SpaceX is reportedly seeking $40 billion of debt to buy Nvidia chips, and Nvidia sits on every side

SpaceX is reportedly in talks to borrow about $40 billion to buy Nvidia chips, a financing that would put one of the largest AI hardware orders on the credit markets instead of on a balance sheet. The Financial Times first reported the plan on October 6, 2026, citing people familiar with the matter, and Bloomberg then reported that talks are underway but early, and could still fail. None of the companies involved has confirmed anything.

What is reported

The reported structure is about $10 billion of bank loans and $30 billion of investment-grade debt. Apollo Global Management is expected to lead and sell the debt on to a wide range of investors, with Pimco among a small group of lenders in talks. Closing is expected in 2027. After hours, according to the reports, SpaceX shares fell about 1 percent and Nvidia rose about 0.5 percent.

The chips would power SpaceX's AI data centers. SpaceX absorbed Elon Musk's xAI and went public in June, and it has said it will use Nvidia hardware exclusively in its Colossus data centers. Musk said on the August earnings call, per Reuters, "we think the Vera Rubin architecture is the best." For a comparison with another chip financing built on a leaseback, see our report on Amazon's $8 billion Nvidia chip structure.

Nvidia on every side

What makes this deal unusual is that Nvidia appears in four roles. It is the seller of the chips. It is a shareholder: Nvidia disclosed about 122.8 million SpaceX shares in August, worth about $21 billion at the June 30 price, a stake that came from its 2025 investment in xAI before SpaceX absorbed that company. It is a financier in effect, because on August 10, 2026 it announced memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to create independent compute financing platforms meant to mobilize more than $500 billion of third-party capital for AI infrastructure over time. And it is the supplier of the chips the debt would pay for.

Huang framed the logic in that announcement: "In AI, compute is revenue." The $500 billion is a target, not a commitment, and the August agreements were subject to final documents. Apollo is both a named partner in Nvidia's platform and the expected lead in the SpaceX deal, which suggests this may be one of the first large uses of that pipeline. That link is our inference. None of the reports says the SpaceX deal is part of the platform.

The numbers behind the ask

Some scale helps. Nvidia's data center revenue was $193.7 billion in fiscal 2026, up 68 percent from $115.2 billion, so $40 billion is roughly a fifth of a year of Nvidia data center sales. SpaceX's capital spending was $18.4 billion in the second quarter of 2026, against $2.9 billion a year earlier, according to TIKR's analysis of company results, which means $40 billion covers a little over six months at that pace. As of June the company held $93.5 billion in cash and $6.5 billion in short-term investments, with total debt of $39.7 billion, up from $30.6 billion three months earlier. A full $40 billion would roughly double that debt.

The backdrop is a cautious bond market. The 10-year Treasury yield was near 5.3 percent, and Morgan Stanley estimates AI infrastructure will need $1.5 trillion in external financing by 2028, while warning that lenders are becoming more cautious. Oracle plans to raise $45 billion to $50 billion in debt and stock this year. The question for this deal is not whether SpaceX can pay, given its cash, but whether investment-grade buyers will absorb another very large tranche of AI-linked paper.

Why a race reader should care

Compute is the physical race to superintelligence, and financing is how that race gets funded once labs outgrow their equity. xAI now sits inside a public company with a large cash pile, which gives it options other labs lack. Debt against chips also changes the risk: if chip prices fall or newer generations like Vera Rubin make older systems less valuable, the collateral loses value, which is why the structure of the lease or loan matters as much as the headline.

There is a read-across to data center power as well. Chips need sites, and our data center power tracker shows how many gigawatts are announced against how many are funded. Debt like this is how announced capacity becomes funded capacity.

What would change the picture

Three things to watch. First, confirmation: a filing or a company statement would turn a report into a deal. Second, pricing: the spread over Treasuries will show how lenders rate AI-collateral risk. Third, whether Apollo or others sell the debt on and who ends up holding it, because the exposure then spreads from a few banks to pension and insurance money.

The rental angle

One more detail from The Next Web shows why lenders may like the collateral. SpaceX rents out this kind of hardware: Google reportedly pays $920 million a month for access to about 110,000 Nvidia GPUs. If that figure holds, contracted rental income gives lenders a cash stream to underwrite, which is different from lending against chips that sit idle. It also shows that the hardware is not only for Colossus and the xAI models. Part of it is infrastructure for other labs and customers, which blurs the line between an AI lab and a compute landlord.

Our take

We would not treat $40 billion as settled. Reports from unnamed sources, early talks and a 2027 close mean the size and structure can move. We would treat the direction as real: the AI race is moving from equity rounds to infrastructure credit, with Nvidia as supplier, shareholder and facilitator at once. That is efficient while the cycle is rising and fragile if it turns. This is reporting, not investment advice.