- Deal: Google will pay SpaceX $920 million a month to rent AI compute capacity for Gemini Enterprise demand.
- Scale: The agreement covers about 110,000 NVIDIA GPUs plus related CPUs, memory, and systems from October 2026 through June 2029.
- Safeguards: SpaceX must deliver the committed GPU access by September 30, 2026; if it misses, Google can terminate after a grace period or accept less capacity for a lower fee.
- Market Impact: The contract supports SpaceX’s compute- services story before a possible high-valuation IPO, but only if SpaceX delivers the hardware behind the deal.</ span>
Google will pay SpaceX $920 million a month under a cloud service agreement for access to AI compute capacity. SpaceX signed the agreement with Google on June 5. Agreement terms give Google rented processing power, memory, and related systems for AI workloads rather than a new public cloud product or a Starlink broadband service.
From October 2026 through June 2029, Google gets capacity tied to roughly 110,000 NVIDIA GPUs, CPUs, memory, and related components. Graphics processing units are chips used for parallel AI calculations, so a block of that size would be meaningful even for a company with Google’s infrastructure base.
Google has framed the purchase as short-term bridge capacity for stronger-than-expected demand for Gemini Enterprise, its agent platform for businesses. For Google, the purchase buys time as much as hardware while its own AI infrastructure program continues to expand.
The Terms That Make the Deal Conditional
SpaceX must deliver the GPU access before the full monthly fee can become routine revenue. Capacity ramps before the main payment period begins, and the contract gives Google remedies if the hardware commitment falls short. Until October 2026, the agreement covers ramp-up capacity at reduced fees, so the early phase tests supply before the full monthly commitment starts.
By the September 30, 2026 deadline, SpaceX must deliver access to the committed GPU amount. If it misses that threshold, Google can terminate after a one-month grace period.
Google can also accept reduced capacity with a pro rata fee reduction. Contract remedies keep the monthly fee tied to delivered hardware rather than treating the full payment as guaranteed regardless of supply.
After December 31, 2026, either side may end the arrangement on 90 days’ notice. A long main term still remains in place if both sides accept the early delivery ramp and continue beyond the first opt-out window.
Google also retains ownership of its AI models, content, and related data. Contract protections give enterprise customers a clearer boundary: SpaceX supplies access to infrastructure, while Google keeps control over the models and information that run on it.
Why Google and SpaceX Need Each Other Here
Alphabet’s spending plan helps explain why Google would still buy bridge capacity from another infrastructure owner. Alphabet has proposed an $80 billion equity raise to expand AI infrastructure and compute, and Google expects 2026 capital expenditures of $180 billion to $190 billion, mostly for technical infrastructure. Those figures put near-term bridge capacity alongside Google’s own buildout, not in place of it.
Google is delivering TPUs directly to select enterprise customers in their own data centers. Tensor processing units are Google’s custom AI chips, and Google Cloud backlog has grown to more than $460 billion, with just over half expected to be recognized over the next 24 months. Google says more than 75% of Google Cloud customers are using its AI products, giving the SpaceX capacity a near-term role while longer-term infrastructure programs catch up.
For SpaceX, the Google agreement puts another outside customer on AI infrastructure it gained through xAI. Anthropic agreed in May to rent Colossus 1 compute from a data center xAI built before SpaceX acquired the AI company in February. That added demand gives SpaceX another customer contract before a possible listing, as it seeks a reported $1.75 trillion-plus valuation despite much lower outside estimates.
Google’s contract strengthens the recurring-revenue argument SpaceX can put in front of investors, but the delivery clauses keep that argument tied to actual GPU access. In the increasingly tight AI compute market, companies are now locking down guaranteed capacity through large commitments rather than waiting for generic cloud supply. If SpaceX slips beyond the grace period, Google’s choice between termination and reduced capacity will decide whether the monthly commitment becomes durable revenue or a conditional bridge that narrows before it reaches full scale.
