News

Washington is loosening financial safeguards just as risks tied to artificial intelligence are mounting. AI companies will need to generate $2 trillion in new annual revenues to pay their mounting bills. It seems increasingly unlikely they will do so. AI companies will need to generate $2 trillion in new annual revenues to pay their mounting bills. It seems increasingly unlikely they will do so.Heather Hopp-Bruce/Globe Staff; RDVector/Francesco Milanese/Adobe

Matt Scherer is a fellow with the Open Markets Institute. Maya Jenkins is a senior policy analyst with Americans for Financial Reform Education Fund. They are authors of a forthcoming report on the potential economic and policy consequences of the AI bubble bursting.

The AI bubble is a speculative frenzy of historic magnitude, with the potential to cause economic harms worse than the 2008 financial crisis if it bursts.

There are steps policy makers should take to better safeguard the financial system from the threat of a systemic shock. Instead, regulators are taking steps that threaten to inflate the bubble still further and make the consequences of its collapse more severe. The stability of our economy depends on them quickly changing course.

The AI bubble stretches throughout the economy. The US stock market has practically become synonymous with the AI boom; all nine of the most valuable US companies are tech companies that are betting heavily on AI. Those corporations are at the center of a $7 trillion spending spree on data centers and related infrastructure to train and run generative AI models.

While the early years of the AI boom were largely financed by big tech profits, the exploding cost of the build-out, combined with comparatively meager revenues from AI products and services, has increasingly pushed AI companies to debt markets. Nikkei Asia estimates that just five tech giants have racked up an estimated $3 trillion in debt, including $1.65 trillion hidden off their balance sheets (financial arrangements similar to those that led to Enron’s spectacular collapse). That is greater than the size of the subprime mortgage market at its 2007 peak — and the subprime bubble triggered the 2008 financial crisis when it burst.

AI companies will need to generate $2 trillion in new annual revenues to pay their mounting bills. It seems increasingly unlikely they will be able to do so. The industry’s margins are being squeezed from several directions at once by frenzied competition, high costs, and persistent signs that most businesses are not seeing any return on their investments in generative AI. With the gap between the industry’s spending and revenues continuing to widen, an AI crash increasingly seems less a question of “if” than of “when.”

The temptation is to think that if a crash happens, it would, at worst, follow the course of the dot-com bubble, which brought down numerous startups but spared both the industry’s giants and the stability of the financial system. But today’s tech giants have gone all-in on AI in a way their dot-com era counterparts never did, having woven a web of circular financing deals with each other and with AI-focused startups so extensive that graphical representations of them are almost comical. Also in contrast to the dot-com boom, tech giants have taken on huge debts that far exceed even their exorbitant revenues.

The skyrocketing debt also means that a crash likely won’t stay contained to the tech sector. Much risky data center debt is being repackaged and sold to insurance companies and other institutional investors in a manner eerily reminiscent of the subprime bubble. Private equity firms have purchased life insurers and loaded them with risky debt, raising the risk of financial contagion. With international investors showing understandable signs that they are less keen on US assets than in the past, a new financial crisis centered on Silicon Valley and Wall Street could send the economy into treacherous and uncharted waters.

Instead of addressing these mounting risks, regulators are adding fuel to the fire. The Federal Reserve is loosening banks’ capital requirements and weakening stress tests. The Trump administration is opening workers’ retirement accounts to the shadow banking system’s opaque markets. These measures reduce the financial system’s resilience and shift the risks of losses to working people, making risk more attractive to companies even as dangerous speculation runs rampant.

Regulators and policy makers should reverse these dangerous changes and instead require greater transparency and stronger risk-management throughout the financial system. They should demand and scrutinize information on major financial institutions’ exposure to AI-linked debt, as well as adjacent economic trouble spots like the $3 trillion private credit market and private equity-owned life insurers. Regulators should also step up oversight of AI-linked company audits, scrutinize how big bank balance sheets would respond if valuations began to tumble, impose new requirements on Wall Street firms whose failure would threaten the economy, and stop inflated AI stock from being automatically included in retirement plans.

Federal agencies, from the Fed to the Financial Stability Oversight Council to the Public Company Accounting Oversight Board, have multiple tools to identify and mitigate sources of systemic risk. They must use them.

State regulators don’t need to wait for Washington to act. They can investigate and penalize bad actors when AI companies use financial trickery to exaggerate their revenues and obscure the extent of their liabilities.

If a crash does come, the government must avoid the mistake of using taxpayer dollars to prop up or bail out AI companies or the financiers behind them. Bailing out failing corporations is almost always a bad idea, because they create moral hazard, undermine economic justice, and erode faith in the democratic system. An AI bailout would be particularly outrageous given industry leaders’ active role in creating and profiting from dangerous risks. Starting now, policy makers should commit to “no AI bailouts” as a principle. In addition to being the right thing to do, such commitments could play a role in limiting the bubble’s further inflation.

Speculative bubbles operate according to an illogic all their own, and there is no telling how big a bubble can get or how long it can last before it bursts. The only certainty is that the market will not come to its senses on its own. It is up to the Trump administration and the Federal Reserve to wake up and safeguard the real economy from excessive risk. If they fail to take action, Congress should step in and do so. And states should step in to protect their residents if the federal government won’t. Policy makers must take on Wall Street and protect working people — not large corporations and their billionaire shareholders and executives — from the severe economic losses that would surely follow an AI crash.