The figure would have been unthinkable for a rocket company a few years ago: $40 billion in fresh debt, most of it for a single order of computer chips. That is what SpaceX is now arranging, according to people familiar with the matter, in a financing led by the asset manager Apollo Global Management to buy Nvidia’s AI processors.
The Financial Times first reported the talks on Tuesday. The package would split roughly into $10 billion of bank loans and $30 billion of investment-grade debt, with the bond manager Pimco among a small group of lenders in discussions to provide capital. The deal is not expected to close until 2027. SpaceX, Apollo and Nvidia did not immediately respond to requests for comment, and Pimco declined.
What makes the deal notable is less its size than what it does to the company’s books. SpaceX has long been valued as a kind of utility for spaceflight, drawing steady revenue from Starlink subscriptions and government contracts. That business is now being used to support a chip order for an AI push that has little to do with rockets. The company’s market value stands near $2.17 trillion, so the borrowing equals a little under 2 percent of it, large in absolute terms but small for a borrower of that scale.
The willingness to lend did not begin with SpaceX. Nvidia has spent the past year assembling what amounts to a lending machine for its own customers. In August it partnered with Apollo and five other institutions, including BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, on platforms meant to funnel more than $500 billion of outside capital into AI infrastructure. Apollo has already worked with Musk’s other venture, xAI, on a $7 billion financing to supply chips for its Colossus 2 data center in Memphis.
For Musk, the logic is familiar: borrow against one business to fund another. He told analysts in August that SpaceX had decided to build its AI efforts entirely on Nvidia hardware, calling the company’s Vera Rubin platform the best available architecture. The chip order, if it lands, would turn that statement into a purchase order.
There is a risk buried in the structure. Debt backed by a rocket company’s cash flow is one thing; debt spent on chips that can lose value quickly is another. Apollo and Pimco are being asked to finance an asset class that barely existed a decade ago, and whose resale value depends on a market that is still taking shape. The 2027 close date gives the parties time to watch how that market behaves before the money actually moves.
For the wider AI buildout, the deal is a data point in a running argument about who pays for it all. Hundreds of billions of dollars in debt have already been raised to fund data centers and the chips inside them, and the borrowers are increasingly the companies with the strongest balance sheets rather than the startups. That is a sign of scale, but it is also a sign that the financing is migrating from venture capital into the credit markets, where the standards are different and the patience is shorter.
Analysts said the transaction, if completed, would rank among the largest corporate borrowings tied to AI hardware, and that its progress will be read as a test of whether bond investors keep funding the buildout as interest rates stay high. For now the market’s answer looks like yes. SpaceX shares slipped only about 1 percent in after-hours trading on the news, a muted reaction to a figure that would have moved the stock sharply in an earlier cycle.
The terms themselves tell part of the story. By structuring most of the raise as investment-grade debt, SpaceX is borrowing at rates reserved for companies whose revenue is judged stable enough to repay, a signal that bond investors now treat the launch business as something close to a utility. The shift is meaningful: a company that spent its early years burning cash on rockets is now borrowing like a blue-chip industrial, and the credit markets are letting it. It also shows how quickly the definition of an AI borrower has widened. A year ago the big names raising money for chips were the model labs and the cloud providers. Now a space company is lining up for $40 billion, and the lenders who once financed launches are being asked to fund a different kind of payload.
What Musk is really doing, people familiar with the talks said, is attaching the launch business’s credit to a bet that AI training will hold its value. If that bet pays off, the chips earn back their cost many times over. If it does not, the company will have borrowed like a utility to buy like a speculator. The next rocket still flies either way; the question is what it is carrying on the books.
