California adopts Claude statewide; South Korea's $880 billion AI bet

Governor Gavin Newsom announced a first-of-its-kind deal making Claude the first AI assistant offered to all California state agencies — plus every city and county — at a 50 percent discount, bundled with free training and Anthropic engineering support. The agreement lands as Anthropic fights a federal "supply-chain risk" label, imposed after it refused Pentagon requests for mass-surveillance and autonomous-weapons tools. California's chief information officer said the federal designation simply "didn't come up."

President Lee Jae-myung unveiled three "mega projects" totaling roughly $880 billion over ten years — about 5 percent of the country's GDP. The package pours $518 billion into new Samsung and SK Hynix memory fabs to ease the AI-driven shortage some call "RAMageddon," funds $356 billion of AI data centers, and aims to lift South Korea's share of the humanoid-robot market from 1 percent to 20 percent, with Hyundai's Boston Dynamics mass-producing Atlas robots by 2028.

Chinese delivery giant Meituan open-sourced LongCat-2.0, a mixture-of-experts model with 1.6 trillion total parameters (about 48 billion active per token) and a 1 million-token context window. It is the first trillion-parameter model fully trained and run on a 50,000-card domestic chip cluster, and it scored 59.5 on the SWE-bench Pro coding benchmark — edging out GPT-5.5. The release is a milestone for China's push to build frontier AI without Western silicon.

Neutral-atom quantum startup QuEra laid out a roadmap to a "gigaquop" machine — one capable of roughly a billion reliable logical operations — targeting more than 1,000 error-corrected qubits and full fault tolerance by 2029. A smaller "megaquop" system, Libra, is due on Amazon Braket in 2028. The plan builds on QuEra's recent result squeezing one logical qubit from just two physical ones, though researchers caution it is still a paper milestone, not a shipped machine.

Arena — the UC Berkeley-born platform where users blind-test rival models and vote on the winner — hit $100 million in annualized revenue, up from $30 million in January. Its free public leaderboard is the field's de facto scoreboard, but the money comes from a paid evaluations service that labs and enterprises use to tune their models. Backed by a $1.7 billion valuation, Arena now competes with human-labeling firms like Scale AI for post-training budgets.

Sunrun, Tesla, and Renew Home are pooling hundreds of thousands of home batteries and more than 8 million smart thermostats into a 16-gigawatt virtual power plant aimed at AI data centers — capacity that needs no new land, water, or hardware and can come online in months rather than years. With US data-center demand projected to climb from 41 gigawatts this year to 66 gigawatts in 2027, the Brattle Group estimates better use of the existing grid could cut US power bills by $110 billion to $170 billion over the next decade. About 300 megawatts is already available in Virginia's Data Center Alley, though the headline figure is theoretical rated capacity, not firm supply.

Cursor released a public-beta iPhone app that lets developers kick off and supervise its autonomous coding agents away from the desk. It reflects Cursor 2.0's bet that people increasingly manage agents rather than type code — Anthropic's Boris Cherny says most of his coding now happens on his phone. The launch is the company's first major product move since its post-IPO acquisition by SpaceX, which is folding Cursor into its expanding software stack.

Starting July 15, the streaming service TIDAL will demonetize fully AI-generated tracks, tag them with an "AI" badge, and automatically remove songs that impersonate real artists. It is among the most aggressive label-and-demonetize stances yet, following Deezer — which says 44 percent of its daily uploads are now AI-generated. TIDAL framed the move as protecting "organic creativity" as synthetic music floods streaming platforms.