SpaceX announced on Wednesday that it has leased its Colossus 1 supercomputer to Anthropic. The partnership grants the AI developer access to a massive cluster of graphics processing units and significant electrical capacity. This infrastructure supports Anthropic’s efforts to scale its Claude models and enhance developer tools.
Anthropic gains massive compute capacity to scale Claude models
The Colossus 1 facility houses over 222,000 Nvidia GPUs. The hardware mix includes H100 and H200 chips alongside next-generation GB200 accelerator systems. The center delivers more than 300 megawatts of compute power. This energy output is sufficient to power 300,000 homes according to performance claims.
Spec comparison
| Spec | Colossus 1 |
|---|---|
| GPU Count | 222,000 Nvidia GPUs |
| GPU Models | H100, H200, GB200 |
| Compute Power | 300 megawatts |
Anthropic is utilizing this increased capacity to adjust service limits for its users. The company doubled the rate limits for Claude Code to 10 hours for all paid tiers. It also removed peak hour restrictions for Pro and Max subscribers. API rate limits for Claude Opus models received considerable increases to support higher request volumes.
Elon Musk stated he approved the lease after reviewing Anthropic’s safety protocols. He noted that he was impressed by their methods for ensuring Claude benefits humanity. Musk remarked that no one set off his evil detector during these evaluations. The deal marks a reversal of previous criticisms Musk held toward the company.
SpaceX claims that terrestrial power, land, and cooling resources cannot meet the timeline requirements for next-generation AI systems. Anthropic has expressed interest in developing orbital data centers with SpaceX to address these constraints. The company also unveiled a new dreaming feature for Claude agents to improve self-reflection between sessions.
Discussion
0 comments
Log in to join the thread with a thoughtful take, question, or correction.