The AI capital cycle hit its peak: Anthropic confidentially filed an S-1 that Wired calls the largest IPO ever, Alphabet announced an $80B equity raise specifically to fund AI infrastructure and compute, and Berkshire Hathaway is now an equity holder in Google. Florida filed a first-of-its-kind state lawsuit against OpenAI and Sam Altman after multiple ChatGPT-linked violent incidents — and the Trump administration is at war with itself over whether to regulate AI at all. NVIDIA opened a second front against Intel and AMD with RTX Spark AI-agent PCs through Microsoft, Dell, and HP, while Intel's Crescent Island chip pitched cheaper and cooler than the incumbents. Agent research moved with the news: Multi-Agent Computer Use formalizes orchestration across CUAs, and two sharp physical-AI papers — Silent Failures and Can Predicted Dynamics Exist — name the runtime-authorization gap robots and VLA models actually fail through.
Anthropic filed for what Wired calls the largest IPO ever, Alphabet announced an $80B equity raise specifically for AI buildout, and Berkshire Hathaway is now a Google equity holder. The public-market AI thesis is being priced live, and water is now a documented IPO risk factor for SpaceX.
Anthropic confidentially filed paperwork to go public — the company that just raised $65B at a $965B valuation is heading to the public market.
Wired calls the Anthropic S-1 a candidate for the largest IPO in history.
Alphabet will issue $80B in equity capital explicitly to expand AI infrastructure and compute (206 HN points).
Stratechery reads Berkshire Hathaway's Google equity stake as signaling sustained AI-cycle demand and a long-duration buy from the most patient capital allocator.
SpaceX's IPO filing lists water access as a 'significant' risk factor for its data-center cooling — the resource fight is now a public-disclosure item.
Florida filed a first-of-its-kind state lawsuit against OpenAI and Sam Altman over ChatGPT-linked violent incidents; the Trump White House is openly split on whether to regulate AI at all; DuckDuckGo's 'no AI' search is growing on the back of user fatigue; and OpenAI published its own AI-policy advocacy framework. Defense, plaintiff, and legislator are all moving.
Florida's AG accuses OpenAI and Sam Altman of 'utter disregard for human lives' in a first-of-its-kind state lawsuit tied to multiple ChatGPT-linked violent incidents.
TechCrunch on the same Florida lawsuit — partial focus on a Florida State University shooting tied to a ChatGPT exchange.
Wired reports the Trump White House is internally split on AI regulation after the administration killed an executive order on the topic.
OpenAI publishes its policy-advocacy stance — transparency about lobbying and an explicit posture on thoughtful regulation.
DuckDuckGo ships Chrome and Firefox extensions for its 'no AI' search mode as traffic grows on the back of user fatigue with AI-everywhere search.
NVIDIA went after the $200B CPU market with RTX Spark AI-agent PCs in partnership with Microsoft, Dell, and HP; Intel countered with Crescent Island as a cheaper, cooler alternative; Mellum2 (12B MoE from JetBrains) and Stargate Michigan (OpenAI's 1GW DC groundbreaking) fill out the build-out side.
Nvidia's RTX Spark targets the $200B consumer-laptop CPU market with AI-agent PCs co-launched by Microsoft, Dell, and HP.
The Verge reads RTX Spark as Windows' M1 moment — fundamentally new performance per watt, with a price tag to match.
Intel's Crescent Island is an air-cooled, LPDDR5-based AI chip pitched as cheaper and cooler than Nvidia/AMD options.
OpenAI broke ground on a 1GW Michigan Stargate data center.
JetBrains releases Mellum2 — a 12B-parameter MoE model — tightening the open-weights coding-model tier.
Meta's own AI support chatbot was exploited to hijack celebrity Instagram accounts; GitHub Copilot's usage-based pricing kept generating angry headlines; Strava restricted API access blaming AI scrapers. The story is buyers and platforms both losing patience with the unconstrained agent surface.
Meta's AI support chatbot was duped into authorizing takeover of pricey, high-profile Instagram accounts before Meta patched the exploit.
Ars Technica chronicles devs blowing through monthly Copilot AI credit allotments in a single day under the new usage-based model.
Strava restricts third-party API access, citing AI scrapers and zero-code apps as the trigger.
Robotics startup faces $12K lawsuit after allegedly damaging Airbnbs while testing robots — the data-collection arms race meets civil liability.
Multi-Agent Computer Use formalizes orchestration across CUAs — the field is now arguing default-multi, not default-single. Two sharp physical-AI papers name the silent-failure and predicted-dynamics-admissibility problems robotics deployment actually fails through.
Argues computer-use agents should be evaluated and built as multi-agent systems by default — task decomposition, parallel execution, replanning.
Survey of runtime action-authorization gaps in physical AI — silent failure modes that current robotics and VLA stacks systematically miss.
Defines physical admissibility as a prediction-control interface — a decoded proposal is treated as candidate dynamics before execution.
Fine-grained self-verification scales test-time compute for agentic search when correct answers are sparse and selection scores are poorly calibrated.