Anthropic Secures 220,000 GPUs from SpaceX’s Colossus Data Center — Eyes Orbital AI Compute

Anthropic has signed a major compute deal with SpaceX, gaining exclusive access to the Colossus 1 data center in Memphis, Tennessee — originally built by Elon Musk’s xAI to train the Grok model. The agreement provides Anthropic with over 300 megawatts of compute capacity and more than 220,000 NVIDIA GPUs, including H100, H200, and the latest GB200 accelerators.

The Colossus 1 facility became available after xAI (absorbed by SpaceX earlier this year) transitioned its training operations to the larger Colossus 2 complex. Anthropic has already leveraged the additional capacity to double rate limits for Claude Code across Pro, Max, Team, and Enterprise plans, remove peak-hour throttling for Pro and Max users, and significantly increase API rate limits for Claude Opus models.

Perhaps most intriguingly, both companies have expressed interest in a long-term collaboration to develop multiple gigawatts of orbital AI compute capacity — signaling a potential future for space-based data centers. The partnership also includes a $1.5 billion joint venture with Blackstone and Goldman Sachs.

Source

Forbes · AI Business · x.ai

Commentary

The irony of Musk — who has publicly criticized Anthropic — renting them his former flagship AI data center says everything about how insatiable the demand for compute has become. Ideology takes a back seat when there’s 300 megawatts of GPU capacity sitting idle.

The orbital compute angle is the sleeper story here. It sounds like science fiction, but SpaceX is one of the few organizations that could actually make space-based data centers economically viable. If power and cooling are the bottleneck for terrestrial AI infrastructure, putting compute in orbit — where solar power is abundant and cooling is free — starts to look less crazy. We’re a long way from that reality, but the fact that two serious companies are talking about it publicly is notable.