A note in the Federal Reserve Bank of Chicago's series Chicago Fed Insights, dated February 2026, is titled "Tail Risk for Banks Posed by Investments in Generative Artificial Intelligence." It looks at how much large banks lend to what it calls AI-adjacent borrowers, and at tail risk, which it defines as "the risk of losses due to extremely rare events, at the left tail of a probability distribution curve (more than three standard deviations from the mean)." A line under the note says the opinions in it are the authors' own. Parts of the note follow, not the whole.
What the Note Counts
The note is a snapshot of lending by large banks, which a footnote defines by their total assets and a Federal Reserve report they file. "AI-adjacent lending consists of commercial & industrial (C&I) and commercial real estate (CRE) bank loans made to AI software companies and AI infrastructure companies, as well as loans for the construction of data centers and loans that are secured by these centers." Its commercial and industrial exposures are identified by the borrower's industry, and a footnote says that "not all companies in these industries have exposure to AI."
The note focuses on direct lending. Banks "most likely have additional exposure to AI-adjacent industries through lending to nonbank financial institutions (NBFIs)," the note says, such as a private credit lender funding a data center. Those channels are hard to quantify with regulatory data, so the authors leave them out.
Direct bank outstanding exposure to AI-adjacent industries "is generally considered relatively low," the note says. By the authors' own estimate, the average bank's outstanding amount to those industries is around 0.8% of bank total assets. "However, there is notable tail risk." In a tail risk event, it says, stress in one AI-adjacent industry "can spill over to multiple interconnected AI-adjacent industries." If software companies were stressed and cut their spending on infrastructure, for example, the note says semiconductor makers, energy companies and data centers "may be affected, impacting their ability to repay their debts to commercial banks."
Where Data Centers Fit
"AI has primarily entered into bank CRE exposure through investments in data centers," the note says, CRE being commercial real estate. "Loan performances for data centers are not currently directly observable because of regulatory data limitations." According to Federal Reserve data, it says, the delinquency rate for the overall industrial property type that the majority of data centers are part of, which was 1.6% as of the third quarter of 2025, "one of the lowest among all property types."
"While there have been large data center transactions, large banks are not materially concentrated (as a percentage of tier 1 capital) in data centers." Tier 1 capital, in the note's words, is "a bank's core capital (including common stock and reserves) that often serves as a measure of its ability to absorb losses."
"Despite the strong loan performances and property fundamentals, there are multiple tail risks from lending to this property type." Those that could impact lenders such as banks, it says, include high upfront capital investment and interconnectivity among companies that create concentration risks for lenders, and potential obsolescence of the equipment and technologies built inside these properties, "resulting in loss of tenants or the need for further capital investment." They also include "exit risk due to the fact that data centers are often built in highly specialized ways and, therefore, tenant replacement may be difficult or infeasible."
Where replacing a tenant may be difficult or infeasible, the loan leans on the first lease. How long that lease runs, set against the debt, is the question that follows.
The Largest Possible Loss
A committed balance, a footnote says, is "the legally binding promise from a lender to make funds available to a borrower," and an outstanding balance is the part already borrowed. The note uses commitments, another footnote says, because borrowers "may draw on lines of credit in the event of extreme stress."
"The fundamentals of the parties receiving the funding look appropriate as of now. That said, there are risks," which "if they were to manifest could impact the standing of those borrowers and in turn the providers of their funding." From a commercial and industrial lending view, the average large bank has a committed exposure to AI-adjacent industries of 25% of tier 1 capital. That is a share of tier 1 capital, where the outstanding figure above is a share of total assets. For commercial real estate, the note says it is difficult to give a percentage of committed exposure, given data limitations, "but it is not material."
"The aforementioned committed amounts would be the maximum loss for large banks for the known direct AI exposure, but that is highly unlikely given that some recovery of losses would occur (albeit this is uncertain) and not all borrowers would default." The authors add that banks' indirect exposure to AI still has to be investigated.
A ceiling on loss is not a measure of what comes back. The note calls the full loss highly unlikely and the recovery uncertain, and gives no figure for either.
The opinions, the note's last line says, are the authors' own, and "do not necessarily reflect the views of the Federal Reserve Bank of Chicago or the Federal Reserve System."
Sources
- Federal Reserve Bank of Chicago. Tail Risk for Banks Posed by Investments in Generative Artificial Intelligence (Chicago Fed Insights, February 2026). February 2026. central bank