SpaceX becomes AI compute landlord as Cursor, Anthropic, and Google sign GPU deals
From 220,000 GPUs for Anthropic to orbital data centers with Google, SpaceX controls the AI compute bottleneck.
SpaceX AI is quietly emerging as the dominant infrastructure provider for the AI industry, leasing massive GPU clusters to model builders who can no longer own the full stack. Cursor, the coding assistant startup, was the first to hit a compute ceiling — their growth outpaced access to training infrastructure, so they turned to SpaceX's Colossus data center to train their composer models. The deal gave Cursor the capacity to scale without building their own hardware.
Anthropic went even bigger. After struggling to meet developer demand and imposing aggressive rate caps, they secured access to over 220,000 NVIDIA GPUs across 300MW of capacity at Colossus 1. The deal was so massive that SpaceX AI moved its own training workloads to Colossus 2. Meanwhile, Google is exploring a project called 'Suncatcher' — talks with SpaceX to launch data centers into orbit via rockets, an extreme bet on latency and energy advantages. The pattern is clear: infrastructure players like SpaceX AI, Google, and Amazon are decoupling compute ownership from model development, making GPU access the new leverage point in AI.
- Cursor used Colossus to overcome compute ceiling for training composer coding models after growth outpaced infrastructure access.
- Anthropic secured 220,000+ NVIDIA GPUs across 300MW at Colossus 1, prompting SpaceX AI to move own training to Colossus 2.
- Google's 'Suncatcher' project involves talks with SpaceX for rocket-launched orbital data centers.
Why It Matters
Access to massive GPU clusters is now a strategic asset, reshaping AI infrastructure ownership and partnerships.