI and Anthropic expect protection for their own IP while their tools can be used to recreate existing games. The remarks are reported secondhand in the post.
Microsoft's Charles Lamanna says the 365 outage was neither a cyberattack nor AI
Microsoft executive Charles Lamanna (@clamanna) wrote that the outage 'was not a cyberattack or an issue caused by AI tooling,' pointing to cascading failures and 'thundering herds' at the network perimeter that made recovery harder. He said a post-mortem with a step-by-step timeline will be published shortly. The post landed as DHS Secretary Markwayne Mullin said 'we can't rule out' AI in the separate FAA disruption, prompting posts asking whether both were AI failures.