SpaceX's AI unit turns itself into an AI cloud firm
The compute market got a new heavyweight this weekend: xAI's old bones are now selling GPUs to everyone, Anthropic is spending to grow its own army of deployers, and Yann LeCun picked a very public fight with the doomer camp.
SpaceX's AI unit has quietly become an AI cloud provider — and its customers include the labs. According to an investigation by The Information, xAI was folded into SpaceX earlier this year and renamed SpaceXAI, and Musk — who once vetoed selling compute on the grounds that it would blur xAI's standing as a lab — has reversed himself, starting with Anthropic as a customer. The money is real: SpaceX CFO Bret Johnsen disclosed a $1.1 billion-a-month compute deal at Goldman Sachs' conference in September, with payments starting in December, which annualizes to roughly $13 billion toward Musk's stated $100 billion year-end run-rate ambition. Anthropic, Google, Reflection AI and Cursor all appear on the customer list, and reporting from CNBC and TechCrunch independently confirms the pivot: Colossus II sat under 40% utilized for months, and surplus capacity looking for a buyer is exactly how neoclouds are born. We flagged the same trade when Oracle just proved the AI cloud trade is real — the bill for it is coming due anyway, and SpaceXAI now walks straight into that market. The catch: nearly every one of these deals carries a short exit clause, so the compute is sold, but never forever.
Anthropic is spending $100 million to train 10,000 "frontier deployed engineers." The company announced the Claude Frontier Academy on October 2 — a multi-day in-person program in San Francisco, New York and London that runs nominated engineers from Accenture, Bain, Capgemini, Deloitte, McKinsey, Morgan Stanley and others through a graded practical deployment, with the first certified graduates expected in early 2027. The pitch is bluntly commercial: enterprises keep saying they can't find people who understand both their business and the models, and Anthropic would rather certify those people itself ahead of an IPO it reportedly wants at a $2 trillion valuation. It's a familiar playbook — vendors manufacturing the talent market for their own platform — and if the badge carries weight, it becomes a distribution channel dressed up as a training program.
LeCun calls Amodei "deluded" and says he has "zero concerns" about AI extinction. In a wide-ranging Fortune interview, the Turing laureate and new AMI Labs founder said he isn't worried "at all" about AI wiping out humanity, attributed rogue-agent incidents like the Hugging Face hack to "leaky and horribly designed" sandboxes rather than model nature, and argued that Amodei and Altman warning publicly that "AI can kill us all" is "the worst marketing campaign you can possibly imagine." He saved his harshest words for effective altruism — "super toxic," "a complete disaster" — while conceding he worries more about regulatory capture than extinction. It's the sharpest version yet of the industry's counter-safety faction speaking from a sitting founder's chair rather than an op-ed page.
What to watch: whether SpaceXAI publishes standard cloud pricing or keeps leasing deal-by-deal — the difference between a real neocloud and a series of private bargains.
Is selling your rival the compute they need to catch you a business strategy or a trap? Tell us in the comments.
