Anthropic on Wednesday announced a sweeping compute partnership with Elon Musk's SpaceX, taking over the entire capacity of the Colossus 1 data center in Memphis, Tennessee, and signaling interest in eventually moving AI workloads off the planet entirely.
The deal gives Anthropic access to more than 300 megawatts of new capacity and over 220,000 Nvidia GPUs, with the company saying the capacity is coming online within the month. Anthropic is also raising usage limits on its Claude API and doubling Claude Code rate limits for Pro, Max, Team, and Enterprise plans as part of the rollout.
A frenemy turned compute supplier
The most striking part of the deal is who is on the other side of it. Musk has spent much of the past year as one of Anthropic's loudest critics, posting on X in February that the company "hates Western civilization." He is also in the middle of a long-running lawsuit against OpenAI, where he is a co-founder and former backer.
In a post on X around the announcement, Musk struck a different tone, saying he had spent significant time with senior Anthropic leadership over the past week and was "impressed." Bloomberg, CNBC, NBC News and Al Jazeera all reported the deal as a notable softening between Musk and another frontier AI lab — and a meaningful boost for SpaceX's still-young data center operation, which had not previously been associated with serving an external Anthropic-scale workload.
Why Anthropic needs the GPUs
The scale of the agreement reflects how quickly Anthropic's compute appetite is growing. CEO Dario Amodei has said the company grew roughly 80x in the most recent quarter, and Anthropic has spent the past several weeks lining up infrastructure on every front: a $200 billion, five-year cloud commitment with Google announced earlier this week, expanded TPU capacity through Google and Broadcom, and ongoing GPU buildouts with hyperscaler partners.
Adding all of Colossus 1 in one stroke is significant because it is a single, coherent training-grade cluster rather than fragmented capacity stitched across regions. For training and serving frontier models like Claude Opus 4.7, dense GPU pools concentrated in one site materially reduce networking overhead.
The space angle
Anthropic also said it has "expressed interest" in working with SpaceX to develop multiple gigawatts of compute capacity in space. That part of the announcement is aspirational rather than contractual — there is no signed buildout, schedule or cost — but it is the first time a frontier AI lab and a launch provider have publicly aligned around the idea of orbital data centers as a serious compute strategy.
What it means
For Anthropic, the deal is another sign that compute, not algorithms, is the binding constraint on its growth, and that it is willing to source from any provider that can deliver power and GPUs at scale — including one whose owner has publicly attacked it. For SpaceX, Colossus 1 just landed an anchor tenant whose demand profile rivals OpenAI's. For Musk, the partnership is a reminder that in the AI buildout, business pragmatism is increasingly outrunning personal feuds.
