SpaceX secures $40B GPU lease with Apollo partnership

SpaceX announced on Monday, October 26, that it has entered into a binding commitment for a financial leasing agreement of approximately $40 billion in Nvidia Vera Rubin GPUs and associated equipment with an entity managed by Apollo Global Management. According to the company’s announcement and the Form 8-K submitted to the SEC, the arrangement will provide the next generation of Nvidia's Vera Rubin chips for the company's terrestrial computing expansion. SpaceX describes this operation as the largest GPU financing ever conducted.

The structure is a lease rather than a direct loan. The Apollo entity purchases the GPU pods and leases them to SpaceX for about five years, for use in datacenters chosen by the company, with deliveries occurring in batches. The financing group was led by Apollo and Goldman Sachs, according to the announcement. SpaceX expects to finalize the contract before the end of 2026, with deliveries and payments starting in the first half of 2027. The fixed interest rate will only be determined at the final pricing stage.

"Closed," according to Musk; commitment, according to the 8-K

Elon Musk stated on X that SpaceX had "closed" $40 billion in financing. However, the regulatory document describes something different: a binding commitment, still subject to closing conditions and without a defined price. Bloomberg reported, based on people familiar with the discussions, that negotiations regarding the pricing were ongoing.

This distinction is not merely semantic for those watching the company since its IPO in June. A listed company carries the weight of the difference between what the controller publishes on social media and what is filed with the SEC, and Musk's history with the regulator, including the 2018 settlement regarding the "funding secured" tweet from Tesla, makes this discrepancy more sensitive than it would be for another issuer.

GPU leasing as an asset class

The format matters more than the value. By purchasing the equipment and leasing it, the lender retains ownership of the chips as collateral and transfers the usage risk to SpaceX. This arrangement mirrors the model that CoreWeave popularized with GPU-backed lines of credit since 2023, now applied on a much larger scale with a borrower that has an investment-grade rating. In June, SpaceX sold $25 billion in bonds of that rating to replace expensive debts inherited from X and xAI, according to Bloomberg and BusinessDay.

Musk used the announcement to criticize the so-called neoclouds, providers that lease GPUs and rely solely on revenues from AI datacenters. According to coverage from Yahoo Finance and Bloomberg, he claimed that the debt market has practically closed for these companies and predicted a wave of consolidation in the segment.

There is a counterpoint to note. Musk has a vested interest: SpaceX competes for the same Nvidia chip supply and the same computing customers as the neoclouds. Moreover, the contract itself depends on a pricing that has not yet been finalized, which means that even the highest-rated borrower in the sector has not managed to lock in the cost of capital so far.

Where credit costs are headed

In the United States, CoreWeave and smaller competitors rely on GPU-backed debt to grow; if Musk's assessment of the market is correct, these companies' capital costs will rise, increasing the advantage of borrowers with diversified balances.

In Europe, operators like the Dutch company Nebius and the British Nscale finance their expansion using the same logic of assets leased to AI clients. For them, the relevant comparison becomes the spread that a lender like Apollo charges to a borrower with satellite and launch revenues versus what it charges to those with only computing contracts.

In Taiwan, this operation strengthens the line for Vera Rubin chips manufactured by TSMC, with deliveries starting in 2027, and each contract of this size reduces the negotiation space for smaller corporate buyers.

For CIOs and consultancies selecting where to run AI workloads over the next three years, the practical question shifts from the GPU price per hour to the provider's solidity. A five-year contract with a neocloud whose credit costs are rising carries counterparty risk that does not appear on the price list.