Google signs $920M-a-month compute deal with SpaceX for xAI capacity
SpaceX disclosed a Google Cloud agreement worth roughly $920M a month for about 110,000 Nvidia GPUs at xAI-tied data centers, running October 2026 through June 2029. It is the second mega-compute deal SpaceX has booked in weeks, after a $1.25B-a-month Anthropic contract.
SpaceX disclosed on June 5, 2026 that Google Cloud will pay roughly $920 million a month for compute capacity at xAI-tied data centers, in a contract running from October 2026 through June 2029. The deal covers about 110,000 Nvidia GPUs plus supporting CPUs, memory, and networking, and lands in the amended S-1 ahead of what filings frame as the largest IPO ever attempted. Across the life of the contract, payments add up to roughly $30 billion. It is the second mega-compute deal SpaceX has booked in weeks, and the most explicit signal yet that even the companies best positioned to build their own AI capacity cannot build it fast enough.
What the deal covers
The contract ramps through September 2026 at reduced fees before the full monthly payment begins in October. Termination clauses are unusually firm for an arrangement of this size: if SpaceX fails to deliver committed GPU capacity by September 30, Google may exit after a one-month grace period or accept the available chips with proportionally reduced payments. After December 31, 2026, either party may terminate with 90 days’ notice. The flexibility runs both ways, which suggests both sides are pricing in the possibility that GPU supply (or demand) shifts before the contract runs its course.
Google framed the agreement narrowly. “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,” a Google Cloud spokesperson told CNBC. That positioning, capacity-backfill rather than long-term partnership, matters for how the deal is read: Google appears to be renting capacity it intends to bring in-house, not outsourcing its compute strategy.
SpaceX did not name the data center. CEO Elon Musk has previously suggested Colossus 2 will be reserved for xAI’s own training runs, which implies Google’s allocation lands at Colossus 1 or one of the smaller xAI sites SpaceX absorbed in the all-stock merger earlier this year.
Where this lands in the compute market
The structural read is that the AI-compute market is consolidating into a small number of capacity blocs that can absorb multi-billion-dollar multi-year contracts. SpaceX now sits on two of them. In late May, Anthropic agreed to pay SpaceX $1.25 billion a month through 2029 to rent all available compute at Colossus 1, the Memphis data center xAI originally built for its own model training. Google’s contract appears to cover roughly half the capacity Anthropic has access to at that site. Two frontier-lab anchor tenants, both routing around their own internal capacity, both signing with the same vendor inside a month.
For developers, the practical shift is that the compute layer underneath the major frontier models is increasingly shared infrastructure. Gemini Enterprise customers will be running on hardware SpaceX provisioned for xAI; Claude API customers already are. The implication is that throughput ceilings during peak windows are no longer purely a Google or Anthropic operational concern: they are a SpaceX delivery concern, and the termination clauses suggest both sides know it.
SpaceX’s own AI numbers underscore the urgency on the supply side. The prospectus reports $7.7 billion of Q1 capital expenditure committed to AI out of a $10.1 billion total, an operating loss of $2.5 billion on $818 million in AI revenue. The segment is being run as a build-now, monetize-later bet, with the Google and Anthropic contracts as the first two large monetizations.
What’s worth watching
-
Whether Microsoft, Amazon, or Oracle signs a comparable deal. If the top hyperscalers are quietly renting from SpaceX while building their own next-gen data centers, the shape of the supply market in 2027 is materially different from what cloud customers have been priced into.
-
Whether the September 30 delivery deadline holds. A miss would activate Google’s reduced-payment fallback and become the first public signal that SpaceX’s AI capex is running into execution friction. A hit would lock in the September IPO narrative and shift pricing power toward SpaceX for the next round of contracts.
-
What the contracts do to xAI’s own runway. xAI is now inside SpaceX, but Grok still has to compete with Gemini and Claude on capability. If the most lucrative use of Colossus 1 capacity is renting it to competitors, Grok’s training budget effectively competes with billable Google and Anthropic load. The internal prioritization call is no longer abstract.
The compute market is the dependency every model vendor and every cloud customer is now exposed to, and the last week made the dependency public.
Stackmaven’s follow-up on the SpaceX compute contracts lands on or around September 5, 2026.