Google signed a deal to pay SpaceX $920 million per month for compute capacity at xAI data centers, gaining access to roughly 110,000 Nvidia Graphics Processing Units (GPUs) from October 2026 through June 2029. The agreement, disclosed in a Securities and Exchange Commission (SEC) filing, includes a cancellation clause allowing either party to terminate with 90 days’ notice after December 31, 2026.
A deal before a historic IPO
SpaceX just keeps printing money. One week before its stock is expected to trade on Nasdaq (aiming to raise $75 billion at a $1.75 trillion valuation), the company announced its second massive compute deal in two weeks.
First, Anthropic agreed to pay $1.25 billion monthly for all compute at SpaceX’s Colossus 1 data center. Now, Google is paying $920 million monthly for roughly half that capacity (110,000 Nvidia GPUs plus CPUs and memory).
Google described it as a “short-term, timely agreement” to meet “unexpectedly high” demand for its Gemini Enterprise agent platform. The irony? Google is already the world’s largest owner of AI compute, thanks to its custom Tensor Processing Units (TPUs). But even the king needs backup.
Most important aspects of this deal
- It’s Huge: We’re looking at over $30 billion total over 33 months. The power levels will ramp up through 2026 before hitting the full rate for the rest of the term.
- The Hardware: About 110,000 Nvidia GPUs, which is roughly half of what Anthropic secured for Colossus 1.
- The Safety Net: If SpaceX can’t deliver the GPUs by late 2026, Google can walk away or take a discount. Plus, after December 2026, either side can cancel if they give 90 days’ notice.
- Privacy Matters: This is strictly a rental. Google keeps total control over all its models and data; no sharing allowed.
- The Big Picture: SpaceX spent billions building this infrastructure, and these deals are the perfect way to show investors they can turn that tech into serious revenue right before they go public.
How Google controls the most AI computing power
Here’s the twist that makes this deal surprising: Google already owns more AI compute than anyone on the planet. According to Epoch AI data, as of Q4 2025, Google held the largest cumulative AI compute among the four tech giants (Microsoft, Meta, Amazon), driven entirely by its custom-designed TPUs.
In contrast to Nvidia GPUs (which other companies fight over), Google’s TPUs are proprietary, optimized for its own workloads, and produced in-house. So why rent from SpaceX? Two reasons:
- First, demand for Gemini Enterprise has been “even higher than we expected,” per Google’s statement.
- Second, TPUs and GPUs are good at different things.
Google likely needs the raw, general-purpose parallel processing power of Nvidia H100s for specific workloads that TPUs handle less efficiently. Plus, Alphabet is already on a spending spree ($180-$190 billion in capex this year, with plans to “significantly increase” in 2027), and they just announced an $85 billion equity sale to fund it.
What’s next for SpaceX, Google, and AI Compute
SpaceX has transformed into a major AI infrastructure provider, not just a rocket company. The xAI merger gave it Colossus 1 and Colossus 2 data centers near Memphis, and CEO Elon Musk has suggested Colossus 2 will be reserved for xAI’s own Grok model.
To this point, the Google deal provides anchor tenancy and predictable revenue, making SpaceX’s AI segment look more credible to IPO investors. For Google, this is bridge capacity. The company is reportedly in talks with SpaceX about building orbital data centers, a long-term, sci-fi-level infrastructure play. But for now, this 33-month deal buys Google time to expand its own TPU production while keeping up with surging demand.
For the broader market, neocloud providers like CoreWeave and Nebius now face a formidable competitor: SpaceX, backed by Google’s checkbook.
