There’s a moment when any technology stops being a curiosity and starts being infrastructure. Call it this week.
Anthropic closed a $65 billion Series H at a $965 billion valuation — edging past OpenAI by roughly $113 billion — and filed a confidential S-1 with the SEC on June 1. The company’s annualized revenue has crossed $47 billion, more than doubling in a single quarter. Dario Amodei, who said two years ago that raising $5 billion felt surreal, now runs a company bigger on paper than Ford, Delta, and Marriott combined. The IPO, targeting a debut as early as this fall, would be one of the largest AI listings ever attempted.
Meanwhile, Sam Altman, Amodei, and Demis Hassabis have all confirmed they’re heading to the G7 summit in Évian-les-Bains, France, this weekend. France holds the rotating G7 presidency this year and has placed AI prominently on the agenda. The summit will include a working lunch with political leaders and tech executives on AI infrastructure, regulation, and frontier risks — with voluntary commitments on youth safety and biosecurity expected to emerge. What’s notable is the symbolism: all three major labs at the same table, for the first time, not testifying before Congress but sitting where economic and security policy actually gets made.
On Thursday, Jeff Bezos’s Prometheus raised a $12 billion Series B at a $41 billion valuation. Prometheus isn’t building a chatbot. Co-CEO Vik Bajaj — previously co-founder of Alphabet’s Verily — calls it an “artificial general engineer”: software capable of automating the design and manufacture of complex physical systems, from jet engines to drug compounds. Bezos is co-CEO; it’s his first operating role since stepping down from Amazon in 2021. Whether “AGE” joins “AGI” in the lexicon of aspirational acronyms is a fair question, but $12 billion from JPMorgan, BlackRock, and Goldman Sachs is a serious answer.
The open-weight frontier closes in
Not everything this week fits the “AI goes establishment” narrative. Shanghai-based MiniMax released M3 — the first open-weight model to combine frontier-level coding, a one-million-token context window, and native image and video understanding in a single architecture. On SWE-Bench Pro it scores 59.0%, beating both GPT-5.5 and Gemini 3.1 Pro. The weights are releasing this week. Anyone with their own hardware is about to have access to a model that outperforms most commercial systems on software engineering tasks, at $0.60 per million input tokens.
Claude Opus 4.8 still leads at 88.6% on SWE-Bench Verified, but the gap between frontier proprietary models and the best open-weight alternatives is now measured in percentages, not multiples. The pattern is consistent: every few months, open-source closes another gap that felt permanent. For most production use cases, the era of closed labs holding an insurmountable technical lead may already be over.
The week’s through-line is about institutions claiming permanence. An IPO, a summit seat, a $41 billion engineering bet, and an open model that anyone can download — all in the same five-day window. The question is no longer whether AI is becoming infrastructure. It’s who controls which layers.