The Anthropic's compute deal that nobody saw coming

On a surprising turn of events, Anthropic has signed an agreement to use the full computing capacity of SpaceX's Colossus 1 data center in Tennessee: over 200 thousand Nvidia GPUs delivering more than 300 MW of raw compute. The deal takes effect within the month, and its impact for paying users is immediate: five-hour rate caps lifted, peak-hour usage restrictions removed, and doubled rate limits for Claude Code on paid plans.

Colossus 1 packs H100s, H200s, and GB200 accelerators in one of the densest GPU clusters EVER assembled, and Anthropic is renting the whole thing, not only a slice of it. This comes on top of an already aggressive infrastructure push: a multi-gigawatt agreement with Amazon (with nearly 1 GW arriving before end of 2026), a 5 GW deal with Google and Broadcom, and a $30 billion Azure partnership with Microsoft and Nvidia. Anthropic's CEO, Dario Amodei, had publicly acknowledged that demand for Claude had created "inevitable strain on infrastructure", particularly during peak hours. The SpaceX deal is the most visible answer to that strain yet.

Interestingly, SpaceX competes directly with Anthropic through Grok, and Musk spent much of early 2026 publicly criticizing Anthropic, at one point calling the company a threat to Western civilization. The turnaround is real, if awkward: Musk spent a week with Anthropic's senior team, came away saying no one "set off his evil detector," and added a clause giving SpaceX the right to reclaim compute if Claude "engages in actions that harm humanity". That's an unusual contractual provision, but for a company whose entire thesis is that AI development should prioritize safety, having a compute partner who at least nominally shares that framing is not nothing.

And additional compute, in Anthropic's hands, more than a commercial win is infrastructure for the kind of model development their work requires. Anthropic isn't a company that treats safety as (pure) marketing; their model welfare research genuinely requires frontier level resources. The orbital data center ambition - Anthropic has expressed interest in partnering with SpaceX to develop multiple GW of space compute - is still TBD, but if it materializes, it raises the question on whether putting AI infrastructure in orbit changes anything in the long run.