SpaceX has turned its GPU infrastructure into a $26 billion-a-year leasing business, signing two of the largest compute capacity agreements in history just days before its Nasdaq debut.
SpaceX has turned its GPU infrastructure into a $26 billion-a-year leasing business, signing two of the largest compute capacity agreements in history just days before its Nasdaq debut.

SpaceX has turned its GPU infrastructure into a $26 billion-a-year leasing business, signing two of the largest compute capacity agreements in history just days before its Nasdaq debut.
SpaceX on Friday disclosed a multiyear cloud services agreement with Alphabet's Google, under which Google will pay $920 million a month from October 2026 through June 2029 for access to roughly 110,000 Nvidia GPUs and related hardware, according to a regulatory filing. The deal follows a separate agreement with AI startup Anthropic, announced in May, valued at $1.25 billion a month for compute capacity at SpaceX's Colossus 1 facility in Memphis, Tennessee, which houses more than 220,000 Nvidia processors and delivers 300 megawatts of capacity.
"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," a Google Cloud spokesperson said.
Combined, the two contracts generate roughly $21.7 billion in monthly revenue for SpaceX, or about $260 billion on an annualized basis. If both agreements run to their scheduled end dates, the aggregate contract value exceeds $70 billion. The Google deal includes termination provisions: if SpaceX fails to deliver the agreed number of GPUs by Sept. 30, Google may terminate after a one-month grace period or accept reduced capacity with a proportional fee reduction. After Dec. 31, either party can exit with 90 days' notice.
The compute leasing business provides a highly visible revenue stream as SpaceX prepares for what is expected to be the largest U.S. initial public offering in history. The company aims to raise roughly $75 billion at a valuation approaching $1.8 trillion when it begins trading on Nasdaq next week. The deals strengthen SpaceX's AI narrative for investors, though the company's own AI subsidiary, xAI, faces significant headwinds.
The leasing business is booming, but xAI's internal struggles raise questions about SpaceX's long-term AI strategy.
According to a regulatory filing, xAI posted a $6.4 billion operating loss last year on revenue of just $3.2 billion. The Information reported that xAI's pre-training team has shrunk to fewer than five people, down from more than 20 a year ago, and that eight co-founders have departed. In one incident, an employee accidentally deleted core training data for an AI coding product, losing the equivalent of two to three weeks of work.
The company has also faced scrutiny over its use of competitors' models. The Information reported that xAI conducted a months-long distillation project using Anthropic's Claude outputs to train its own coding models, and that some engineers continued accessing Claude through personal accounts after Anthropic cut off direct access. Elon Musk acknowledged in a May court deposition that xAI had "partially" used OpenAI models to train its Grok chatbot, calling the practice industry-standard.
SpaceX's S-1 filing describes the compute leasing structure as a way to "monetize unused compute capacity" while retaining the ability to reallocate capacity for internal initiatives. Musk has characterized the Anthropic deal as a "short-term arrangement" on X, adding that SpaceX "may need to take the compute back at some point." The company said it continues training its next major Grok model but has not disclosed a release timeline.
The deals reshape competitive dynamics in the AI infrastructure market. Nvidia, whose GPUs power both agreements, benefits from surging demand that keeps its data center revenue growing — the company reported $39.3 billion in data center revenue in its most recent quarter. But the concentration of compute capacity in SpaceX's hands also creates a potential competitor to traditional cloud providers like Amazon Web Services and Microsoft Azure, which have been the primary lessors of Nvidia's H100 and B200 chips. Google's willingness to pay a premium for SpaceX capacity rather than building its own suggests the hyperscaler market is supply-constrained even for the largest players.
For investors, the compute deals provide a quantifiable anchor for SpaceX's IPO valuation. At $26 billion in annualized leasing revenue, the business alone would justify a significant portion of the $1.8 trillion valuation if valued at 30 to 40 times run-rate revenue — a multiple in line with high-growth infrastructure peers. However, the termination provisions in both contracts mean the $70 billion aggregate figure depends on both parties choosing to honor the agreements through maturity. Google retains full ownership of its AI models and data under the deal terms, limiting SpaceX's ability to use that compute activity for its own AI development.
This article is for informational purposes only and does not constitute investment advice.