Government goes all-in on AI
The day's biggest story came out of Sacramento: California signed a first-of-its-kind deal with Anthropic to put its Claude assistant in the hands of state workers — at roughly half price. Governor Gavin Newsom's administration is framing the agreement, reported across Politico, CBS News, citybiz and others, as an expansion of AI inside government, rolling the assistant out across state agencies at a steep discount. It's a notable vote of confidence in deploying generative AI for public-sector work, and a marquee win for Anthropic.
Claude had a good day on the infrastructure front too. Anthropic's models are now generally available inside Microsoft Foundry, hosted on Azure and running on Nvidia's GB300 'Blackwell Ultra' GPUs — bringing several of its models to one of the biggest enterprise clouds on the newest silicon.
But there isn't enough compute to go around
That hunger for chips is the day's quieter through-line. Google has reportedly told Meta it can't supply all the Gemini capacity the Facebook parent wants, blaming a shortage of compute. When two of the largest companies on earth are rationing AI horsepower between them, it underscores just how tight the supply of cutting-edge capacity has become — the same scarcity that makes Blackwell Ultra availability newsworthy.
Meta's chatbot espionage
In a stranger Meta story, hundreds of contractors working on a Meta project posed as teenagers and prompted rival chatbots — including Google's Gemini and OpenAI's ChatGPT — to engage on high-risk topics like suicide, sex and drugs. The effort, in effect, probed how competitors' safety guardrails hold up under adversarial, minor-impersonating prompts.
The money and the metrics
The business side stayed busy. Investor Chamath Palihapitiya stepped out from behind the checkbook to name himself CEO of his AI coding startup, 8090, which has raised fresh funding. And Arena, the company behind a widely used free AI leaderboard, revealed it has quietly built a $100 million business in under a year — proof that even 'free' AI tooling can mint real revenue.
There was a reality check, too. New research that put AI models through traditional human intelligence tests found a telling split: the models ace verbal and word-based exams but stumble badly on visual and picture-puzzle reasoning. Fluent on paper, far less so when they have to actually see.
Creators, consumers and cops
TIDAL is moving against the flood of AI-generated music — not by deleting it, but by changing its new policy so AI tracks can't earn a payout, cutting off the money rather than the catalog. Google, meanwhile, is widening access on the consumer side, bringing Gemini's personalized AI image generation to eligible free users in the U.S.
Finally, a look at AI's expanding real-world footprint: police forces are leaning ever harder on AI to investigate and solve crimes, and Axon — the Taser maker reinvented as a policing-tech powerhouse — sits at the center of that gold rush, with an associated stock trade now drawing scrutiny.
The theme of the day? AI is being bought, deployed and monetized faster than the chips, the guardrails, and the oversight can keep up.