Google to pay SpaceX $920M/month for AI compute

SpaceX has signed its second massive compute deal in a month, this time with Google. Under the terms disclosed in a regulatory filing on Friday, Google will pay SpaceX $920 million per month from October 2026 through June 2029 for access to approximately 110,000 Nvidia GPUs, CPUs, memory, and related components. At that rate, the contract is worth roughly $30 billion over its lifetime.

The deal lands one week before SpaceX’s stock is expected to begin trading on Nasdaq. The company is aiming to raise around $75 billion at a valuation of approximately $1.75 trillion, which would make it the largest IPO in history.

The second Colossus deal

The agreement is similar in structure to the one SpaceX announced with Anthropic in late May. That deal saw Anthropic agree to pay $1.25 billion per month through 2029 for the entire output of Colossus 1, the Memphis data centre that xAI originally built for its own chatbot, Grok, before the AI company was merged into SpaceX in February.

Google’s deal appears to cover roughly half the compute that Anthropic has access to at Colossus 1. SpaceX did not specify which data centre Google would use. CEO Elon Musk has previously suggested the Colossus 2 facility would be reserved for xAI’s own work.

Both deals include 90-day cancellation clauses after 31 December 2026. Google’s access will ramp up through September at a reduced fee. If SpaceX fails to deliver the committed GPU capacity by 30 September 2026, Google can terminate immediately after a one-month grace period, or accept whatever hardware is available at a proportionally reduced rate.

Why Google needs this

The deal is notable because Google is not compute-poor. Some estimates name it as the world’s largest single owner of AI compute, thanks in large part to its custom TPU chips. It has already committed to more than $180 billion in capital expenditure this year and expects that figure to “significantly increase” in 2027. Alphabet recently announced an $80 billion equity sale to help fund that spending.

A Google representative described the SpaceX deal as a short-term bridge. “This is a short-term, timely agreement to ensure we have bridge capacity to meet surging customer demand for our agent platform, Gemini Enterprise, which has been even higher than we expected,” the company said in a statement.

That framing suggests Gemini Enterprise is consuming GPU capacity faster than Google’s own data centre buildout can supply it. For a company spending $180 billion a year on infrastructure, needing to rent Nvidia GPUs from a rocket company is an unusually public admission of demand outstripping supply.

From rockets to revenue

For SpaceX, the compute deals are transforming a cost centre into a revenue engine. xAI built Colossus to train Grok, but usage of the chatbot has reportedly dropped. Rather than let the GPUs sit idle, SpaceX is renting them to the two companies that need compute most urgently: Anthropic, which was severely capacity-constrained before the deal, and Google, which is building out its own infrastructure but cannot build fast enough.

Between the Anthropic and Google contracts, SpaceX is now collecting approximately $2.17 billion per month in compute revenue, or roughly $26 billion annualised. That figure will feature prominently in the IPO prospectus. Google is also a longtime SpaceX investor, with a stake expected to be worth more than $100 billion after the listing. The companies are also reportedly in talks to build orbital data centres, a major component of SpaceX’s post-IPO plans.

The broader picture is striking. The world’s largest search company is renting Nvidia GPUs from the world’s largest rocket company because demand for AI compute has outpaced even the most aggressive infrastructure buildouts. Whether that demand is sustainable, or whether it represents the peak of a cycle that will correct, is the question hanging over SpaceX’s $1.75 trillion valuation.