Summary: An SEC filing disclosed on June 5, 2026 that Google will pay SpaceX $920 million monthly from October 2026 through June 2029 for roughly 110,000 NVIDIA GPUs and related components. The deal is described as bridge capacity for Gemini Enterprise after demand outran expectations. Combined with Anthropic’s contract, SpaceX’s compute rental revenue reaches about $2.17 billion per month.
On June 5, 2026, a compute contract came to light through an SEC filing from SpaceX: Google will pay $920 million each month, from October 2026 through June 2029, in exchange for the use of roughly 110,000 NVIDIA GPUs along with CPUs, memory, and related components. Reuters has confirmed the agreement is a multi-year cloud services deal, and the filing became public exactly one week before SpaceX stock is expected to begin trading on Nasdaq.
Three angles make it worth attention. One: the tenant in this arrangement is Google, which Epoch AI pegs as the largest single owner of AI compute on the planet — and it is borrowing GPU capacity from a rocket maker. Two: stack the $1.25 billion per month Anthropic agreed to in May on top, and SpaceX’s compute-leasing operation is collecting on the order of $2.17 billion every month. Three: the deal confirms what large-scale compute turns into when power and delivery timelines bind — a strategic commodity available for fleet-scale rental.
Contract Terms: $920M Monthly, 110,000 GPUs, a 90-Day Off-Ramp
TechCrunch, which reviewed the filing, highlights these key clauses:
- Price and term: $920 million per month, starting October 2026 and running through June 2029
- Fleet size: approximately 110,000 NVIDIA GPUs, together with CPUs, memory, and related components
- Delivery: SpaceX must complete GPU delivery by September 30, 2026, with a one-month grace period; past that point, Google can terminate right away or take fewer GPUs in exchange for lower monthly fees
- Walking away: once December 31, 2026 is behind them, either side can end the deal with 90 days’ notice
- Ramp: usage builds up through September at a reduced fee
Where this capacity will physically sit, SpaceX has not said. Musk has stated before that Colossus 2 is held back for xAI’s own use, and the May agreement with Anthropic already claimed every available cycle of Colossus 1 near Memphis, Tennessee. By TechCrunch’s estimate, Google’s rental comes to about half of Anthropic’s holdings at Colossus 1, so the balance presumably comes from newly added capacity or from sites that have not been disclosed.
Why the Biggest Compute Owner on Earth Went Looking for Rented Capacity
Google characterizes the arrangement as a “short-term, timely agreement” supplying “bridge capacity” for Gemini Enterprise, its agent platform, after demand outran expectations. The surrounding numbers show why urgency won out: Alphabet has already pledged more than $180 billion in capital expenditures this year, expects the figure to “significantly increase” in 2027, and just lined up an $80 billion equity sale to help cover the bill. In-house capacity, however fast it expands, still trails the inference demand that agentic products are generating.
The Anthropic comparison is instructive. Anthropic leased Colossus 1 out of genuine scarcity — by April, it was already conceding infrastructure strain. Google has no long-term shortage; what it lacks is capacity in the here and now, and the entire point of the contract is paying cash to a quicker-moving third party to buy time. Musk’s version is blunter: companies head to SpaceX because power supply is stretched thin, and SpaceX builds its own generation plants and moves faster.
SpaceX’s Compute Business: the Second Leg of the IPO Story
Combine the two agreements — Anthropic at $1.25 billion a month, Google at $920 million — and SpaceX has locked in roughly $26 billion a year of compute rental revenue, with both contracts running through 2029. For a company about to raise roughly $75 billion at a valuation near $1.75 trillion — on course to stage the largest IPO in history — these multi-year deals are tangible proof of the AI second growth curve its prospectus describes.
Two footnotes round out the picture. Google has been a SpaceX investor for years, and Bloomberg estimates its stake will top $100 billion in value once the IPO is done — the company will be tenant and shareholder at once. The Wall Street Journal, for its part, reports the two sides are negotiating over orbital data centers, seen as a marquee project for after the listing.
What Developers and Product Teams Should Take Away
Three lessons stand out. First, “bridge capacity” has graduated into routine procurement: if Google will spend $920 million a month to cover a roughly 33-month gap, inference demand has grown too volatile for any capacity plan to react in time — which makes rented transitional capacity a legitimate line in your architecture assumptions. Second, the supplier roster is being rewritten: SpaceX, a fast-delivering newcomer that builds its own power plants, can now serve customers at the Anthropic and Google tier side by side. Third, building your own chips and renting capacity are not mutually exclusive: OpenAI is advancing a Broadcom-designed chip of its own while still needing external capacity, and Google runs its own TPUs next to these rented GPUs. In 2026, the compute strategy question has more than one defensible answer.