Win McNamee
WASHINGTON, DC - OCTOBER 08: Tesla and SpaceX CEO Elon Musk attends the Golden Age of American Innovation Summit at the Institute of Peace on October 08, 2026 in Washington, DC. Trump is expected to award some of the nation's highest scientific honors to technology leaders including Elon Musk, Sergey Brin, Jensen Huang, Lisa Su, Michael Dell, and Satya Nadella.

US space and artificial intelligence company SpaceX (Nasdaq: SPCX) said on Monday (October 26) that it had signed a binding commitment letter with an entity managed by Apollo Global Management (NYSE: APO) to enter into a master GPU financing lease worth about US$40 billion (approximately HK$312 billion). The deal will be used to acquire Nvidia (Nasdaq: NVDA) next-generation Vera Rubin chips. SpaceX described it as the largest committed GPU financing to date.

A day before the announcement, SpaceX chief executive Elon Musk said on social media platform X that the company had "completed" US$40 billion in GPU debt financing. He attributed the success to SpaceX's investment-grade ratings and revenue from Starlink and its rocket-launching business. In the same post, he warned that debt markets had in recent weeks almost completely shut their doors to companies without significant revenue beyond AI data centres. More than 200 emerging cloud computing providers, or neoclouds, would face widespread failures and consolidation, he said: "Some will make it through, but not many."

Apollo to buy equipment as SpaceX leases it in instalments

According to SpaceX's announcement and filings with the US Securities and Exchange Commission (SEC), the two sides signed the commitment letter on October 24. The master lease agreement itself has not yet been executed and remains subject to specified closing conditions. Completion is expected by the end of this year. Musk's description of the deal as "completed" does not match the wording in the company's filings.

Under the arrangement, a lessor managed by Apollo will purchase GPU systems and related equipment, then lease them to SpaceX for use at data centres selected by the company. The equipment will be delivered in batches, with a separate lease for each batch. Lease terms will run for about 5 to 5.25 years, with the first batch expected to be delivered and placed under lease in the first half of 2027. SpaceX will make quarterly payments, generally covering one-twentieth of the equipment financing cost, plus interest calculated at a fixed annual rate and certain fees. The US$40 billion represents the estimated total principal and excludes interest, meaning SpaceX's actual payments will be higher. The fixed interest rate will be set before the master agreement is signed and has not yet been disclosed.

The financing commitment is being led by Apollo and Goldman Sachs, with Goldman Sachs acting as sole structuring arranger. BlackRock, funds and accounts managed by Goldman Sachs Asset Management, and capital accounts advised by KKR have also made cornerstone commitments. The amounts committed by each party were not disclosed.

The financing lease allows SpaceX to secure the use of a large amount of hardware with limited immediate cash outlay. The company had about US$93.5 billion in cash at the end of June, according to its financial data, so the transaction is not intended to address a cash shortage. Instead, it allows SpaceX to retain the proceeds from its public listing as a buffer. SpaceX listed on Nasdaq on June 12 this year and received investment-grade ratings from Moody's, Fitch and S&P in the same month: Baa1, BBB+ and BBB respectively.

It remains unclear whether the debt will be kept "off balance sheet". Some reports have said the financing will be arranged through a special-purpose vehicle (SPV) to protect SpaceX's credit rating. However, SpaceX itself has described the transaction as a "financing lease", which under US accounting standards generally must be recognised on the balance sheet as a lease liability. The final accounting treatment will depend on the full agreement and future financial reports.

Why bet on Vera Rubin?

At SpaceX's first earnings call after its August listing, Musk said the company's AI computing capacity would use Nvidia products exclusively because he believed Vera Rubin was the best architecture available. SpaceX aims to have more than 2 gigawatts of computing capacity by the end of this year and nearly 10 gigawatts by the end of next year.

Vera Rubin is no longer a product on paper. Nvidia said shipments began in the third quarter of this year, with volumes increasing in the fourth quarter and rising further in the first half of 2027. The delivery schedule for SpaceX's equipment therefore coincides with the expected peak in Vera Rubin shipments. With global AI computing capacity still in short supply, signing early helps ensure the company is not pushed to the back of the queue when production capacity is allocated.

SpaceX is itself a neocloud

The more than 200 neoclouds referred to by Musk are in fact SpaceX's competitors. After acquiring xAI in February, SpaceX took over xAI's Colossus data centre in Memphis, Tennessee, and began leasing computing capacity to other AI companies. Anthropic has reportedly committed to paying about US$1.25 billion a month to lease roughly 325,000 Nvidia GPUs. According to filings submitted by SpaceX to the SEC, Google has also committed to paying US$920 million a month from October 2026 to June 2029 to lease about 110,000 GPUs. Either party may terminate the agreement from 2027 by giving 90 days' notice.

Musk's warning therefore comes from a direct competitor backed by Starlink cash flow. His argument is that financing access is becoming polarised: large companies with strong credit and diversified revenue can still borrow heavily to expand computing capacity, while operators that rely solely on renting out computing power cannot raise funds. He also said on X that large numbers of half-built data centres could emerge in future, leaving computing capacity stranded and unusable. Asked by users whether SpaceX planned to buy competitors at distressed prices, he said the company would not acquire any neocloud.

That forecast does not mean every cloud service provider will be forced out. Operators focused on vertical sectors such as healthcare and finance retain a degree of customer stickiness. Those under the greatest pressure are providers that depend on flexible general-purpose GPU capacity while lacking sufficient capital of their own.

Who benefits and who faces pressure?

Nvidia is positioned on almost every side of the transaction. It is the chip seller and, having invested in xAI at an early stage, also holds about 122.8 million SpaceX shares. In August, Nvidia teamed up with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to establish multiple computing-capacity financing platforms, with the aim of mobilising more than US$500 billion in third-party capital. Apollo, BlackRock, Goldman Sachs and KKR are all on that list, but the parties have not said whether this transaction is part of those platforms. Based on US$40 billion in principal, purchases of GPUs and related equipment would run into the tens of billions of dollars, making Nvidia the biggest beneficiary.

For Apollo and the other investors, the transaction is a stable cash-flow asset linked to the AI expansion cycle and running for about five years. The key risk is the second-hand value of the Vera Rubin equipment at the end of the lease, as a newer generation of chips is likely to have been launched by then.

The structure itself is not unusual. This summer, Anthropic entered into a US$36 billion structured-note transaction led by Apollo and Blackstone and backed by Broadcom to purchase and lease GPUs. Apollo had also previously provided US$3.5 billion and US$3.4 billion in financing for GPUs leased to xAI. What is difficult to replicate is not the transaction structure but the borrower's credit quality — the distinction Musk was referring to.

What to watch next

  • Closing and terms: Whether the master lease agreement can close as scheduled before the end of the year, and the full terms disclosed in the definitive documents.
  • Interest-rate pricing: The fixed rate will be set before signing and will show how the market prices AI-related debt.
  • Delivery and infrastructure: Whether deliveries in the first half of 2027 proceed on schedule, Nvidia's volume ramp-up runs smoothly, and power and cooling infrastructure are ready at the same time.
  • Neocloud developments: The financing costs and consolidation trends among computing providers such as CoreWeave and Oracle Cloud Infrastructure (OCI).
  • Accounting and regulation: How the lease is ultimately recorded, as well as policy changes in different jurisdictions covering data-centre construction, energy use and chip export controls.