Anthropic just signed a $45 billion agreement with Nscale, a British AI infrastructure company founded only in 2024, to rent compute capacity powered by Nvidia’s upcoming Vera Rubin chip system. A source familiar with the deal told TechCrunch that the six-year arrangement will draw from Nscale’s flagship data center in West Virginia, with capacity expected to come online in late 2027.
The Vera Rubin architecture is notable for combining six different chips working in concert — Nvidia’s most advanced design yet. Nscale has already cut deals with Microsoft, which lends some credibility to a company that barely existed eighteen months ago. But betting $45 billion on infrastructure from a startup still in its infancy is a remarkable show of both desperation and conviction in the current AI moment.
This is hardly Anthropic’s first big swing. Just this month, the company signed a $10 billion deal with Volta, a cloud startup founded in January, for six years of compute from Norway. In July came a $5 billion AMD partnership. May brought a SpaceX arrangement reportedly worth $1.25 billion per month in capacity — a strange-bedfellows moment given Elon Musk’s feud with OpenAI CEO Sam Altman. April saw an Amazon expansion adding 5 gigawatts of compute, plus a broadened Google and Broadcom relationship the same month.
Add it up and Anthropic has committed well over $60 billion to compute in roughly eight months. The scale is staggering, but the logic is straightforward: without massive compute reserves, no lab can realistically compete with OpenAI on frontier models. Google, Meta, and OpenAI are all in the same buying spree. What’s less clear is whether any of this capacity actually arrives on schedule — Vera Rubin chips don’t ship until 2027, and betting billions on unproven startups like Volta and Nscale carries real execution risk. Anthropic is buying insurance against falling behind, but the premiums are getting astronomical.
