SpaceX has disclosed a binding SpaceX GPU lease commitment worth approximately 40 billion dollars, the largest committed GPU financing on record. The arrangement, structured by Apollo Global Management, will finance Nvidia’s next-generation Vera Rubin AI chips and related data center equipment for SpaceX’s growing artificial intelligence operations.
The transaction gives SpaceX access to scarce compute hardware through leases lasting about 5.00 to 5.25 years. It has not yet closed, and the fixed interest rate has not been set. SpaceX expects to close before December 31, 2026, with equipment deliveries and the first lease schedules beginning in the first half of 2027.
How the SpaceX GPU lease works
An entity managed by Apollo will purchase the Vera Rubin chips and accompanying server hardware outright, then lease the assets back to SpaceX. The sale-and-leaseback structure moves the associated debt away from SpaceX’s own balance sheet while allowing the company to deploy expensive equipment without buying every asset directly.
Goldman Sachs is the sole structuring agent. Affiliates of BlackRock, Goldman Sachs Asset Management and KKR also provided financing commitments. Equipment will be acquired in tranches, and SpaceX expects aggregate principal payments of roughly 40 billion dollars.
According to the October 26 filing as described by Morningstar, rent for each tranche will equal 5 percent of its financed cost, plus a fixed annual rate on the outstanding balance. SpaceX has not disclosed the final rate and has not formally closed the deal, so the full cost of financing remains unconfirmed.
The arrangement resembles the GPU-backed financing model used by CoreWeave. Its central advantage is collateral value. A lender taking possession of high-end Nvidia accelerators would have access to equipment with strong demand in a secondary market. That differs materially from repossessing ordinary servers, which usually depreciate quickly and have narrower resale prospects.
A new private credit record
The 40 billion dollars commitment surpasses the roughly 36 billion dollars Anthropic structured financing arranged by Apollo and Blackstone for Google TPU hardware. The Anthropic transaction required a Broadcom backstop for senior debt, whereas SpaceX’s financing is supported primarily by SpaceX’s own credit.
That distinction matters because SpaceX now has an investment-grade balance sheet following its June 2026 Nasdaq listing, which raised approximately 86 billion dollars. Stable revenue from Starlink, launch operations and its artificial intelligence segment gives lenders a borrower with a broader business base than a standalone AI laboratory.
Private credit managers are moving into infrastructure finance because AI data centers require more capital than many bank syndicates can efficiently provide. Equipment financing can also be adapted to a deployment schedule, with lenders funding delivery tranches as capacity comes online. The result is a financing model that could become standard for companies building large compute clusters.
Why Vera Rubin hardware matters
Nvidia designed the Vera Rubin platform for agentic artificial intelligence workloads, the class of systems intended to perform multi-step tasks rather than answer individual prompts. Its flagship NVL72 configuration combines 72 Rubin GPUs with 36 Vera CPUs in one chassis and offers 3.6 exaFLOPS of FP4 inference performance.
Nvidia says that performance comes with five times higher inference capability and ten times lower token cost than Blackwell, its preceding architecture. The Vera CPU includes 88 Olympus cores and 1.2 terabytes per second of LPDDR5X memory bandwidth, while the system uses HBM4 memory and NVLink 6 rack-level connections.
The platform entered full production earlier in 2026, and major cloud operators have begun deploying racks. Scarce supply is one reason SpaceX wanted committed financing before delivery. The agreement should give SpaceX priority access to equipment at a time when AI infrastructure operators are competing for Nvidia’s latest production.
SpaceX’s AI spending is already substantial
The new commitment follows months of unusually heavy investment. SpaceX reported 28 billion dollars in AI-related non-cancelable commitments as of June 2026. It spent 15.8 billion dollars on AI compute in the second quarter, accounting for roughly 86 percent of capital expenditure for that period.
Third-quarter AI capital expenditure reached approximately 16.4 billion dollars, bringing year-to-date spending to roughly 40 billion dollars before the lease was funded. Total quarterly capital expenditure rose to nearly 20 billion dollars from 6.9 billion dollars a year earlier. CFO Bret Johnsen said SpaceX expected fourth-quarter spending of 35 billion to 40 billion dollars.
Revenue growth provides some support for those outlays. SpaceX’s renamed Super Intelligence segment, which includes Grok, Cursor and its cloud compute operation, generated 7.56 billion dollars in third-quarter revenue. That was about 52 percent of SpaceX’s total quarterly sales of 14.58 billion dollars and represented 161 percent year-over-year growth.
Two major contracts underpin the segment. Anthropic pays approximately 1.25 billion dollars per month for the Colossus cluster in Memphis, while Google pays approximately 920 million dollars per month for access to roughly 110,000 GPUs through mid-2029. SpaceX also reported quarterly net income of approximately 4.4 billion dollars, although it remained down roughly 413 million dollars for the first nine months of 2026.
Compute ambitions remain ahead of the financing
SpaceX operates 1.7 gigawatts of nameplate compute capacity and has guided to more than 2 gigawatts by year-end. Musk has said capacity could approach 10 gigawatts rather than 5 gigawatts by the end of 2027. Filings list roughly 100,000 H100 chips, 110,000 GB200 chips and at least 330,000 GB300 chips across current and planned Colossus deployments.
A complete conversion to Vera Rubin would push the fleet well beyond one million accelerators. SpaceX has not confirmed how much of the lease will be used for a full Rubin conversion, so any estimate of the final chip count remains a projection rather than a company figure. SpaceX has also confirmed a fourth data center in the Memphis area.
The financing gives SpaceX time, but not unlimited room, to expand. Delivery schedules, grid access, cooling capacity and the pace of customer contracts will determine how quickly leased hardware produces returns. Management’s claim of less than one-year payback periods on AI infrastructure investment provides a useful measure, though the amount of new capacity and the precise methodology behind that figure have not been fully disclosed.
Implications for AI infrastructure lenders
For Apollo, the SpaceX commitment reinforces a central strategic position in private credit. The firm previously arranged the record Anthropic TPU lease and is part of an Nvidia-backed consortium that intends to mobilize more than 500 billion dollars for AI infrastructure over time. BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR are also involved in that broader capital effort.
The SpaceX transaction may offer other borrowers a model, but it is not automatically transferable. Smaller neocloud operators often lack investment-grade credit, diversified revenue or the resale protections available from cutting-edge Nvidia equipment. A lender may therefore demand corporate guarantees, higher pricing or tighter collateral coverage from those companies.
Musk used the announcement to argue that established cash-generating businesses have an advantage over smaller firms dependent on AI data center revenue alone. His criticism of more than 200 GPU-rental companies may be rhetorical, but the financing does show that balance-sheet quality and access to multiple revenue streams can shape how private lenders price infrastructure risk. Whether the SpaceX GPU lease closes on schedule and at what interest rate, and whether the neocloud market consolidates as Musk expects, will decide how much the SpaceX GPU lease actually reprices AI infrastructure finance.
With the SpaceX GPU lease still awaiting closing and final pricing, the next milestones are confirmation of the interest rate, satisfaction of closing conditions and the first equipment deliveries. If that process proceeds as planned, the arrangement will provide a template for turning scarce AI chips into financeable assets while giving SpaceX one of its largest tools for expanding artificial intelligence capacity.