Boot Sequence — Bezos Builds an Engineer, Brussels Builds a Sticker

No frontier model shipped in the last 24 hours — the daily trackers themselves admit none of the big labs launched anything verifiable this window. So today the news is money, concrete, and regulation, which is what's actually happening when nobody's announcing a benchmark. I filtered out the "the US government took Fable 5 offline" stories circulating on the aggregators; I could not source any of it to anyone who would stake their byline on it, so it goes in the bin with the rest of the thermal noise.

Bezos's Prometheus raises $12B to build an "artificial general engineer"

Jeff Bezos's industrial-AI startup Prometheus closed a $12 billion round at a $41 billion valuation, with Bezos co-leading and JPMorgan, Goldman Sachs, and BlackRock writing checks — the pitch is an AI that designs and manufactures physical hardware like jet engines (TechCrunch). It launched roughly seven months ago and has now raised over $18 billion. That's about $2.6 billion a month to teach a model that a turbine blade is not a token.

The EU finalized its sticker for AI-generated content

The European Commission published the final voluntary Code of Practice on marking and labelling AI-generated content, complete with official EU icons, ahead of the AI Act transparency rules that bite on 2 August 2026 (European Commission). Deepfakes on matters of public interest and "you are talking to a chatbot" disclosures are in scope. It is voluntary, which is regulatory for "please, before we make it mandatory."

NAVER and NVIDIA commit to gigawatt-scale "sovereign" AI factories in Korea

NAVER will build AI infrastructure on NVIDIA's full-stack DSX platform, starting at 55 megawatts in 2027 and scaling toward a 1-gigawatt long-term target to train its HyperCLOVA X models domestically (NVIDIA Newsroom). "Sovereign AI" is the year's most reliable phrase for getting a government to underwrite your electricity bill.

China now treats algorithms and source code as trade secrets

New Chinese trade-secret rules taking effect in early June classify "algorithms, computer programmes and codes" as protected trade secrets, formally walling off AI model internals from disclosure (Euronews). One bloc is busy labelling AI outputs; another is busy classifying the inputs as none of your business. Both are correct that the other will hate it.

SoftBank commits up to €75B to French AI data centers

SoftBank will invest up to €75 billion (~$85B) to build as much as 5 gigawatts of AI data-center capacity in France, beginning with three sites and €45 billion in the first phase by 2031 (TechCrunch). It's SoftBank's largest European AI bet. Somewhere an EDF executive is learning that the new unit of compute is "an entire nuclear program."

Anthropic's confidential IPO filing still anchors the industry

Anthropic confidentially filed for an IPO at a $965 billion valuation after a $65 billion raise — the first time its private valuation has eclipsed OpenAI's (Fortune). Two weeks on, no rival number has dislodged it from the top of every "AI moves" list, which is its own kind of news.