Anthropic on Tuesday announced a wide-ranging compute deal with SpaceX under which it will use all of the available capacity at the Memphis-based Colossus 1 data centre, giving the company access to more than 300 megawatts of new power and over 220,000 Nvidia accelerators within the month, according to a statement from the company's developer conference in San Francisco.
The agreement immediately translates into higher usage limits for Claude's paid tiers. Claude Code's five-hour rate cap is being doubled for Pro, Max, Team and seat-based Enterprise customers, peak-hour reductions are being lifted for some accounts, and rate limits on the company's API for Claude Opus models have been raised significantly.
Anthropic said its enterprise revenue base grew roughly eighty-fold over the past quarter, and that the SpaceX agreement was needed to keep pace with demand from large customers including in financial services, healthcare and the public sector. The company has separately disclosed plans for international expansion, particularly in regulated jurisdictions where data-residency requirements limit the use of US-hosted infrastructure.
In an unusual extension of the deal, the two companies said they had "expressed interest" in jointly developing multiple gigawatts of compute capacity in space. Earlier on-orbit data-centre proposals from other firms have remained at concept stage; SpaceX's Starship vehicle is widely considered the leading contender to make the economics workable for the first deployments.
The Colossus 1 facility was originally built by xAI, the AI company founded by Elon Musk, which is now being folded into SpaceX as a product line. Industry analysts said consolidating compute under SpaceX gave the combined entity unusually flexible access to power, since it can colocate generation with the campus.
Anthropic continues to contest a Pentagon supply-chain risk designation that has so far excluded it from a separate set of US Department of Defense classified-network awards announced this week. Chief executive Dario Amodei has said the company's acceptable-use policies still allow a broad range of national-security applications.
Other model developers welcomed the deal as supportive of overall industry compute supply, even as some flagged concentration concerns. The Federal Trade Commission said earlier this year it was monitoring large compute partnerships among frontier AI firms but has yet to open any formal probes.