Starting in October, Google will send SpaceX a check for $920 million every single month. The deal runs through June 2029, covering access to roughly 110,000 NVIDIA GPUs — and adds up to about $29.4 billion over the full term. Google's stated reason: "unexpected demand for its recently launched AI products." That's about as candid an admission of unpreparedness as a company this size tends to offer publicly.
The compute is housed in SpaceX's AI data center network, originally built to power xAI. SpaceX didn't name which facility Google will use, but its flagship site — Colossus 1 near Memphis, Tennessee — is already fully committed to Anthropic. That deal, struck in late May, runs to $1.25 billion per month through the same 2029 end date and covers all available capacity at Colossus 1. Google gets roughly half those resources. SpaceX's second complex, Colossus 2, was originally announced for xAI's own operations.
Add both contracts together and SpaceX now has $2.17 billion per month locked in from just two AI customers — more than $26 billion a year. For a company whose primary businesses are rockets and satellite internet, that's an extraordinary revenue stream, and one that required no new cloud platform to build. SpaceX constructed these data centers for its own AI ambitions and is now monetizing the capacity at rates that would make any cloud provider envious.
There are exit clauses on both sides. Either party can terminate with 90 days' notice after December 31, 2026. If SpaceX doesn't deliver the committed GPU count by September 30, Google can walk away immediately or accept reduced capacity with a proportional fee cut. That flexibility signals this is bridgework while Google scales its own infrastructure — not a permanent dependency on Musk's data centers.
SpaceX is meanwhile preparing for an IPO targeting a $1.75 trillion valuation. Locking in two of the world's most prominent AI labs as paying customers through 2029 creates the kind of predictable, high-margin recurring revenue that institutional investors find compelling. For Google, the picture looks different: a company with massive TPU investments and a global data center footprint still ran short on GPU capacity and had to go shopping. Whether SpaceX can simultaneously fulfill both commitments without degrading service is a question nobody in this arrangement is asking publicly yet.
