At the end of a previous lending boom, back in 2007, then-Citigroup CEO Charles Prince famously said: "As long as the music is playing, you've got to get up and dance."
This time around, with artificial intelligence now a driving force, banks are definitely up and dancing. But they are also eyeing possible seats and listening more keenly for a pause in the music.
In the third quarter, both big banks and midsize regionals reported sharp jumps in their traditional business-loan books, known in the industry as commercial-and-industrial lending. The AI data-center build-out was often cited as a driving force.
The biggest banks are where giant "hyperscaler" tech companies and big developers are going to get loans or bond offerings for data-center build-outs. But regional lenders are also active supporting construction companies and others benefiting from the boom.
At the 25 largest banks in the country, "C&I" loans grew almost 10% year-over-year in the third quarter, according to Federal Reserve data. Outside of extreme swings in lending around the Covid-19 pandemic, that is the fastest growth in over a decade. Big banks' commercial-real-estate construction lending also rose quarter-over-quarter for the first time since early 2024.
This ought to be a moment of vindication for bankers after seeing years of commercial lending shift to nonbank competitors. It is even happening in the face of rising interest rates. But bank shares are sharply underperforming the market, with the KBW Nasdaq Bank index up around 5% this year, versus a nearly 13% rise for the S&P 500.
With doubts swirling about the sustainability of the AI build out, banks aren't talking up their growth potential from here, but their caution.
"In this moment where AI is such an overwhelming theme, it's producing so much innovation in so many different pockets, there's capital being tapped through all kinds of different mechanisms," JPMorgan Chase Chief Financial Officer Jeremy Barnum told analysts earlier this month. "We just need to be really, really on top of things to make sure that nothing slips through."
Even several months ago, big banks were becoming reluctant to keep lending more to specific clients, such as Oracle, The Wall Street Journal has reported. Beyond individual concentration risks, they now have to monitor the complex interlinking of many AI-financing arrangements and the various kinds of lending that goes into each project.
Wells Fargo CFO Mike Santomassimo told analysts earlier this month that his bank was very focused on ensuring it understands "who the lessees are, where the guarantees are, all the different sort of layered risks that could potentially come into play there."
"We have very hard limits in terms of what we're willing to do, and what we're not willing to do, and are sticking to that," he said. "And it's a conversation that we're very focused on, from the top down across all the different businesses that are impacted by it."
About four years ago, Providence-based Citizens Financial acquired DH Capital, a boutique advisory firm that specializes in tech infrastructure like data centers. It was fortunate timing as AI took off, and Citizens has since put its own balance sheet to work in these kinds of deals.
Still, the bank says it has several guardrails around that lending. This month, executives told analysts that most of its exposure is for data centers being leased to hyperscalers and to data centers with committed power and regulatory approvals.
"We have well-defined buckets as to how much we want to hold, and we're also hedging some of those exposures as well," Citizens Chief Executive Bruce Van Saun told analysts. "Because even though it seems like you can't miss right now in the data-center space, you're kind of always wondering, does the pin go in the bubble someday? And so, we have to think about those scenarios as well."
A lot of data-center financing is being done by private-asset managers and developers, who set up special entities to borrow. Those tend to be clients of the Wall Street banks. But even on Main Street, regional banks' core lending to manufacturing and construction firms is also picking up along with AI capital expenditures.
Cleveland-based KeyCorp Chief Executive Christopher Gorman described the AI build-out as having "long tentacles."
Tim Spence, chief executive of Cincinnati-based Fifth Third Bancorp, said in an interview: "We're not lending money to the people sponsoring the build of data centers. We are lending money to people who construct data centers."
This kind of lending isn't directly tied to risks such as demand for computing power falling short of expectations. But there is still risk, for instance for a construction company that takes on new workers or equipment for a data-center project that doesn't ultimately happen. A key question then will be if the broader economy is strong enough to generate other uses for those services and equipment.
"It's hard to come up with an answer on how much of the AI capex is driving loan growth broadly across the economy, because there's clearly tangential effects on any sort of heavy equipment, transportation, energy component producer," PNC Financial Services CEO Bill Demchak told analysts this month. "So if the capex broadly slows down related to AI, do we think -- even though we don't lend directly to that -- that it would slow our loan growth? Yes, we do."
There are indicators that the non-AI economy is more tepid. For example, while spending in the U.S. on power-related construction was up about 9% in August, spending on residential construction was down about 5%, according to Census Bureau data.
Still, there is a broader need for physical investment that the AI build-out has spotlighted. KeyCorp's Gorman told analysts that the marginal driver of power-utility investment, for example, might be AI. But Gorman also said that "the investment cycle for these publicly owned utilities is one that's literally decadeslong, and we've been deficient in the amount of power that we've had. So I don't think that stops."
Fifth Third's Spence noted that a lot of demand in construction and materials industries is also still coming from government-backed infrastructure investments, related to the 2021 Bipartisan Infrastructure Law.
In general, banks' reticence to keep directly financing data centers is one reason that many companies have turned to the bond market to raise money for the kind of construction projects usually financed via bank loans. That, in turn, has put the spotlight on markets' pricing and risk appetite. Banks' hedging activities can involve buying credit-default swaps or selling off portions of loans. That adds to the supply of credit in the market and can push up prices, too.
It also means that this might be a different kind of lending boom from the one leading up to the 2008 crisis, when banks were far more willing to retain direct risks.
But with the scale of the spending planned on AI, banks won't be able to totally insulate themselves from the build-out's ups and downs.
Comments