Elon Musk's SpaceX turns Google into a $920 million-a-month compute customer
The deal, disclosed a week before SpaceX's expected Nasdaq debut, would run through June 2029 unless either side exits early.
By Staff Writer ยท Published
Why it matters
SpaceX is using AI compute contracts to broaden its IPO story beyond rockets and satellites, while Google is signaling that even hyperscalers still need outside GPU capacity.
Elon Musk's SpaceX has signed Google to a compute rental agreement worth $920 million a month, according to a SpaceX regulatory filing on Friday that was first reported by TechCrunch.
The deal gives Google's AI organization access to "approximately 110,000 NVIDIA GPUs, CPUs, memory, and other related components" from October 2026 through June 2029, according to the filing. On a simple monthly basis, that implies $30.36 billion of gross contracted payments if the agreement runs for the full 33 months.
That number comes with an important caveat. TechCrunch reported that both SpaceX and Google can terminate the agreement with 90 days' notice after December 31, 2026. In other words, the headline total is not the same thing as a locked, noncancelable backlog. But the timing is still the point: SpaceX disclosed the agreement one week before its stock is expected to begin trading on Nasdaq, with SEC paperwork showing SpaceX aiming to raise about $75 billion at a roughly $1.75 trillion valuation.
For Musk, who founded SpaceX as a rocket company and has since pulled xAI into the same corporate orbit, the filing recasts SpaceX as something public-market investors can more easily model: not only a launch, Starlink, and satellite manufacturer, but a landlord for scarce AI infrastructure.
Google is not supposed to need someone else's cloud
The counterintuitive part of the deal is the buyer. Google is one of the few companies on Earth with its own global data-center footprint, custom AI chips, and cloud business. If Google is willing to rent a large block of NVIDIA-based capacity from SpaceX, it says something about the continuing imbalance in AI compute: even hyperscalers are still chasing usable, near-term capacity wherever they can get it.
The filing's wording also matters. It does not say Google is buying 110,000 GPUs. It says Google is paying for access to a bundle that includes NVIDIA GPUs, CPUs, memory, and related components. That distinction makes the agreement harder to compare directly with a standard cloud instance contract or a chip purchase order. The filing does not identify the NVIDIA GPU generation, the cluster topology, networking specifications, power profile, or utilization commitments.
Still, the broad competitive implication is clear. The AI infrastructure market is no longer just Amazon, Microsoft, Google, Oracle, and a handful of specialist GPU clouds selling capacity to startups. SpaceX is positioning itself as a compute supplier to the largest AI labs and hyperscalers, using physical infrastructure originally tied to Musk's own AI ambitions.
TechCrunch reported that the Google agreement resembles a late-May deal in which Anthropic agreed to pay SpaceX $1.25 billion per month through 2029 to rent compute from a Colossus data center near Memphis, Tennessee. That facility was originally built by xAI for its own AI work, according to TechCrunch, and xAI is now part of SpaceX.
Taken together, the two disclosed monthly figures amount to $2.17 billion a month, or $26.04 billion on an annualized basis, while both agreements are in force. That is arithmetic based on the disclosed monthly payments, not a statement of recognized revenue. The termination clauses mean public investors will need to distinguish between contracted demand, durable backlog, and capacity that can be reallocated if a customer walks.
The IPO incentive is visible
A week before an expected listing is not a neutral moment to disclose a contract of this size. SpaceX benefits from showing investors that its AI compute business can attract blue-chip counterparties. Google benefits from locking in capacity while retaining a cancellation option after the end of 2026. NVIDIA benefits indirectly because the scarcity around its GPUs continues to support strategic buying and long-duration capacity deals.
Google also has a second exposure. It is a longtime SpaceX investor, and Bloomberg previously reported that Alphabet's stake could be worth more than $100 billion after the IPO. That makes Google both a customer and an investor in the company selling it compute. The arrangement may be economically rational for both sides, but it also complicates the clean story investors usually prefer: a simple vendor-customer transaction priced at arm's length between fully independent parties.
For SpaceX, the contracts may help bridge two investor narratives. The first is the traditional SpaceX story: launch dominance, Starlink connectivity, reusable rockets, and a founder whose companies routinely internalize supply chains. The second is the AI-infrastructure story, where the market has rewarded companies that can turn power, land, chips, and networking into contracted capacity.
The Google deal points toward the second narrative. A rocket company with massive capital requirements can be valued partly as an infrastructure platform if it can show recurring payments from companies that themselves operate at hyperscale.
The competitive pressure moves upstream
The agreement also shows where competition in AI is shifting. Model labs and cloud providers are still competing on software, data, distribution, and developer ecosystems. But the scarcest input remains large-scale, ready-to-use compute. That scarcity gives unusual leverage to anyone who can assemble chips, power, networking, and operations faster than the incumbent clouds can bring equivalent capacity online.
That is why SpaceX's move matters beyond SpaceX. If Google is renting external NVIDIA capacity at this scale, smaller AI companies face an even tighter market. The best clusters are increasingly tied up in multi-year arrangements with companies that can write billion-dollar monthly checks or offer strategic value to infrastructure owners.
Anthropic's SpaceX agreement already suggested that leading AI labs are willing to rely on Musk-controlled infrastructure even as they compete in a market where Musk's xAI is also active. Google's agreement widens that pattern. SpaceX is not merely serving xAI's internal needs; it is commercializing compute for external counterparties that have their own strategic reasons to avoid falling behind.
The competitive question for Amazon, Microsoft, Oracle, and other cloud providers is not whether SpaceX becomes a general-purpose cloud overnight. The question is whether large AI customers increasingly treat specialized compute campuses as strategic supply, separate from their primary cloud relationships. If that happens, the cloud market becomes less about where software runs and more about who can guarantee the next tranche of accelerated capacity.
The missing details are the financial ones that matter most
The filing answers the headline question - how much Google will pay each month and roughly what hardware pool it gets access to - but leaves open the details investors would need to value the contract precisely.
The filing excerpt cited by TechCrunch does not specify SpaceX's gross margin on the agreement, power costs, depreciation schedule, financing structure, customer prepayment terms, or utilization commitments. It also does not specify whether Google receives dedicated capacity or priority access within a broader pool. Those details determine whether $920 million a month is a high-margin infrastructure lease, a lower-margin way to monetize a capital-heavy buildout, or something in between.
The cancellation option is also central. If both sides can exit with 90 days' notice after December 31, 2026, SpaceX can present major customer demand while preserving flexibility for both counterparties. That is useful before an IPO. It is less definitive than a long-term take-or-pay commitment with no early termination right.
Even with those gaps, the filing gives SpaceX a powerful pre-IPO message: two of the most important AI buyers, Google and Anthropic, are willing to commit enormous monthly sums for access to its compute. For a company about to ask public investors to underwrite a historic valuation, that is the kind of customer list that can make a new business line look less theoretical.