Anthropic taps SpaceX's Colossus 1 supercomputer to lift Claude Code limits

Anthropic struck a deal at its Code with Claude conference to consume the full compute capacity of SpaceX’s Memphis data center, gaining access to more than 300 megawatts of capacity backed by Colossus 1’s 220,000+ NVIDIA GPUs spanning H100, H200, and GB200 accelerators. CEO Dario Amodei framed the agreement as fuel for higher usage ceilings on paid Claude tiers.

On the back of the deal, Anthropic doubled Claude Code’s five-hour window limits for Pro and Max subscribers, scrapped peak-hours throttling for those plans, and raised API limits on the Opus model. The two companies also floated joint interest in building multiple gigawatts of orbital compute capacity, citing terrestrial constraints on power, land, and cooling as a forcing function for off-planet infrastructure.

The partnership is a sharp reversal of tone from Elon Musk, who as recently as February attacked Anthropic on X by amplifying a false claim about its Claude constitution and accusing the company of hostility toward Western civilization. The capacity arrangement now binds two AI rivals into a shared infrastructure dependency at the exact moment frontier model training is becoming a power-and-real-estate problem rather than a chip-availability one.