Earnings Signal: Google to pay SpaceX $920 million a month for compute capacity at xAI data centers

Google has agreed to pay SpaceX $920 million a month for AI compute capacity under a 32-month contract, according to a regulatory filing.

The agreement was also described in a separate report, but the filing provides the clearest confirmed details: a long-term capacity reservation, a lower-fee ramp before full service, and a hard delivery deadline tied to Google’s right to walk away.

The scale is striking even before the ramp period is counted. At the stated monthly rate, the core term implies about $29.4 billion in revenue over 32 months. Annualized, that is a little more than $11 billion a year, or over $30 million a day once the contract reaches its full run rate.

The filing says Google will use about 110,000 Nvidia GPUs, along with CPUs, memory and other components housed in SpaceX’s data center. The full-rate portion runs from October this year through June 2029, with capacity ramping through September at a reduced fee.

On a rough math basis, the monthly payment works out to more than $8,000 per GPU, though that figure is only a blunt yardstick because the deal includes much more than graphics chips.

That distinction matters. What is confirmed is not just a chip purchase, but a broader infrastructure package that includes the surrounding hardware needed to make the systems usable. What is still unclear is how the economics break down between GPUs and everything else, how quickly the capacity becomes productive, and whether the ramp unfolds as planned.

The contract’s most important clause may be the delivery test. SpaceX said that if it fails to provide access to the committed number of GPUs by Sept. 30, 2026, Google can immediately terminate the agreement.

After a one-month grace period, Google can also choose to take fewer GPUs at a reduced fee, which means the headline value is real but not unconditional.

That makes this both a demand signal and an execution test. The confirmed part is easy to read: a customer as large as Google is still willing to lock in multiyear AI infrastructure at an extraordinary dollar amount.

The harder part is whether all that promised capacity arrives on time, because in projects this large, delayed delivery can matter as much as weaker demand.

The base-case scenario is that the build-out largely stays on schedule and the contract becomes a durable marker of sustained AI spending through 2029. If that happens, the agreement would support the view that hyperscalers still see compute as scarce enough to justify utility-style commitments measured in the tens of billions.

It would also suggest that access to working infrastructure, not just access to chips on paper, remains one of the industry’s key bottlenecks.

The upside scenario is narrower but meaningful. If this deal turns out to be part of a broader pattern of large prepaid or reserved-capacity agreements, investors may start treating AI compute contracts more like recurring infrastructure revenue than one-off tech procurement.

That would not erase execution risk, but it would strengthen the case that a significant slice of AI spending is becoming locked in over years rather than quarters.

The downside scenario is the simplest: delivery slips. If the full 110,000-GPU commitment is not available by the deadline, Google has contractual ways to reduce the economics or leave altogether.

In that case, the lesson would be that headline demand for AI compute does not automatically translate into recognized revenue, especially when the physical build-out still has to catch up.

So far, the verified facts point in one direction without settling every broader question. Google has made an unusually large commitment, SpaceX has set a timetable it must meet, and the money only fully materializes if the hardware comes online as promised.

For markets watching earnings and guidance across the AI supply chain, that is enough to make this deal worth taking seriously — but not enough to treat the most optimistic outcome as assured.

Published at 2026-06-05T20:00:40.698973+00:00 UTC

Related Symbols

• NVDA — Nvidia
• Selection note: Alphabet/Google is the direct counterparty to the $920M-per-month compute agreement, impacting both listed share classes, and Nvidia is explicitly named as supplying the GPUs used for the capacity.

References

• CNBC
• Bloomberg
• Bloomberg