Nvidia has spent the past three years selling the picks and shovels of the AI boom. Now it wants to help pay for the mine.
Riot Platforms signed a 20-year, $9.1B deal to supply 191 MW of AI compute to a lab identified as Anthropic.
Cambridge Aerospace, a two-year-old UK air-defense startup, raised $300M at a $3.4B valuation to shoot down cheap drones.
Boeing agreed to hand Wisk Aero, Insitu and SkyGrid to Archer Aviation, taking a nearly 20% stake instead of funding them itself.
Flagler Health raised a $50M Series B led by Bessemer to build an AI operating system for the $400B MSK care market.
Anthropic is weaving imperceptible watermarks into Claude's text and C2PA metadata into its images to meet EU rules.
Meta shipped Muse Spark 1.2 on August 6 with little fanfare, but paired with the open on-device Muse Glimmer it signals a real open-model return.
Alibaba's largest model yet arrives with 2.4T parameters and a bold 'second only to Fable 5' claim - but no benchmark table.
A new arXiv paper turns multi-agent reasoning into a market, auctioning off each reasoning step to the best-calibrated solver.
Ask an LLM to check its own work and it often repeats the same mistake. MAR fixes it with a debate.
Ask an LLM to help diagnose a dead pump and tell it you suspect the pressure switch, and it will usually run with your wrong theory.
OpenAI's only dedicated ethicist has left, and the company does not plan to replace her, amid an autonomous-cyber scare.
You can carry a smartphone into an English courtroom. You cannot, as of this week, carry Meta's camera glasses.
Britain handed a slice of its most powerful supercomputer to Cosine, a ~30-person London startup, betting focus can beat scale.
Security vendors and investors are bracing for a world where AI attacks other AI, and humans are too slow to sit in the loop.