Google Pays SpaceX $920M a Month for AI Compute in $30B Deal

Google has agreed to pay SpaceX $920 million per month for access to approximately 110,000 NVIDIA GPUs, CPUs, memory, and related components, according to an SEC filing disclosed on June 5, 2026. The deal runs from October 2026 through June 2029 and is worth approximately $30 billion in total, according to TechCrunch.

The agreement includes a ramp‑up period through September 2026 at a reduced fee. If SpaceX fails to deliver the committed GPU count by September 30, Google can terminate immediately or accept fewer GPUs at a reduced monthly rate. After December 31, 2026, either party can cancel with 90 days' notice, as detailed by.²

Google, by some estimates, owns more AI compute than anyone, including custom TPUs and global data centers. But demand for Gemini Enterprise, its AI agent platform launched in October 2025, has surged past expectations.

A Google Cloud spokesperson told ¹ the deal provides "bridge capacity to meet surging customer demand for our agent platform, Gemini Enterprise, which has been even higher than we expected." Alphabet has already committed over 80 billion in capital expenditures for 2026, alongside an equity sale of approximately 5 billion, per.²

As Startup Fortune put it: "Google agreeing to rent compute from SpaceX is not just another cloud contract. It is a signal that the old playbook, where hyperscalers relied mainly on their own data centers and procurement muscle, is no longer enough for the AI market they helped create."

This is SpaceX's second major compute deal in two weeks. In late May, TechCrunch reported that Anthropic signed a $1.25 billion per month deal for the entire output of the Colossus 1 data center in Memphis, Tennessee. Together, the two deals position SpaceX as a major AI infrastructure provider.

SpaceX's AI segment posted an operating loss of $2.5 billion on just $818 million in revenue in Q1 2026, per.² The Google and Anthropic contracts transform that narrative: SpaceX can now argue its data center buildout is a commercial infrastructure business.

Five years ago, Google supplied cloud resources to SpaceX for Starlink. Thomas Kurian, CEO of Google Cloud, said at the time: "They chose us because of the quality of our network and the distribution and reach of our network," according to CNBC. Now the roles are completely reversed.

SpaceX's AI arm, xAI, competes directly with Google through its Grok chatbot. Elon Musk has acknowledged Grok needs a rebuild after a talent exodus, per.² Yet Google is paying billions to the same corporate entity. Alphabet's 6.11% stake in SpaceX, dating back to a 2015 investment at a $12 billion valuation, could be worth over $100 billion post‑IPO, as noted by TechCrunch.

If even Google needs to go outside for GPU capacity, smaller companies and individual builders face an even steeper challenge. The AI infrastructure market is behaving less like a commodity cloud service and more like a strategic resource market.

As Startup Fortune observed: "AI compute contracts now resemble power deals, aircraft leases and strategic supply agreements more than ordinary cloud purchasing."

For builders, the practical implications are clear: model inference costs are unlikely to drop significantly in the near term, API availability may become constrained during demand spikes, and the companies that secure long‑term compute contracts will have a structural advantage.

The deal caps a remarkable transformation for SpaceX. In February 2026, it merged with xAI at a $1.25 trillion valuation. Now, days before what is expected to be the largest IPO in history, SpaceX has demonstrated its data centers are a commercial infrastructure business with blue‑chip customers.

Neocloud providers like CoreWeave and Nebius now face competition from SpaceX itself. Both stocks fell Friday amid a broader tech selloff before rebounding on the SpaceX-Google announcement, per.²

For Google, the deal is a pragmatic admission: in the AI race, availability beats purity. As Startup Fortune put it, "the scarce resource is not only the chip. It is the complete site" — the land, power, cooling, and networking that turns a pile of GPUs into a working AI supercomputer.