Anthropic signs Colossus compute deal with SpaceX — 100,000+ NVIDIA H100s in Memphis target multi-quarter Anthropic training capacity
Anthropic announced on May 6 a compute deal to use SpaceX's Colossus supercomputer in Memphis, Tennessee — a 100,000+ NVIDIA H100 GPU facility — for AI model training. The deal extends Anthropic's compute footprint beyond AWS and signals SpaceX's emergence as a third-party compute supplier to frontier labs.
Memphis is the meaningful geographic detail. The Colossus facility was originally built to support xAI's Grok model training; SpaceX's all-stock merger with xAI (completed February 2026) brought the data center into the SpaceX portfolio. The Anthropic deal demonstrates that the post-merger SpaceX entity is selling third-party compute capacity — the facility is no longer captive to a single lab.
The strategic implication for the frontier-lab compute market is meaningful. Anthropic's prior compute story was AWS-led with Trainium-and-H100 mix; the Colossus deal opens a second-source capacity line that does not depend on AWS scheduling. Combined with SpaceX's S-1 filing targeting $75B raise, the post-merger SpaceX+xAI entity is positioning to compete with AWS, Microsoft Azure, and Google Cloud for frontier-lab compute spend.
Anthropic — Compute partnership announcement → · Reuters — Anthropic SpaceX Colossus compute deal Memphis →