On June 5, 2026, SpaceX disclosed a new compute contract in an SEC filing: Google will pay $920 million per month from October 2026 through June 2029 for access to roughly 110,000 NVIDIA GPUs, along with CPUs, memory, and related components. Reuters confirmed the multi-year cloud services agreement. The disclosure landed exactly one week before SpaceX stock is expected to begin trading on Nasdaq.
Three things make this deal worth your attention. First, Google — which Epoch AI estimates is the largest single owner of AI compute on the planet — is renting GPUs from a rocket company. Second, stacked on top of the $1.25 billion per month Anthropic agreed to pay in May, SpaceX’s compute leasing business now pulls in roughly $2.17 billion a month. Third, it confirms that under power and delivery constraints, large-scale compute has become a strategic commodity you can lease by the fleet.
The Deal: $920M a Month, 110,000 GPUs, 90-Day Exits
Per the filing, as reported by TechCrunch:
- Term and price: $920 million per month, October 2026 through June 2029
- Scope: approximately 110,000 NVIDIA GPUs, plus CPUs, memory, and related components
- Delivery deadline: SpaceX must deliver the GPUs by September 30, 2026, with a one-month grace period; after that, Google can terminate immediately or accept fewer GPUs at reduced monthly fees
- Exit: either party can walk away with 90 days’ notice after December 31, 2026
- Ramp-up: Google’s access scales up through September at a reduced fee
SpaceX did not say which facility Google will use. Musk has previously said Colossus 2 is reserved for xAI’s own workloads, and the May deal already handed Anthropic every available cycle of Colossus 1 near Memphis, Tennessee. TechCrunch estimates Google’s allotment is roughly half of what Anthropic holds at Colossus 1 — meaning the capacity likely comes from new builds, or from facilities that have not been disclosed.
Why Google, the Largest Compute Owner, Is Renting
A Google representative framed the deal as a “short-term, timely agreement” providing “bridge capacity” for Gemini Enterprise, the company’s agent platform, whose demand exceeded expectations. The surrounding numbers explain the urgency: Alphabet has committed to more than $180 billion in capital expenditures this year, expects that to “significantly increase” in 2027, and recently announced an $80 billion equity sale to help fund the spending. In-house capacity is expanding as fast as money allows — just not as fast as inference demand.
The contrast with Anthropic matters. Anthropic rented Colossus 1 because it was genuinely compute-constrained; by April it was already admitting infrastructure strain. Google is not short of long-term capacity. It is short of capacity right now, and paying cash to a faster-moving third party for time is the entire point of the contract. Musk’s own explanation is blunter: companies are turning to SpaceX because power is tight, and SpaceX builds its own plants and moves faster.
SpaceX’s Compute Business: The IPO’s Second Act
Add the two contracts together — $1.25 billion a month from Anthropic, $920 million from Google — and SpaceX is sitting on roughly $26 billion a year of contracted compute revenue, running through 2029. For a company about to raise roughly $75 billion at a valuation near $1.75 trillion, which would make it the largest IPO in history, these multi-year contracts are concrete proof of an AI second growth curve for the prospectus.
Two side signals are worth noting. Google is a longtime SpaceX investor, and Bloomberg estimates its stake will be worth more than $100 billion after the IPO — tenant and shareholder at once. Separately, the Wall Street Journal reports the two companies are in talks on orbital data centers, described as a major SpaceX initiative after the listing.
What It Means for Developers and Product Teams
Three takeaways. First, “bridge capacity” is now a standard procurement tool: when Google will pay $920 million a month for a 33-month stopgap, it tells you that inference demand spikes outrun any capacity plan — so treat rented transitional capacity as a legitimate architectural assumption, not a hack. Second, the supplier list is changing: a newcomer that builds its own power and delivers fast can simultaneously serve customers the size of Anthropic and Google. Third, owning silicon and renting capacity are not mutually exclusive — OpenAI is pushing its own Broadcom-designed chip while still shopping for outside capacity, and Google runs its own TPUs alongside these rented GPUs. In 2026, there is no single answer to the compute strategy question.
Sources
- Google will pay SpaceX $920M per month for compute — TechCrunch
- SpaceX lands Google AI compute deal after Anthropic pact ahead of IPO — Reuters
- Google to pay $920 million to SpaceX-xAI monthly for AI data center capacity — Datacenter Dynamics
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.