Silicon Valley Daily
Data center operator Hut 8 fell 23% on Wednesday while Amazon rose 8%. A day earlier, voters sided against data centers in 40 of 41 local measures across 10 states, Texas elected a governor who wants a construction pause, and Musk started talking about data centers in space. Investors bought the cash-rich clouds and sold the developers that depend on new sites. Inside AI companies the math is getting sharper: OpenAI spent about $1.2 million in API credits to produce 948 Linux root exploits, Apple cut some engineers' daily AI budgets from $300 to $200, and Uber said AI agents booked four rides in its busiest week. Talent and capital kept moving: Claude Code's product lead Cat Wu joined OpenAI, Isomorphic Labs is in talks at about a $50 billion valuation, and Anthropic paid a news publisher, the AP, for the first time.
Voters side against data centers in 40 of 41 local ballot measures
Ballotpedia and CNBC counted 41 data center measures in 10 states on Tuesday; voters sided against the projects in all but one, in Harwood, North Dakota. Ohio held 21 of them, and Urbana and Hubbard passed outright bans with 76% and 77% of the vote, Engadget reported. Semafor says banks and asset managers already treat local opposition as a credit risk, and TechCrunch notes the Ohio county where Meta planned a 700,000-square-foot facility has just passed a ban.
Hyperscalers rally while data center stocks slide after Democratic sweep
Amazon rose 8%, Google more than 6% and Apple nearly 7% on Wednesday, CNBC reported, as JPMorgan Asset Management's David Kelly said divided government removes uncertainty. Data center names fell: Hut 8 dropped 23%, IREN 11% and Core Scientific 10%, Reuters reported, and CoreWeave lost 10%. Needham's Laura Martin wrote that higher regulatory barriers entrench incumbents. In Australia, NextDC jumped 43% after Goodman Group and Digital Realty moved to buy stakes of up to 9.9%.
OpenAI pipeline built 948 Linux root exploits for $1.2 million in API costs
OpenAI said a mid-October run of its vulnerability pipeline, using an upcoming cyber model, reproduced 1,724 Linux kernel vulnerabilities and produced 948 working local privilege-escalation exploits against a stock desktop distribution. The run cost $1,213,430 in API credits, about $1,279 per exploit; human researchers reviewed the output and are reporting bugs with candidate patches. OpenAI says a low-privilege account on an unpatched system should now be treated as equivalent to root against well-resourced attackers. It also made Daybreak Red, built on GPT-5.6 Cyber, available in AWS GovCloud.
Texas blue wave puts the AI buildout and Musk's home base in question
Democrat Gina Hinojosa unseated Gov. Greg Abbott, James Talarico beat Ken Paxton for Senate, and Democrats will control the Texas House for the first time in 24 years, MIT Technology Review reported. Hinojosa campaigned for a moratorium; projects underway include the 1,820-megawatt Stargate Frontier site. Republicans still hold the Texas Senate, and experts told Wired that developers enjoy strong vested rights. Musk, who put about $474 million into the midterms with nearly half in Texas, posted that data centers in space look 'pretty appealing right now.'
Claude Code product lead Cat Wu joins OpenAI
Cat Wu, who led product for Claude Code and left Anthropic last week, said she is joining OpenAI to lead product management for its core products team, Business Insider reported. She will work with Nick Turley and Ashley Alexander and report to Tibo Sottiaux, an OpenAI representative said. The move comes weeks before Anthropic's planned IPO; Wu wrote that she had 'no idea' whether leaving would let her sell shares earlier and that it 'wasn't a factor.'
Isomorphic Labs in talks to raise about $5 billion at a $50 billion valuation
Alphabet's AI drug discovery company Isomorphic Labs is in talks to raise roughly $5 billion at a valuation of about $50 billion, Bloomberg reported, citing people familiar with the matter. The Google DeepMind spinout last raised $2.1 billion in May.
Google defends school AI policy after Sanders, Khan and Mamdani object
Under Google's plan for 2027, schools that block its AI features entirely also lose Google Search; education VP Chris Phillips told the New York Times that 'Search and AI are deeply integrated.' Senator Bernie Sanders called it 'extortion,' former FTC chair Lina Khan called Google an 'unregulated monopolist,' and NYC mayor Zohran Mamdani also objected. Google replied that Gemini, Gemini Notebook and Search are off by default and schools can turn each on; CNBC notes Search comes back with AI Overviews. DuckDuckGo is now pitching schools on switching their default search engine.
Pro-AI super PACs spent heavily and lost the races they contested
American Mission and Building the Future spent $4.8 million backing South Carolina Senate candidate Darline Graham, who lost to Annie Andrews, TechCrunch reported. Wired says Leading the Future started with about $100 million, spent about $26 million on primaries and ends the midterms with just over $20 million, mostly backing safe candidates; the Anthropic-backed Public First did better with Democrats. The Verge counts $47 million from Musk's America PAC aimed at James Talarico, who won with 55%. Senator Ron Wyden reacted on X with one word: 'Subpoenas.'
Goldman and Apollo shop pieces of SpaceX's $40 billion chip financing
Goldman Sachs and Apollo are in talks with investors to sell portions of SpaceX's roughly $40 billion chip financing, asking some for up to $500 million each before broader syndication, Bloomberg reported, citing people familiar with the matter. On the same day SpaceX shares fell 3% as Musk-linked and data center stocks slid after the midterms, CNBC reported.
Anthropic signs its first news licensing deal, with the AP
Anthropic struck a content and data licensing deal with the Associated Press just before the midterms, Adweek reported, citing two people familiar with the matter. It is the first time Anthropic has paid a digital news publisher; Claude surfaced AP election data on Election Night. Neither side disclosed terms. OpenAI has more than a dozen such deals.
Apple cuts engineers' daily AI spending caps
Apple lowered some engineering teams' daily AI spending cap from $300 to $200 in October and plans further cuts to as little as $100, Bloomberg reported, citing people familiar with the matter. Outside engineering, Apple is weighing a limit as low as $25 a day, according to TechCrunch's account of the report. Bloomberg says the cost cutting has stoked staff fears of layoffs.
Qualcomm handset chips fall 26%; Arm royalties miss estimates
Qualcomm's revenue fell 8% to $10.3 billion, Bloomberg reported, with handset chips down 26% to $5.14 billion as memory costs push up phone prices; shares fell 7% after hours on a weak forecast, CNBC reported. CEO Cristiano Amon expects $5 billion in data center revenue in fiscal 2027 and said 'it's a supply issue.' Arm's revenue rose 27% to $1.44 billion, ahead of estimates, but royalties grew 8% to $670 million against $720.8 million expected, Reuters reported.
Investors expect Anthropic IPO before Thanksgiving; a critic urges the SEC to block it
Investors who have backed both OpenAI and Anthropic told Newcomer that Anthropic's IPO still looks on track; one described 'infinite demand' from long-term funds and expects a listing before Thanksgiving, while Newcomer says OpenAI's latest financing has drawn tepid interest. In a Tech Policy Press op-ed, Matt Scherer argued the SEC should block the IPO, citing at least $582 billion in non-cancelable commitments to five companies and unaudited 2026 results, against $96 billion in cash.
Meta raises €9 billion in debut euro bond on record €53 billion of orders
Meta raised €9 billion in its first euro-denominated bond sale, which drew a record €53 billion of orders, Bloomberg reported. It paid about 1.4 times the risk premium Alphabet offered on similar bonds this week.
Uber says only four rides were booked by AI agents in its busiest week
Uber handled about 320 million trips and orders during Halloween week, but only four rides and 500 Uber Eats orders came through AI agents, CFO Balaji Krishnamurthy said on the earnings call. CEO Dara Khosrowshahi noted a ride takes about 50 seconds to book in the app and called traction 'very, very early'; DoorDash reported only a few hundred agent orders last week. On an a16z podcast, David Pawlan, who has tested more than 140 AI assistants, argued the most compelling agents may do more than save time.
Meta buys Conductor, maker of a multi-agent coding tool
Meta acquired Conductor, which lets developers run several AI coding agents in parallel; CEO Charlie Holtz said the team will join Meta Superintelligence Labs and ship 'something new' early next year. Conductor has stopped new sign-ups, will shut its cloud service in the coming weeks and plans to wind down the local app in early 2027. Meta confirmed the deal but did not disclose terms.
Figma beats and raises, but shares fall as AI credits squeeze margins
Figma's third-quarter revenue rose 43% to $391.4 million and it lifted its 2026 forecast to $1.497 billion to $1.499 billion, yet shares fell 7% after hours, CNBC reported. Gross margin of 80% missed the 81% estimate, per Reuters, as Figma gives customers more AI credits at the same price. CFO Praveer Melwani said the company is trading near-term revenue for durable growth; the stock is down about 41% this year.
AMD will pay about $2.9 billion for inference chip startup Taalas
AMD disclosed in a filing that it will pay about $2.9 billion in cash and stock for Toronto-based inference chip startup Taalas, with $1.4 billion dependent on unspecified conditions, the Globe and Mail reported. Taalas, founded in 2023 by former AMD executives, had raised over $225 million, and its 2025 Series B implied a valuation above $750 million. AMD announced the deal in August without a price.
Snap misses on users; Snapchat+ passes 30 million subscribers
Snap's third-quarter revenue rose 14% to $1.72 billion and daily users reached 495 million, short of the 499 million expected; shares fell more than 10% after hours. Snapchat+ passed 30 million paid subscribers, up 5 million since February. Snap said monthly active users will become its primary audience metric starting in 2027.
OpenAI to acquire Pierre, whose libraries already power Codex
OpenAI agreed to acquire Pierre Computer Company, whose Trees and Diffs libraries already handle file browsing and code review in Codex. OpenAI says it wants to adapt Git workflows to teams working alongside coding agents, building on Pando, its in-house source control backend. Pierre's open-source projects will stay open; closing requires regulatory approval.
OpenAI ships Images 2.6 with animated GIF creation
OpenAI released Images 2.6 to ChatGPT and Codex users and the API, saying it generates images 50% faster and can make animated GIFs from prompts or photos, with frame-by-frame editing. OpenAI says users now generate more than 4 billion images a week, up from 3 billion in September. The API adds GPT-Image-2.6 Flare and Sunburst variants.
Apple will scan public iCloud links for 'harmful content'
Starting with iOS 27.3 and companion releases, Apple will check iCloud items shared with 'Anyone with the link' for harmful content and may delete the content or disable accounts, according to a support document reported by 9to5Mac. Such links are not end-to-end encrypted, The Verge notes, and Apple has not defined 'harmful content.'
Sequoia raises about $10 billion and shuts Arc, Scouts and London
Sequoia said in a memo that it raised roughly $10 billion in new funds while closing its London office, shutting its Arc accelerator and Scouts program, and merging early-stage and growth teams. The firm says a 20-year expansion created hierarchy. European partner Luciana Lixandru declined a move to San Francisco and will invest personally, TechCrunch reported.
Tech lobby says the vote was about the economy, not AI
Chamber of Progress CEO Adam Kovacevich said the results 'offer little evidence of a voter mandate to crack down on AI or data centers,' Ars Technica reported. NBC exit polls ranked cost of living and corruption above AI, but Blue Rose Research found Americans would rather live near a prison or a methadone clinic than an AI data center. On the All-In podcast, comedian Tim Dillon framed the blue wave as a vote against the cost of living.
OpenAI's Chris Lehane pitches an AI 'New Deal' for 2028
In a new essay, OpenAI global affairs chief Chris Lehane compared AI's economic impact to the New Deal and urged 2028 candidates to address voters' concerns, TechCrunch reported. His proposals include incentives for workers to learn AI skills and new forms of employee ownership, and he warned that data center moratoriums could jeopardize America's lead over China. AI Now's Kate Brennan called the New Deal framing 'ahistorical at best.' Platformer reports frontier labs have also set up election war rooms against deepfakes.
Tencent releases Hy4 with vision and computer-use skills
Tencent released Hy4, a 770-billion-parameter model that reads charts and documents and operates command-line and graphical tools, with targeted training on applications such as Blender and CAD software. It is available in Tencent products and on OpenRouter at unchanged API prices of $0.834 per million input tokens and $2.501 per million output tokens; Tencent plans to publish weights and a system card shortly.
Golden Dome and other Pentagon priorities at risk after Democratic sweep
Republicans had relied on reconciliation to fund projects such as Golden Dome, and that route is now a long shot, Axios reported. A TD Cowen note said 'Trump branded projects are dead,' and AEI's Todd Harrison expects the defense budget to run on a continuing resolution into next year. Reps. Pat Ryan and Thomas Massie have floated impeaching Defense Secretary Pete Hegseth.
Revolut valued at $115 billion in secondary sale ahead of US bank launch
Revolut's secondary share sale, led by investors including Glade Brook, Coatue and T. Rowe Price, valued the fintech at $115 billion, 53% more than a year ago, TechCrunch reported. No new money went to the company. Revolut has conditional OCC approval for a US national bank it plans to open in the first half of 2027, and posted a 38% profit margin on $6 billion of revenue in 2025.
Moonshot releases Kimi K3.1 Pro open weights
Moonshot AI published weights for Kimi K3.1 Pro, a natively multimodal agentic model it calls its most capable 3-trillion-parameter-class model, with a 1-million-token context window. It replaces Kimi K3's full-attention layers with Gated DeepSeek Sparse Attention for more efficient long-context work.
Lame-duck Senate moves on an FCC seat and an energy permitting bill
Sen. Ted Cruz scheduled a November 10 hearing for FCC nominee Danielle Thumann Severs, which would give Republicans a 3-1 majority on the commission before Democrats take the Senate in January, Ars Technica reported. The Senate is also set to vote November 9 on the Bipartisan American Affordability and Jobs Act, a White House-backed overhaul of energy project permitting, MIT Technology Review reported. Democrats lack the two-thirds majority needed to override vetoes.
Ricursive's Ripple-65B claims top US open-weight coding score for $32M
Ricursive Intelligence released Ripple-65B Preview, a dense open-weight model it says scores 80.52 on TerminalBench 2.1, ahead of larger US open models such as Reflection Beam 501B at 80.1. The company says its full research cost from scratch was $32 million, with pretraining on 12 trillion tokens; the model runs on Hugging Face, Together AI and OpenRouter.
Uber beats on freight; Waymo will end Uber exclusivity in two cities in 2028
Uber's revenue beat estimates on 34% freight growth to $1.76 billion, but ride-hailing revenue fell 2% to $7.52 billion; shares rose more than 7%, CNBC reported. Waymo co-CEO Tekedra Mawakana told Bloomberg it will end exclusivity with Uber in Austin and Atlanta in January 2028 and offer rides in its own app. Uber says it has committed about $10 billion to bring driverless vehicles onto its platform.
Anthropic extends text watermarks; a sampling tweak defeats them on older models
Anthropic told API customers that Claude Opus 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 will carry text watermarks starting November 16, citing the EU AI Act's December deadline, The Decoder reported. On those models the watermark applies only when temperature, top_p and top_k stay at defaults, so changing any one of them avoids it; Anthropic's email responds with a compliance warning.
Suspected Kremlin hackers used patched iOS and Android bugs in watering-hole attacks
Security firm Sekoia said a group possibly linked to the Kremlin injected malicious JavaScript through proxy sites including a copy of america250.org and a site imitating a London hotel near UK government offices, Ars Technica reported. One exploit chain was DarkSword, an iOS kit previously tied to a Turkish surveillance vendor. All the flaws were already patched, so only outdated phones were exposed.
Hinton warns of a near-term AI incident; Hugging Face's CEO puts p(doom) near zero
Geoffrey Hinton told the Guardian that humanity might have been better off never developing AI and warned of a possible large-scale incident, such as a temporary blackout of financial or power systems, this year or next. At WIRED World Fair, Hugging Face CEO Clément Delangue said his p(doom) is 'close to zero' and that most risk comes from a couple of players building behind closed doors; he said the platform now has more than 20 million AI builders. Hugging Face policy chief Irene Solaiman discussed how companies profit from open models on TechCrunch's Equity podcast.
Users flag Gemini chat training and Codex data-sharing settings
A Hacker News post citing Google's support pages notes that personal Gemini subscribers' saved chats can be used to train models unless chat history is turned off, and that a subset of chats is read by human reviewers. In a separate thread, developers at one company said the 'Improve the model for everyone' setting in Codex had been switched back on without anyone opting in.
DreamHost says hackers stole customer files and inboxes over months
DreamHost said in a breach notice that hackers exploited an unspecified vulnerability starting in August to obtain session tokens and access customer emails and files; it detected the attack only last week. A spokesperson said affected customers are 'well under 10%' of its roughly 400,000 paying customers, TechCrunch reported.
Apple opens Siri AI app integration to developers in beta
Developers can now test Siri AI app intents in the iOS 27.3, iPadOS 27.3, macOS 27.3 and visionOS 27.3 betas, and Apple expects to accept App Store submissions next spring, nearly three years after promising deeper third-party Siri integration, The Register reported. A TechCrunch reviewer found the new Siri Recaps useful within a day but noted privacy reservations.
Nothing says its first screen-free AI device ships in December
Nothing CEO Carl Pei said on X that the company's first 'AI-native product' ships in December; it has no new screen and 'couldn't have been an app.' Pei called it a productivity tool and distanced it from the pendants, pins and glasses others are pursuing.
Nvidia RTX Spark laptops run AI at a fifth of the speed on Linux
A MindStudio test found GPU clocks on an RTX Spark laptop drop to about 780 MHz on wall power under Linux, versus roughly 2.2 GHz on Windows, because the firmware's power table is encrypted and only the Windows driver can decrypt it. GPT-OSS-120B ran at 13.9 tokens per second on native Linux against 65.4 on Windows; WSL matched Windows.
Nintendo's main cartridge ROM supplier Macronix exits the business
Macronix said its ROM chip production 'has reached its end' and it will gradually decouple from the game industry, shifting capacity to NOR, NAND and eMMC memory, Ars Technica reported. ROM output fell 39% in a year and is now 5% of Macronix's business. Digital sales were 61.5% of Nintendo's software business in the second calendar quarter.