Google will pay SpaceX $920M per month for compute

Google Secures Bridge Compute in Multi-Billion Dollar SpaceX Deal

In a significant move highlighting the intense, industry-wide demand for AI compute, Google has entered into a major agreement with SpaceX to rent a substantial cluster of NVIDIA GPUs. Announced in a regulatory filing just a week before SpaceX's historic IPO, the deal commits Google to pay $920 million per month for access to the hardware. The timing underscores the critical nature of securing AI training and inference capacity, demonstrating that even the world's largest cloud providers are seeking external partners to meet immediate demand.

The Deal's Architecture

The agreement provides Google with guaranteed access to a massive hardware installation from late 2026 through mid-2029. This contract is comparable in structure to a similar deal SpaceX recently signed with Anthropic, though it appears to be for roughly half the compute capacity. While SpaceX did not specify which data center would be used, the terms of the deal are clearly defined.

• Monthly Cost: $920 million
• Contract Duration: October 2026 through June 2029
• Hardware Access: Approximately 110,000 NVIDIA GPUs, CPUs, memory, and related components.
• Ramp-Up Period: Access will scale up through September 2026 at a reduced fee.
• Termination Clause: Both parties have the option to terminate with 90 days’ notice after December 31, 2026.

Surging Demand and Strategic Positioning

While Google is already one of the largest single owners of AI compute, a company representative framed the deal as a necessary stopgap. The agreement is described as a “short-term, timely agreement” to provide “bridge capacity” for surging customer demand for its Gemini Enterprise agent platform, which has exceeded expectations. This move occurs as parent company Alphabet is on a massive spending initiative, with over $180 billion in capital expenditures committed for this year. For SpaceX, this deal further establishes its new role as a key wholesale provider of AI compute, turning the infrastructure originally built for its own xAI division into a powerful new revenue stream ahead of its public offering.

The AI compute market has become so constrained that even hyperscalers are now customers for specialized, fast-moving infrastructure providers. This deal signals that the velocity of enterprise AI adoption is outpacing even the most aggressive internal build-out schedules, creating a new and lucrative market for wholesale compute access.