SpaceX Google deal: $US920m a month for AI compute
SpaceX Google deal puts a $US920m-a-month price on AI compute as Gemini demand pushes cloud buyers to lock in scarce GPU capacity.

SpaceX has signed a large AI compute agreement with Google days before its planned IPO, putting a monthly price on scarce graphics processors. The rocket and satellite company said in a free writing prospectus that Google will pay $US920 million (about $1.4 billion) a month for cloud-services capacity from October 2026 to June 2029.
The contract covers about 110,000 NVIDIA GPUs. For SpaceX, that turns part of an infrastructure build-out into booked revenue rather than spare capacity. For Google, it buys time while demand for Gemini Enterprise and agent workloads runs ahead of its own data-centre schedule, a gap that cannot be solved quickly if power, chips and networking are all tight.
SpaceX presented the deal as evidence that compute is becoming a commercial asset in its own right, not merely an internal cost for training and serving models.
“We believe our compute infrastructure and related strategy provides us with substantial flexibility in how we allocate and monetize capacity.”
Source: SpaceX free writing prospectus
Reuters reported that the Google agreement follows a separate Anthropic pact and lifts the combined annual value of the two contracts to about $US26 billion (about $40 billion). The wire service put the aggregate disclosed value above $US70 billion (about $108 billion), a large figure for a company still asking public-market investors to back a capital-intensive growth story.
Why the capacity matters
Large AI buyers are no longer only comparing model quality, API pricing or cloud discounts. They are trying to reserve clusters that can support live agent products, where usage can spike faster than a new data centre can be permitted, powered and fitted out. In that market, capacity that looks surplus in one quarter can become leverage in the next.
CNBC quoted a Google Cloud spokesperson saying the SpaceX capacity would help meet demand that had exceeded forecasts.
“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.”
Source: Google Cloud spokesperson, CNBC
For Australian enterprise buyers, the immediate effect is indirect. The deal does not mean local Gemini customers get new capacity in Sydney next quarter, and it should not be read as a guide to domestic cloud pricing. Procurement teams are more likely to feel it in feature rollouts, usage limits and the premium charged for guaranteed capacity.
That caveat is important. A US pre-IPO contract does not reset the funding market for Australian cloud or AI start-ups. It does show why the global hyperscalers keep spending heavily on GPUs, energy procurement and regional infrastructure: the constraint for AI services is increasingly physical supply, not software demand alone.
The timing also gives SpaceX a cleaner investor story. Ahead of an IPO, recurring cloud-services revenue can sit beside launches, Starlink subscriptions and long-range space projects. Execution risk remains, including delivery of the capacity, customer concentration and the economics of running large GPU clusters. The Google contract still gives the prospectus a useful second line: SpaceX can sell scarce compute while it builds the space business.
Soren Chau
Enterprise editor covering AWS, Azure, and GCP in the AU region, plus the SaaS shaping local IT. Reports from Sydney.
Related

Anthropic locks in SpaceX Colossus 1 for 300MW of new Claude compute

Nvidia earnings show the AI chip boom is still accelerating

IREN signs $US3.4b Nvidia AI cloud contract, founder slams Australian regulation

AWS and Versent sign five-year AI cloud pact
