Anthropic is scaling Claude inference on SpaceX’s Colossus 2 cluster using GB200 capacity throughout June 2026.
Key Takeaways
Anthropic and SpaceX are expanding an existing partnership; GB200 ramp on Colossus 2 begins within days of the May 20 announcement.
Tom Brown framed the compute need as moving “a lot of atoms” to keep up with AI demand, crediting SpaceX’s hardware execution speed.
Claude inference workloads are being migrated to Colossus infrastructure, suggesting significant dependency on third-party GPU clusters for production serving.
Hacker News Comment Review
Commenters are skeptical of the political optics: Musk has publicly criticized Claude as “too woke” yet is now hosting its inference, leading to cynicism about financial motives over ideology.
Several observers interpret xAI renting Colossus capacity to Anthropic and Cursor as a signal that xAI is deprioritizing the AGI race and pivoting Colossus toward revenue generation.
The dominant read is that this is a capital arbitrage play for SpaceX ahead of an IPO, with AI compute rental booked as revenue regardless of which lab’s models run on the hardware.
Notable Comments
@aurareturn: Frames Colossus 1 going to Anthropic and Colossus 2 capacity following as evidence xAI is ceding ground in the AGI race.
