A new report scheduled for presentation at the Brookings Institution today puts the total cost of the AI buildout through 2032 at $10.3 trillion. That equals 3.63% of GDP each year, many times the scale of previous major infrastructure investments.

Those earlier projects included canals, railroads, telephone and electric grids, and the interstate highway system. Unlike AI, whose benefits remain speculative and unproven while its risks are difficult to fathom, those investments delivered productivity gains across the broader economy.

The Brookings report warns that much of the AI expansion is being financed through debt. Some of that borrowing remains obscured through financing kept outside corporate balance sheets, while the immense debt burden crowds out more productive borrowing and pushes interest rates higher.

Some of the debt will never be repaid, according to the article. Meanwhile, much of the stock market boom and the continued growth in GDP and employment rests on what the author describes as the artificial and unsustainable stimulus of the AI bubble.

Citizen resistance to enormous data centers is also growing. Combined with concern among some AI executives that the industry has overreached, higher interest rates driven by Federal Reserve policies, and bond market panic, the conditions could produce a financial collapse comparable in scale to 2008 or 1929.

A warning sign arrived Thursday, when Oracle disclosed trouble involving a massive AI data center planned for New Mexico. Oracle notified property developer Blue Owl Capital that it would be unable to make its payments on time, with the company carrying $18 billion in debt on the deal.

Oracle invoked “force majeure,” a standard contractual provision generally used when wars or weather disasters prevent a party from fulfilling its obligations. Here, Oracle cited delays in obtaining necessary regulatory permits and protests from citizens, rather than some uncontrollable calamity.

Robert Kuttner writes that courts have viewed similar claims skeptically when creditors bring litigation. Several force majeure claims emerged during COVID, but courts rejected them after finding that the debtors still had enough money to meet their obligations despite the pandemic.

Oracle and Blue Owl Capital issued statements presenting the dispute in a favorable light, but shares in both companies fell sharply. The market was apparently less impressed by the corporate gloss than the companies might have hoped.

Oracle faces pressure well beyond the New Mexico project. Its stock is down 50% over the past year, while its long term debt has climbed above $117 billion, a 43% increase from the previous year.

S&P downgraded Oracle’s credit rating to one notch above junk bond status. Oracle must also produce billions more to satisfy the terms of its recent settlement in an antitrust lawsuit, even as CEO Larry Ellison has assisted his son David with $40 billion in personal guarantees to build the Paramount empire.

Whether Oracle’s force majeure maneuver becomes a Bear Stearns moment remains uncertain. The alternative is that investors keep whistling past the graveyard, as the broader stock market was still up about half a percentage point at the time of writing.

Citigroup CEO Chuck Prince once defended speculative trading shortly before a market collapse by telling the Financial Times, “As long as the music is playing, you’ve got to get up and dance.” That logic remains a fitting anthem for supposedly efficient, lightly regulated private markets dancing through another potential disaster.

The debt beneath the AI frenzy looks even more alarming under closer examination. Recent Goldman Sachs research found that between 33% and 37% of all capital expenditures by AI hyperscalers are financed by debt, without counting borrowing kept outside balance sheets.

An unusual alliance is now taking shape among citizens fighting data centers, AI executives seeking a reason to pause expansion plans, and supporters of regulation aimed at preventing rogue AI catastrophes. President Trump’s aggressive support for unrestricted AI expansion puts him at odds with increasingly vulnerable Republican House and Senate candidates and much of the MAGA base.

The AI industry, meanwhile, is promoting a cartel built around soft self regulation. Kuttner argues that regardless of how the regulatory debate unfolds, an almost certain retreat from inflated expansion plans will reveal that much of the accumulated AI debt cannot be paid.

That reckoning could carry dire consequences for the wider financial sector. As Warren Buffett once said, “You never know who is swimming naked until the tide goes out,” and the AI boom may soon put that warning to another brutal test.