- Key insight: Banks have a great deal of exposure to America's AI industry, which could become a major problem in the event of a growth slowdown or crash.
- Supporting data: The banking sector has already lent about $800 billion to businesses involved in building out AI's physical infrastructure, a Columbia finance professor has estimated.
- Expert quote: "You do not need a big shock to the demand for that debt to be in trouble." — Stijn Van Nieuwerburgh, professor of finance and real estate at Columbia University
In recent weeks, fears about AI have veered toward the apocalyptic. But even if the technology doesn't end the human race, it could still hobble the global economy.
It's not hard to imagine the chain of events: AI companies fund the construction of data centers partly with bank loans. Then, as those firms roll out their products, revenue fails to meet expectations. The AI companies default on their debts, and banks take the hit, depleting their capital and hurting their ability to lend to other businesses.
"Banks play a central role in the functioning of the broader economy," Stijn Van Nieuwerburgh, a professor of finance and real estate at Columbia University, told American Banker. "So if banks get exposed to data-center loans that are going bust, then that impacts their ability to lend to the rest of the economy."
Whether that causes a minor recession or
Adding up all the debt they've issued — including mortgages, syndicated loans and project debt — Van Nieuwerburgh estimates that banks currently have about $800 billion riding on the fate of AI. And as the buildout continues, that number is only expected to rise.
"A lot of banks are pretty much up against their concentration limits in their data-center lending, so they have tons of exposure," Van Nieuwerburgh said.
And there are many potential threats to American AI revenue. One is that demand for the products may simply not prove strong enough. Another is that Chinese competitors may produce cheaper, popular alternatives, driving prices lower. Yet another is that
Each one of these dangers has shown signs of coming to fruition. And in Van Nieuwerburgh's view, each of them could be enough to push AI-related loans underwater.
"A lot of the time … the debt could be as much as 90% of the overall project cost, and the equity is only 10%," Van Nieuwerburgh said, "which means that you do not need a big shock to the demand for that debt to be in trouble."
Power banking
Data centers eat up enormous amounts of electricity, sometimes requiring the expansion of local power grids just to meet their demands. A number of banks have stepped up to finance that expansion.
One notable example is KeyCorp. The $191 billion-asset bank's loans to utility companies have soared by almost 40% in three years, reaching $10.1 billion in the second quarter of 2026. Over the past year, utilities accounted for 61% of the Cleveland bank's loan growth.
"I don't think most people realized how big it had gotten," Brian Foran, a bank analyst at Truist Securities, told American Banker.
Much of that loan growth, Foran said, was "turbocharged" by the AI boom.
"AI is clearly driving a wave of capex and investment for utilities," he said. "I think they've benefited from the trickle-down effects of those AI investments into the utility grid."
So what would happen if the buyers of all that extra electricity started struggling to pay their bills? Foran noted that, historically, utilities have been a "low-risk" sector. Their revenue is regulated, and a certain amount of demand is guaranteed — but removing the AI element would likely take away the explosive growth.
"I want to say they'd be okay," Foran said. "You never really know until the tide goes out."KeyCorp declined to comment for this story.
AI's energy needs are also a political liability. Americans across the ideological spectrum have
"I think many of the technology companies have been deeply surprised by the level of opposition and the speed at which that opposition has mobilized," Jeremy Fisher, principal advisor for climate and energy at the Sierra Club, told American Banker.
For businesses that took out loans to build data centers — and their lenders — government policies that curtail construction could wreak financial havoc.
"If there's a clamp-down on new data-center construction, but you've already started, and you've already issued the debt to finance it, now you're stuck," Van Nieuwerburgh said. "You don't have any revenue to pay that debt."
Ripple effects
Even banks that are not directly involved in building or powering data centers have some stake in their future, simply because AI's influence is everywhere in the U.S. economy.
"It's just an enormous ecosystem supplying that whole chain of AI infrastructure," Van Nieuwerburgh said. The economy is "very dependent" on it, he said, "all the way upstream to chip manufacturers and memory producers, downstream to all the energy suppliers, the plumbers, the electricians."
Some banks have said they're aware of this interconnectivity, and are working to minimize its risks. During the most recent earnings call of Regions Financial, the Birmingham, Alabama-based bank, an analyst asked how the company
"We're trying to have discussions on a routine basis, just in terms of understanding what's in our portfolio, what the connectivity is and doing different kinds of stress analysis to say: If this particular sector has some weakness, how does that affect us?" Regions CEO John Turner answered.
It's an important question, considering the sheer size of AI's role in the economy. Van Nieuwerburgh has estimated that if the AI buildout continues at its current pace and finishes in 2032, the total investments would average about 3.6% of U.S. gross domestic product per year — more than any previous infrastructure boom, including the creation of the railroads and the internet.
So if the AI bubble were to burst, every bank — whatever its level of involvement in the construction of data centers — would almost certainly feel the ripple effects.
"Any time you take out a major source of growth for the economy, then you get into the broader" impacts, Foran said. "What about credit-card exposure? What about unemployment rates? What about broader C&I exposures?"
Postponing the pop
Not everyone sees trouble ahead. In recent months, bank executives have generally made bullish comments about the state of the U.S. economy. During the latest round of
Brendan Coughlin, president of Citizens Financial Group, doesn't believe an AI crash could derail all that.
"Even outside the AI economy, the non-AI sectors of the U.S. are still solid," Coughlin told American Banker. "So even with an AI slowdown — if it happened, which I don't expect it will — it could take a little bit of the air out of the balloon, but the rest of the rest of the U.S. economy is still on pretty solid ground."
Others, remembering previous tech investment cycles, have indicated they believe an eventual slowdown is likely. In July, Goldman Sachs CEO David Solomon said AI spending was
"Ultimately, you will have a recalibration, a reset, a drawdown and then a further acceleration," Solomon said. "That's what the path generally looks like."
Read more:
IDscan's breach shines light on vendor risk in the age of AI Lessons from 9/11: How one NYC community bank rebounded Safety and soundness rule puts nonmaterial risk onus on banks Banks, credit cards fall out of favor for bill pay
Even if a crash is on its way, Van Nieuwerburgh does not think it will happen immediately. So many things are slowing down the buildout — from energy bottlenecks to manufacturing challenges to tech leaders' own recent pleas for a slowdown — that the supply of AI computing has remained somewhat constrained. That allows prices, and therefore revenue, to remain high for now.
"It gives us time to maybe see the crash coming ahead of time, and maybe slow things down," Van Nieuwerburgh said. "I don't have a crystal ball, and the world is changing very quickly. But at least for the foreseeable future, I don't see sort of an imminent crash around the corner."
At the same time, looking back at previous tech advances can be instructive — and not reassuring. The dot-com boom of the late 1990s, for example, was followed quickly by the stock-market crash and recession of the early 2000s.
"I think if we look at history, these types of buildouts are often associated with financial crises, or with crashes," Van Nieuwerburgh said. "I think that tends to occur later in the cycle."
There's also another chapter of history that both Van Nieuwerburgh and Foran pointed to: the Global Financial Crisis. In that case, a housing bubble eventually burst as massive numbers of subprime mortgages went bad — and banks were a key link in the contagion to the rest of the economy.
It remains to be seen whether AI will cause economic problems anywhere near that magnitude. But one thing the two booms have in common is their deep connections to banks — and therefore to many other businesses.
"I'm sure there are other industries, other books, other exposures that are more tied to AI than people realize," Foran said, before recalling a quote from 2007 by Citi's then-CEO, Charles Prince: "As long as the music's playing, you've gotta keep dancing."
