Google commits $920 million per month to SpaceX for 110,000 NVIDIA GPUs to meet surging Gemini Enterprise demand

Google has agreed to pay SpaceX $920 million per month from October 2026 through June 2029, securing access to approximately 110,000 NVIDIA GPUs housed in a SpaceX data center facility, according to a regulatory filing SpaceX disclosed on June 5. The deal is explicitly framed as a capacity bridge: Google attributed the arrangement to demand for Gemini Enterprise that has outpaced the company's internal compute forecasts.

What's new

The agreement runs roughly 33 months at the stated rate, with a ramp-up period at reduced prices through September 2026. The full GPU count must be accessible by September 30 or Google has the right to terminate within 30 days following a one-month grace period, according to the filing. Either party can exit the agreement with 90 days' notice after December 31, 2026.

The GPU inventory is primarily NVIDIA data-center hardware. The specific facility was not disclosed, though the arrangement is structurally similar to SpaceX's concurrent compute deal with Anthropic—which runs at $1.25 billion per month and provides roughly twice as much capacity.

Google explained the decision in a statement: "This is a short-term, timely agreement to ensure we have bridge capacity to meet surging customer demand for our agent platform, Gemini Enterprise, which has been even higher than we expected."

Context

SpaceX built and operates the Colossus data center in Memphis, Tennessee, originally to support xAI's Grok model training. After securing Anthropic as a compute tenant, SpaceX has turned that infrastructure into a commercial GPU rental business aimed at AI labs that need capacity on timelines shorter than building data centers allows.

Google is by most estimates the world's largest single owner of AI compute capacity—its custom TPU fleet and NVIDIA GPU inventory are enormous by any measure. The decision to rent from SpaceX rather than accelerate its own build-out is a direct signal of how quickly AI demand is outpacing even the most aggressive internal infrastructure programs.

Why it matters

This deal puts a concrete dollar figure on what it looks like to close a compute gap at Google's scale, and it carries weight for the broader AI infrastructure market. At $920 million per month, the arrangement is the second-largest known AI compute rental in history (after Anthropic's deal with SpaceX). It validates SpaceX's ambition to become a neutral AI infrastructure provider. For Gemini Enterprise customers, it is a hard commitment from Google that capacity constraints will not bottleneck service availability through mid-2029. And for the ecosystem, it confirms that third-party compute markets—SpaceX, CoreWeave, Lambda Labs, and others—remain commercially essential even for hyperscalers, sustaining demand for GPU-rental infrastructure as a viable business.