Will the debt behind AI chips outlive the chips?

Borrowing for hardware that could be old news before the last payment clears.

Every borrower learns the same rule early: do not take a loan that runs longer than the thing you are buying keeps its value. A car loan that outlasts the car is a problem. So what happens when the thing you are buying is a stack of AI chips?

Those chips earn their keep only while they are the fastest machines in the room. Each new generation makes the last one less attractive, and the next one, in Nvidia's case the Rubin line, is already on the way. Meanwhile the financing discussed for SpaceX's 40 billion dollar chip order is a mix of bank loans and investment-grade bonds, which run for years. The chips are bought now, the bills keep arriving for years, and by the time the final payment clears, the same kind of machine may cost a fraction of what was paid.

Scale makes it heavier. Big Tech issued roughly 200 billion dollars of investment-grade debt in the first half of 2026, almost double everything it issued in 2025. Whoever holds that paper is betting that the hardware keeps earning money for longer than its market price suggests.

The cushion, and how big it really is

SpaceX reportedly held about 100 billion dollars in cash and short-term investments as of June. Against that, its total debt is reported in sources ranging from about 40 billion to about 65 billion. Under the lower figure, a 40 billion dollar loan adds to a manageable balance. Under the higher one, it roughly doubles the debt. That gap is why the cushion looks comfortable on one reading and thin on another.

And here is the uncomfortable question: who decides how long a chip is useful? The company that owns it, in its own accounts. The longer that number, the better this year's profit looks.

That is the lever. The useful life of hardware is an accounting choice, and stretching it flatters current earnings while the real value drops faster. Nobody has to lie for this to work. They only have to pick a generous number and keep picking it.

Is this a crisis by itself? Not necessarily. Plenty of infrastructure is financed against assets that lose value, from aircraft to fibre networks. The difference here is that the useful life is short and uncertain, the debt is long, and the amounts are big enough that a shortfall would be felt well beyond the companies involved.