SpaceX Wants to Borrow $40 Billion to Buy Nvidia Chips: The AI Race Just Became a Debt Race
October 8, 2026 · 5 min read
This is the week the AI boom started printing IOUs. On October 6, the Financial Times reported that SpaceX is in talks with banks and asset managers about raising $40 billion — earmarked for a single purchase: Nvidia AI chips. That's not a rounded-up estimate or a multi-year pledge; it's a financing package being assembled right now, for chips. Roughly $10 billion would come from bank loans and $30 billion from investment-grade bonds, led by Apollo with PIMCO participating, with the deal expected to close in 2027. (via Channel News Asia)
Stop and look at the number again: $40 billion, for chips. SpaceX builds rockets and Starlink terminals, but this money isn't for any of that — it's reserved, up front, for Nvidia hardware. The detail that matters isn't the headline figure; it's the structure. Loans plus bonds means this is debt, not cash from the balance sheet. And debt has a repayment schedule. When a company borrows $40 billion to buy processors, the processors stop being a purchase and become a bet — the bet being that the AI models they train will be worth more than the interest. The AI arms race just grew its own financing arms race.
Forty billion dollars, one shopping list
Break the $40 billion down and it's a very conventional Wall Street package: about $10 billion in bank loans — flexible, expensive — and about $30 billion in investment-grade bonds, the kind of paper pension funds can hold. Apollo is leading the raise, PIMCO is participating, and the money is earmarked: lenders know exactly what it's for — Nvidia chips. Earmarking is the tell. You don't earmark $40 billion unless the thing you're buying is both urgent and non-negotiable. This is what buying compute looks like when the buyer can't wait for cash flow to catch up with demand. The deal is expected to close in 2027, which means the chips — and the data centers they'll fill — are already being planned around money that hasn't arrived yet.
Why Vera Rubin — and why Colossus 2 doubles down
The chips aren't generic. In August, Musk said xAI's data centers would go all-in on Nvidia's Vera Rubin architecture — Nvidia's newest platform — and that Colossus 2's chip count would double before year-end. That's the context for the $40 billion: it's not a stockpile, it's a build plan. When the head of a company publicly commits his data centers to one architecture and then starts raising the largest chip-financing package on record, the two announcements are one strategy. The doubling of Colossus 2 is the proof of concept; the $40 billion is the scale-up.
The $1.5 trillion behind the headline
Zoom out and the $40 billion starts to look small. Morgan Stanley estimates that by 2028, AI infrastructure will need $1.5 trillion in external financing. Trillion, with a T. This deal is a rounding error on the coming capital bill. That's the real story: the chip race is turning into a financing race. The bottleneck isn't just Nvidia's supply chain anymore; it's who can borrow, at what rate, and how fast. The companies that win AI may turn out to be the ones that can sell the most bonds, not the ones with the best models.
Who's footing the bill
Follow the money and you land on Wall Street's biggest credit desks. Apollo is leading the raise; PIMCO is participating. These are not venture capitalists betting on a startup — they're fixed-income giants lending against the world's most valuable private company to buy the world's most in-demand chips. That lineup tells you how the market now treats AI compute: not as a speculative tech bet, but as infrastructure — closer to a toll road than a startup. When Apollo and PIMCO start financing GPUs the way they finance airports, the arms race has officially become an asset class.