Anthropic has expanded its compute partnership with SpaceX, committing to roughly $1.25 billion per month through the end of 2029 for guaranteed access to the rocket and satellite company's Colossus supercomputing cluster. The headline figure was disclosed in the SpaceX IPO prospectus on Tuesday and confirmed by Anthropic in a brief blog post.
The deal totals more than $60 billion in committed payments through the contract's term, making it one of the largest single compute commitments ever announced and the anchor customer for the Colossus build-out. Colossus, located on three sites in Tennessee and Texas, is among the largest contiguous Nvidia Hopper and Blackwell deployments outside the major hyperscalers.
The arrangement is structured around a guaranteed annual capacity floor of roughly 12 gigawatts of equivalent training capacity, with the right of first refusal on incremental Colossus expansions over the contract's term. The technical specifics — chip mix, networking topology and the share of new versus second-life GPUs — are redacted in the prospectus and the Anthropic post.
For Anthropic, the deal converts a major recurring cash outflow into a long-dated contract, providing the visibility required for the increasingly debt-heavy financing structure that has paid for Claude's training runs over the past eighteen months. For SpaceX, it underwrites the Starlink-adjacent terrestrial-compute business that Mr Musk has been building since 2024.
It also realigns the competitive landscape: Anthropic is moving towards single-vendor compute concentration at a moment when its largest rival OpenAI is splitting compute across Microsoft Azure, Oracle, CoreWeave and its own Stargate build-out. Industry analysts at SemiAnalysis described the trade-off as "concentration for capability versus diversification for resilience".