Gemini Enterprise Demand Drives $30B SpaceX GPU Contract

Google has secured an immediate infusion of AI compute capacity, signing a 33-month infrastructure agreement with SpaceX. The deal guarantees Google access to roughly 110,000 NVIDIA GPUs housed within SpaceX data centers. The disclosure appeared in an amended S-1 registration statement filed by SpaceX ahead of its June 12 initial public offering.

Contract Scope and Hardware Details

Google will pay $920 million per month for the compute cluster starting in October 2026. The contract structure includes a phased ramp-up period through September 2026 at a reduced fee. If the agreement runs through its scheduled expiration in June 2029, it will generate over $30 billion in revenue for SpaceX.

The arrangement provides access to hardware originally built for xAI and its Grok models. Google gains utilization of the 110,000 GPUs, associated CPUs, memory, and high-speed networking fabric. Under the contractual terms, Google retains full ownership of its AI models, content, and related data throughout the execution phase.

The Shift in Hyperscaler Infrastructure

Google attributes the hardware acquisition to surging customer demand for Gemini Enterprise. Internal projections significantly underestimated the compute required to support parallel agentic workloads at scale. A Google representative characterized the SpaceX hardware as a short-term bridge capacity while the company builds out its internal infrastructure.

This agreement highlights a severe supply bottleneck in the AI compute market. Google typically relies heavily on its own data centers and custom Tensor Processing Units. Renting third-party NVIDIA capacity at a $30 billion scale indicates that immediate deployment speed currently overrides the margin benefits of proprietary silicon.

The contract mirrors the recent $45B Colossus deal where Anthropic secured exclusive access to SpaceX hardware in Memphis.

Metric	Google Agreement	Anthropic Agreement
Monthly Cost	$920 million	$1.25 billion
Hardware Allocation	110,000 NVIDIA GPUs	220,000 NVIDIA GPUs
Contract Duration	33 months	Multi-year

Timeline and Termination Clauses

The SEC filing details strict delivery and termination clauses for the massive deployment. SpaceX must provision the committed hardware capacity by September 30, 2026. Failure to meet this deadline triggers a one-month grace period. After that window, Google can immediately terminate the agreement or accept a pro-rata reduction in monthly fees.

The contract also includes an exit hatch for both organizations. Starting December 31, 2026, either party can cancel the agreement with 90 days of notice.

For developers building on top of Google infrastructure, this capacity squeeze signals that base model inference and agent orchestration will remain highly resource-constrained through the end of the year. You should architect systems to reduce LLM API costs in production and handle strict rate limits, as hyperscaler compute shortages directly impact downstream API availability and latency.