SpaceX is moving beyond rocket science. The company is now leasing access to its massive Colossus 1 compute cluster to Anthropic under a "compute as a service" model. The deal provides Anthropic with roughly 300 megawatts of power and about 220,000 GPUs for AI training and inference workloads. To put that in perspective, a typical hyperscale data center runs on about 100 megawatts; SpaceX is offering triple that in a single cluster.
The arrangement allows Anthropic, the maker of the Claude AI models, to bypass the significant bottleneck of building and powering its own data centers. For SpaceX, the deal creates a new revenue stream from infrastructure it already operates. The company has also signaled ambitions to eventually build orbital data centers, scaling to multiple gigawatts of space-based compute.
This development has implications for the broader AI compute market, including decentralized networks like Render and Akash, which have built their offerings around the same GPU scarcity. The concentration of such massive compute resources in the hands of a few players could raise the barrier to entry for new AI labs.