Anthropic Taps SpaceX's Colossus 1 for 220,000 GPUs and Eyes Orbital Compute
Anthropic has secured the entire compute capacity of SpaceX's Colossus 1 data center in Memphis — over 300 megawatts and 220,000 NVIDIA GPUs — and the two companies are exploring multiple gigawatts of orbital AI compute.
Anthropic has signed a deal to take all of the compute capacity at SpaceX's Colossus 1 data center in Memphis, Tennessee, gaining access to more than 300 megawatts of new capacity and over 220,000 NVIDIA GPUs — including H100, H200, and next-generation GB200 accelerator systems — with the additional capacity arriving by the end of May. The agreement, announced May 6 alongside an Anthropic blog post, marks one of the largest single-site compute leases in the AI industry's short history.
The arrangement is notable not just for its scale but for its participants. Colossus 1 was built by xAI, the rival lab founded by Elon Musk, who said he greenlit the lease only after spending time with Anthropic's senior leadership to "understand what they do to ensure Claude is good for humanity." His verdict, reported widely after the announcement: "No one set off my evil detector." The deal layers atop Anthropic's earlier commitments to AWS Trainium, Google TPUs via a five-year $200 billion arrangement, and a $30 billion Microsoft/NVIDIA Azure partnership.
Anthropic is using the new capacity to immediately raise customer-facing limits. The company doubled the five-hour rate limits on Claude Code for Pro, Max, Team, and seat-based Enterprise plans; removed the peak-hours limit reduction that previously throttled Pro and Max accounts during heavy demand; and substantially raised API rate limits for Claude Opus models. The company framed the changes as long-overdue relief for power users who had been hitting ceilings on agentic coding workflows.
The most striking element of the partnership, however, looks past terrestrial infrastructure entirely. Anthropic and SpaceX said they have "expressed interest" in developing multiple gigawatts of orbital AI compute capacity — a previously theoretical concept that pairs SpaceX's launch economics with the cooling and solar advantages of free space. Neither company committed to a timeline or capital figure, but the language goes further than any prior public discussion of space-based training clusters from a frontier lab.
For Anthropic, the deal underwrites the compute demands of a roadmap that has pushed aggressively into agentic coding, cybersecurity, and enterprise workloads — categories where the lab has gained ground on OpenAI through 2026. For SpaceX and xAI, leasing Colossus 1 to a direct competitor signals that the economics of AI data centers now reward whoever can fill capacity fastest, even at the cost of strategic positioning. The next test is whether Colossus 2 — reportedly far larger — ends up serving xAI exclusively or follows the same multi-tenant playbook.