The $300 Billion Question: Is AI's Data Center Boom Built on Hidden Debt?

The trade that tipped people off wasn't in AI chips. It was a loan. Oracle reportedly put together an $18 billion financing package for its data center ambitions, and soon after it closed, reports had the loans changing hands below par — the market's cold way of saying the buyers are already sitting on losses. Fresh corporate debt isn't supposed to behave like that. When it does, credit desks notice, and the question they ask is the one hanging over the whole AI boom: how much of this is borrowed money — and how much of the borrowing is even visible?

How big the bet really is

The reported figures are staggering: Big Tech is said to have poured about $625 billion into AI infrastructure in 2026 alone. Amazon and Alphabet are each said to be spending over $100 billion this year. Then there's Oracle — a database company, not a server-farm builder — reportedly committed to $300 billion. The scale is what makes the financing question matter. This isn't a side project funded out of petty cash. It's one of the largest capex waves in corporate history, and borrowing isn't cheap right now.

Where the debt goes to hide

The clever part is where the borrowing sits: rather than loading debt onto their own balance sheets, the big builders route it through special-purpose vehicles — SPVs. The SPV borrows the money, builds the data center, and leases the finished facility back to the tech giant. Legally the debt belongs to the SPV, so the parent's published leverage looks lighter than its real obligations.

None of this is fraud — it's standard project finance, the same plumbing behind toll roads and power plants. What changes at this scale is the cost of being wrong: with hundreds of billions routed through entities designed to keep debt at arm's length, the true leverage behind the AI trade is bigger than the balance sheets admit. If AI revenues arrive late, the SPVs can't service their debt, and the lease payments and usage guarantees the tech giants signed snap straight back onto their books — at the worst possible moment.

What the watchdogs are saying

The IMF has noticed, warning that debt-financed AI investment could trigger sharp repricing if returns disappoint — the kind of shock that doesn't stay in the stock market. It also flagged "circular financing": a small cluster of firms acting as each other's customers, investors and financiers at once, so one company's capex cut becomes another's revenue hole. Finimize's write-up of the IMF warning cites outside estimates putting global private-sector AI investment above $2 trillion this year. This isn't a prediction of crisis. It's a demand for better visibility. You can't price risk you can't see.

Steel-manning the other side

To be fair to the optimists, history is on their side. Every great infrastructure buildout looked like a debt-fueled mania in the moment — railroads, electrification, the 1990s fiber overbuild. All were financed with borrowed money and all were overbuilt; the debt got restructured, the winners emerged, and the world kept the railways and the fiber.

My read: the optimists' real point isn't that this time is different — it's that overbuilding is how infrastructure gets built, and the biggest builders throw off enough cash to survive their own excesses. The bear case isn't that AI fails; it's that the financing breaks before the revenues arrive.

Three questions I'd want answered

That leaves three questions nobody has answered convincingly. First: are off-balance-sheet AI commitments disclosed well enough to measure the real leverage? Second: do AI revenues compound fast enough to service this debt before refinancing turns painful? Third: in a stack where everyone's each other's customer, what happens when one giant blinks?

The bond market is already voting — that Oracle loan is the tell.