Google signed a compute-as-a-service agreement with SpaceX valued at $920 million per month, or roughly $11 billion annually, to secure dedicated access to a GPU cluster SpaceX operates. The arrangement became effective in May and runs through December 2027, with two twelve-month renewal options at Google's discretion. SpaceX will provide compute capacity equivalent to approximately 285,000 H100-class GPUs along with associated CPU and memory infrastructure. Google gains reserved access during contracted windows; SpaceX retains the right to sell unused capacity to third parties during off-peak periods.

The deal represents SpaceX's first commercial offering as an infrastructure lessor. SpaceX built the cluster using revenue from Starlink and defense contracts, investing an estimated $4.2 billion in hardware procurement between late 2024 and early 2026. The company installed the equipment across three ground facilities in Texas, Nevada, and an undisclosed location outside the continental United States. Google's payment structure includes a $1.84 billion upfront capacity reservation fee and the monthly recurring charge. SpaceX disclosed the partnership in a filing required under its ongoing pre-IPO disclosure process, which began in March.

Google's move answers a capacity constraint that became acute after its internal Tensor Processing Unit fabrication timelines slipped in Q4 2025. The company faced a 22% quarter-over-quarter increase in AI inference workloads from Search, YouTube recommendations, and Gemini API calls, but TPU delivery from its primary foundry partner fell 11 weeks behind schedule. Rather than compete for incremental cloud capacity from AWS or Microsoft Azure—where spot pricing for H100 equivalents rose 63% year-over-year—Google structured a dedicated arrangement with a non-traditional provider. The SpaceX cluster gives Google predictable cost and latency, and eliminates the noisy-neighbor problem inherent in shared cloud environments. Effectively, Google paid a 38% premium over comparable multi-tenant cloud rates to secure exclusive windows and service-level guarantees SpaceX can credibly enforce through contract penalties.

For SpaceX, the arrangement generates $33 billion in contracted revenue through the initial term, assuming Google does not exercise early termination clauses. That figure exceeds SpaceX's entire 2025 revenue by $11 billion and substantially de-risks the company's valuation ahead of a rumored IPO in late 2026 or early 2027. The deal also signals that SpaceX views its satellite infrastructure and ground-station real estate as platform assets, not just telecom endpoints. If SpaceX can replicate this model with two or three additional hyperscalers, it enters 2028 with a recurring revenue base that rivals its launch and Starlink businesses combined. The risk is execution: SpaceX has no prior experience operating as a tier-one compute vendor, and cooling, uptime, and security obligations in the Google contract carry steep financial penalties for non-performance.

Operators should monitor SpaceX's customer-acquisition pace in this vertical and whether Amazon or Meta respond with similar arrangements. If SpaceX announces a second major compute customer before September, the company's pre-IPO valuation will likely reset upward by 15% to 20%. Google's Q2 earnings call, scheduled for late July, will clarify whether the company books this expense as capital lease or operating expense, which affects free cash flow calculations. Any disclosure of actual utilization rates—Google is required to report only aggregate compute spend—will indicate whether the deal pencils out or represents expensive insurance.