SpaceX S-1 discloses $1.25B/month Anthropic compute deal through May 2029 — Starlink revenue compounds into AI infrastructure spend
SpaceX filed its S-1 on May 20 ahead of a public listing, disclosing $18.7 billion in 2025 revenue, $6.6 billion in adjusted EBITDA, and a $1.25 billion-per-month compute purchase agreement with Anthropic running through May 2029. The number puts SpaceX-Anthropic at roughly $52.5 billion in committed compute spend over four years — and reframes SpaceX as one of the largest AI-infrastructure customers in the world.
The strategic logic is Starlink. SpaceX's launch and satellite revenue compounds into a balance sheet that can fund AI-scale infrastructure without dilutive capital raises. Anthropic gets a long-dated compute commitment from a customer that uses generative AI internally (Grok competitor narratives notwithstanding, Anthropic's training compute is now partially backstopped by Starlink's free cash flow). The S-1 also discloses Anthropic-trained AI workloads inside SpaceX engineering — which is the part the analyst community didn't know was that operationalized.
The contract structure is unusual at $52.5B / 4 years for a single customer-vendor pair. Microsoft-OpenAI's compute commitments through Stargate are larger in aggregate but spread across infrastructure spend rather than direct compute. SpaceX-Anthropic looks more like Crusoe-OpenAI's Abilene operator deal — direct customer-vendor commitment with no infrastructure-as-middleman markup. The Anthropic public-IPO narrative now has a long-dated revenue floor that bond investors will price.
Singh Ajit — Dev Weekly May 18-24 2026: Google IO Anthropic Stainless SpaceX OpenAI IPO Meta layoffs → · IG — SpaceX OpenAI Anthropic upcoming IPOs 2026 → · NewsBytes — SpaceX and OpenAI plan IPOs as Anthropic nears profit →