What deal links Google and SpaceX compute?

Google and SpaceX: the compute contract

Google has agreed to pay SpaceX about $920 million per month for compute capacity, in a cloud-services deal tied to access to roughly 110,000 Nvidia GPUs and related infrastructure. The arrangement is described as “short-term” in one report, framed as a bridge to cover surging customer demand for Google’s Gemini Enterprise.

The structure highlights how tightly model demand and chip supply have become coupled. Instead of waiting for internal capacity buildouts or third-party capacity reservations, Google is contracting for direct access to GPU infrastructure through SpaceX, effectively outsourcing part of the scaling problem to an external infrastructure provider.

The agreement is also linked to SpaceX’s broader IPO timeline and visibility as an infrastructure player, not just a rocket company. Regulatory filings and reporting around the deal describe it as running through mid-2029, which—if accurate—suggests Google is not merely experimenting but planning for sustained compute needs.

Why it matters

Compute is the limiting reagent for training and deploying frontier AI systems. A multi-year, high-dollar compute contract like this signals:

• Gemini Enterprise demand is ramping faster than available in-house capacity
• GPU supply and infrastructure contracts are becoming strategic and financial
• SpaceX could build a new customer base on top of its traditional launch business

For the market, these contracts can shape expectations about who has the fastest path to scale model offerings.

• Monthly cost is very large
• Includes large-scale GPU access
• Used to bridge near-term demand