Google signs $920 million-a-month compute deal with SpaceX to bridge AI capacity

Alphabet’s Google has agreed to pay SpaceX $920 million per month for AI compute capacity, in a deal that is intended to provide “bridge” infrastructure as demand for Google’s AI offerings accelerates, according to multiple reports citing SpaceX regulatory disclosures.

The arrangement is described as running for 32 months, starting in October 2026 and extending through June 2029, with capacity ramping up through September at a lower fee. Reports say the contract is tied to compute equipment housed at data centers associated with xAI, Elon Musk’s artificial intelligence company, following SpaceX’s integration with xAI.

A key detail reported in connection with the filing is the use of roughly 110,000 Nvidia graphics processing units (GPUs), along with additional components such as central processing units (CPUs), memory, and related infrastructure. GPUs are the specialized chips that train and run many modern AI models, and hyperscalers such as Google compete intensely to secure enough of them alongside power, cooling, and data center space.

SpaceX also indicated in the filing that Google can terminate the agreement immediately if SpaceX does not deliver the committed amount of GPUs by September 30, 2026. Alternatively, the contract would continue with a reduced amount of GPUs for a reduced fee after a one-month grace period, according to the reports.

After the initial term, the agreement can be terminated by either party with 90 days’ notice, the reports add. Separately, a Google Cloud spokesperson told CNBC that the deal was structured to ensure it had bridge capacity to meet surging customer demand for Google’s “Gemini Enterprise,” which is Google’s subscription service aimed at large organizations and rolled out in October.

For SpaceX, the reported compute contract arrives as the company positions itself for a major liquidity event. CNBC and other outlets characterized the timing as notable because SpaceX is preparing for an initial public offering. A compute deal at this scale would represent a significant non-launch-related revenue stream, assuming it holds under the delivery and ramp terms described.

The broader context is that AI infrastructure capacity has become a bottleneck beyond chip supply. Reports that emphasize Google’s motivation point to demand for ready-to-run capacity while Google builds out its own systems, but they also underscore the practical constraints of power and data center availability, even for companies with substantial capital resources.

What remains unclear from the public reporting is the exact contractual structure beyond the reported monthly pricing and compute commitments, including whether the GPUs are dedicated, how pricing changes during ramp periods, and what service levels are guaranteed. The reports reference regulatory language, but the full filing text and any side letters were not included in the articles reviewed here, so the precise commercial remedies and operating details may be more granular than has been described. Investors and customers will likely watch for further disclosures if additional terms surface in filings or official statements.