Google Signs $920M Monthly Compute Deal With SpaceX

SpaceX disclosed in a regulatory filing that Alphabet's Google will pay $920 million per month from October 2026 through June 2029 for access to approximately 110,000 Nvidia GPUs, along with CPUs, memory, and other components, according to Bloomberg, CNBC, The New York Times, and TechCrunch. CNBC reports the capacity is hosted at xAI data centers. The filing describes a reduced-fee ramp while SpaceX brings hardware online and a delivery clause that lets Google terminate, or accept fewer GPUs at a pro-rata fee reduction, if SpaceX does not meet the GPU commitment by September 30, 2026. At the stated monthly rate, outlets including Bloomberg and the NYT calculate the contract at roughly $30 billion over its term.

A roughly $30 billion, multi-year compute contract is one of the largest disclosed arrangements of its kind and signals that even a hyperscaler with its own data centers will rent large external GPU fleets to bridge short-term demand. CNBC reports a Google Cloud spokesperson framed the purchase as bridge capacity for surging demand for its Gemini Enterprise agent platform. For practitioners planning capacity and cloud economics, it underscores how tight accelerator supply has become and how AI demand is reshaping infrastructure sourcing.

Reporting in TechCrunch, Bloomberg, and CNBC places the deal alongside a similar large compute arrangement between SpaceX and Anthropic announced in late May, and notes the capacity sits at xAI data centers tied to Elon Musk's AI operation. SpaceX disclosed the Google contract in its filing ahead of a planned IPO that multiple outlets expect to be among the largest ever, making contracted, recurring compute revenue a notable part of its investment story.

Class B analysis: Buying third-party data-center capacity to bridge peak demand is a growing pattern as training and inference workloads spike. Large offsite deals commonly raise engineering questions around network egress, workload colocation, and mixed-generation GPU fleets, so teams typically validate performance across hardware mixes and plan for staged ramps rather than instant availability.

Further filing detail on which data centers host the capacity and the mix of GPU generations; published utilization or benchmark results from early workloads; and how SpaceX's IPO pricing reflects the value investors place on contracted, recurring compute revenue relative to its launch and Starlink businesses.