Google to pay SpaceX $920 million a month for xAI GPU capacity, filing shows

Google will pay SpaceX $920 million per month from October 2026 through June 2029 for access to roughly 110,000 NVIDIA GPUs and related components at SpaceX's xAI data centers, according to a SpaceX SEC filing. The arrangement is worth approximately $30 billion over its term, a sum that signals how acute the scramble for AI compute has become.

The deal surfaced in filings tied to SpaceX's planned market debut; SpaceX entered the cloud service agreement with Google on June 5, 2026, roughly a week before an expected Nasdaq listing reported to target a valuation of about $1.75 trillion. Neither Google nor SpaceX published a standalone announcement, and the figures here derive from that regulatory filing rather than a company press statement.

What makes the deal striking is the direction of the compute. Google, which builds its own AI chips, is renting capacity sited at xAI's facilities — Elon Musk's AI venture — to feed its enterprise products. Google Cloud has described the agreement as bridge capacity to meet customer demand for its Gemini Enterprise agent platform, framing the rental as a stopgap rather than a strategic shift.

The terms come straight from the filing. Google's access ramps through September 2026 at a reduced fee, and if SpaceX fails to deliver the committed GPU count by September 30, 2026, Google may terminate or accept reduced capacity after a one-month grace period, the filing states. Either side can exit with 90 days' notice after December 31, 2026, and Google retains ownership of its content, AI models and related data.

The deal is the second large SpaceX compute commitment to surface in weeks. Anthropic agreed in late May 2026 to pay SpaceX about $1.25 billion per month through 2029 for compute from its Colossus 1 data center near Memphis, Tennessee — turning SpaceX's xAI infrastructure into a wholesale supplier to rival AI labs.

The pattern points to a compute market tight enough that even chip-designing giants are leasing GPUs from a competitor's data centers to keep their AI services running.