Anthropic has agreed to pay SpaceX $45 billion over three years for computing capacity to power its Claude AI models, according to a securities filing disclosed Wednesday, as reported by Bloomberg. The contract runs $1.25 billion per month through May 2029 and covers capacity at two SpaceX data centers, including the Colossus 1 facility in Memphis.
The deal is the clearest signal yet of how much raw compute is required to operate a frontier AI lab at scale. At $1.25 billion per month, Anthropic’s SpaceX contract alone represents roughly $15 billion in annual compute spend. For context, Anthropic raised $3.5 billion in its most recent funding round. The company is now committing more to a single compute vendor each quarter than it raised in that round.
The structural terms deserve attention. Either party can terminate the agreement with 90 days’ notice. That clause is standard in enterprise compute contracts, but it complicates the “nearly $45 billion” framing. The full value only materializes if both sides hold for all 36 months. SpaceX has an incentive to stay in; Anthropic has an incentive to stay in so long as Claude’s compute demand justifies the cost. The 90-day exit is a floor, not a ceiling.
The Memphis data center known as Colossus 1 was the initial site, with capacity starting in an earlier deal inked earlier this month. This expanded agreement adds a second SpaceX data center and ramps capacity in May and June 2026 at a reduced fee, according to the SpaceX IPO filing. The ramp structure suggests Anthropic is absorbing incremental capacity as SpaceX builds it out, not buying a fully provisioned block.
Comparing this contract to other frontier-lab compute arrangements puts the number in sharper relief. OpenAI and Oracle announced a deal in early 2025 valued at up to $300 billion over five years, or roughly $60 billion per year at the headline rate. Anthropic’s SpaceX contract annualizes at $15 billion. OpenAI’s revenue trajectory and Microsoft’s existing Azure commitment give it a different cost structure, but the comparison illustrates the compute-cost ceiling that any frontier lab now operates under.
Anthropic’s other compute commitments are not fully public. The company has ongoing infrastructure agreements with Google Cloud, which has committed $2 billion to Anthropic and is its cloud partner, and with Amazon Web Services, which invested up to $4 billion. Adding a $15 billion per year SpaceX contract on top of those relationships means Anthropic is balancing multiple infrastructure dependencies simultaneously. Whether this SpaceX deal is additive to its cloud spend or partially replacing it is not disclosed in the filing.
The SpaceX IPO filing is the source. Bloomberg reported the details on May 20. The disclosure is notable because it came through SpaceX’s own paperwork rather than an Anthropic announcement, giving the figures more credibility than a press release would.
For teams evaluating Anthropic’s pricing stability or model availability over the next 18 months, the compute picture matters: Anthropic now has contractual access to two large-scale SpaceX data centers plus its cloud partnerships. Supply is not the near-term constraint. Whether Anthropic’s revenue can cover $1.25 billion per month in compute costs before that 90-day notice clause becomes relevant is the question to watch.
Reported by Bloomberg on 2026-05-20.
