Big banks acknowledge AI financing risks but keep funding the boom

America’s biggest banks spent their third-quarter earnings calls listing everything that could go wrong with financing artificial intelligence.

Then they explained why they’re going to keep doing it anyway.

Executives at JPMorgan Chase, Wells Fargo, Bank of America, Goldman Sachs, Citigroup, and Morgan Stanley all addressed the risks of bankrolling AI infrastructure, from data centers to the power plants that keep them running. None signaled an immediate retreat. The fee income, as they see it, is simply too good to walk away from.

The risks banks named, and the money they’re chasing

Bank leaders pointed to two main worries. First, demand for computing power remains uncertain. Second, some of these projects may not deliver the returns their backers expect.

Estimated bank exposure to AI-adjacent sectors sits at around $800 billion. That figure includes traditional bank lending plus money flowing through private credit structures and dedicated project vehicles.

For context, that exposure is approximately 0.8% of bank assets.

Total AI-related capital expenditure by leading tech firms is estimated to exceed $1 trillion over 2025 and 2026.

What individual executives said

Wells Fargo CEO Charlie Scharf noted that the AI buildout is increasingly resembling previous infrastructure booms.

JPMorgan CEO Jamie Dimon indicated the bank is tightening credit terms in response to potential project risks.

Goldman Sachs’ CEO cited evolving uncertainties around how much capital the AI sector will ultimately need.

Bank of America emphasized how important AI investment has become for its investment-grade debt and leveraged finance businesses.

Across the calls, executives also raised concerns familiar from earlier buildouts: high leverage, potential stress from interest rates, and public opposition to where data centers get built.

How banks are trying to have it both ways

The banks are using several tools to keep the risk from piling up on their own books.

One is syndication. Rather than holding a whole loan, a bank splits it up and sells pieces to other lenders and investors.

Another is significant risk transfer, or SRT. In these transactions, banks pay investors to absorb a portion of the losses on a pool of loans, offloading concentrated data-center debt.

Moving debt off bank balance sheets doesn’t eliminate it. It relocates it, often into private credit funds and other investors that face less regulatory scrutiny.

Echoes of past tech booms

The Bank for International Settlements has flagged potential financial stability risks tied to the AI buildout, risks that mirror previous tech booms.

What this means

With exposure at approximately 0.8% of bank assets, a stumble in the sector would hurt but likely wouldn’t define a bank’s year on its own.

A sizable piece of that estimated $800 billion runs through private credit and project vehicles. If demand for compute disappoints or interest rates bite, losses could surface in corners of the market that are harder to monitor. That is precisely the sort of scenario the BIS has warned about.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.