SpaceX rented Colossus 1 to Anthropic because it couldn’t make the data centre work for Grok

The business move

SpaceX rented out its Colossus 1 data centre to Anthropic after struggling to get the facility to work for its own AI needs. Instead of having surplus capacity, SpaceX faced latency problems linking the Memphis location to its other data centres more than 10 miles away. This connectivity issue prevented SpaceX from running its Grok AI models effectively, forcing the company to lease the space to Anthropic.

Why it matters

Data centre latency can make or break AI performance, especially for real-time or large-scale models. SpaceX’s failure to integrate Colossus 1 into its AI infrastructure exposes critical challenges in scaling AI hardware beyond cloud providers. It shows that owning advanced computing assets still comes with significant operational risks, time, and costs. For companies building or expanding AI infrastructure, this highlights the importance of network architecture and integration capability—not just raw compute power.

Who gains and who gets squeezed

Anthropic benefits by gaining access to a high-profile data centre best suited for AI workloads that SpaceX could not yet optimize. It’s a win for Anthropic’s infrastructure ambitions without having to build or source its own from scratch. SpaceX, meanwhile, must pause or rethink how it scales AI internally, which could slow down Grok development. Competing AI ventures watching this might question the reliability of in-house data centres versus cloud providers for running the latest models.

What to watch next

Watch if SpaceX improves Colossus 1’s network and latency issues and reclaims the facility for its AI models. This situation also puts a spotlight on Anthropic’s AI development speed and infrastructure choices. Investors and operators should track whether other AI builders face similar connectivity roadblocks with owned data centres and how they pivot strategies for infrastructure investment.

AI Quick Briefs Editorial Desk