The International Monetary Fund just named a $1.4 trillion financing loop the biggest risk to world markets. The AI technology might be fine. The way we are paying for it is not.

Every financing story ends with the same question: who pays if the bet sours? Start with your power bill.

Investor-owned utilities have asked regulators for $23.1 billion in rate increases so far this year, according to data from PowerLines, filings affecting 45.5 million customers, driven in large part by data-center electricity demand. The AI boom runs on power, and the power gets billed to households, first.

Next is your town’s tax base. Indianapolis just handed about $240 million in tax breaks to a data-center developer promising 75 jobs. That is $3.2 million in foregone public money per job. Cities are bidding against each other to host these facilities, offering discounted electricity and infrastructure upgrades for returns that may never arrive. A federal bill to protect ratepayers failed in the Senate last month.

Then comes risk to your retirement. Morgan Stanley estimates data centers need $1.5 trillion in outside money through 2028, with about $800 billion coming from private credit, bankrolled substantially by pension funds, insurers, and ordinary investors in non-traded funds. These loans barely trade. They are hard to value. Apollo is about to start pricing $850 billion in credit assets daily after regulators raised questions. Fitch says the private-credit default rate just hit a record 6.3%

This is where the money goes. Now look at how it gets there.

The IMF looked at the artificial intelligence boom this month and decided the technology is not the scary part. The financing is.

In its October Global Financial Stability Report , the IMF flagged the way AI is being funded as the top trigger for a global market shock. It examined 14 major AI deals since 2023 and found something it called unusual: a loop.

Here is the loop. The big cloud companies, the hyperscalers, have pumped about $300 billion in equity into smaller AI firms, plus roughly $650 billion in debt, some of it parked off the balance sheet. In return, those AI firms have promised $1.4 trillion in future revenue straight back to the hyperscalers: data-center contracts, compute purchases, service agreements.

Read that again. The money that is supposed to pay back the debt has been promised to the companies that lent it. Jason Wu of the IMF called the downside a “triple whammy”: “the equity investments will deplete in value, the debt may be difficult to service, and the revenue commitments in the future may not materialize.” His summary of the arrangement: “What makes this a little bit unusual compared to previous deals is the future revenue commitments, hence the circularity.”

The IMF’s term for the structure is “wrong-way risk.” Everything gets impaired at once.

For most of their history, big tech companies paid for things out of pocket. The Bank for International Settlements titled its January report on this shift “Financing the AI boom: from cash flows to debt.” Translation: the cash machine days are over. The buildout is now funded the way everything else gets funded, with borrowed money.

BIS chief Pablo Hernandez de Cos said last month that the biggest AI companies are now spending more than $1 trillion across 2025 and 2026, increasingly beyond what their cash flows cover, bridged by debt and private credit. Private credit to AI firms hit $40 billion last year. In 2010, it was $3 billion.

The numbers keep climbing. Barclays counts $346 billion raised by hyperscalers, data-center vehicles, and AI cloud firms so far this year, more than double all of last year. Alphabet sold a 100-year bond in February , the first tech century bond since Motorola in 1997, and investors lined up ten times over for a piece of it. When lenders fight to fund you for a century, ask what they think they know.

JLL, the real estate firm, calls this an infrastructure investment supercycle: nearly 100 gigawatts of new data centers by 2030, requiring up to $3 trillion . Nine U.S. corporate bond deals of $25 billion or more have priced this year, the most ever, and most of them are tech.

Even The Lenders Are Nervous

Here is the tell. The same Wall Street arranging this financing is starting to hedge in public. Goldman Sachs CEO David Solomon told analysts this week the U.S. data-center buildout could need “north of $10 trillion” through 2032, about 3.6 percent of GDP every year. Then he added: “There will be a point in time when this recalibrates, resets, for sure. And when that does happen, of course, there will be some bumps.”

Asked whether financing is the choke point for AI spending, Solomon said Goldman is “highly selective about the risk we underwrite and the capital we commit, particularly as client financing needs become larger and more complex.” Read that as: we see it too, and we are picking our spots.

The credit markets agree. Bloomberg reports traders now price a greater than 20 percent chance Oracle defaults within five years . Sixteen percent for SpaceX. More than 7 percent for Nvidia, the most valuable company on earth. These are market-implied odds, not predictions. But fear has a price, and this is it.

Ray Dalio, speaking at a Forbes conference in Singapore last week, called AI “a classic bubble” and said “I think we’re close to” the breaking point. His kicker: “Everybody says ‘I’m worth a billion dollars’ but OK, try to spend that.”

Tie it together: the technology changes in months. The assets are financed over decades. The BIS warned this kind of repricing has “the potential to be equally disruptive” as the credit shock of 2008. Not the same crisis. The same shape.

“What’s troubling is that ordinary people are being exposed to the same speculative bet in several different ways,” said Tonantzin Carmona, director of economic policy at The Century Foundation. “They’re already helping finance the infrastructure through higher utility bills and public subsidies, while their retirement savings may also be tied to the investments themselves. And if those investments don’t pay off, it’s not at all clear how far the financial consequences could spread, or how much of the burden could ultimately fall on households.”

Nobody has to predict a crash to see the problem. The IMF didn’t. It just read the contracts and found a machine that only works if everything goes right at once. The AI boom might be the future. But right now it is a $10 trillion bet placed with borrowed money, and the people holding the ticket are households, towns, and pension funds. Ask who signed them up