Banks and the AI boom: Measuring direct exposure
In this paper, Greg Cohen, Cooper Killen, and Simon Lau explore how exposed are large US banks to losses from the AI investment boom, and specifically, what is the “tail risk” if AI spending or AI company valuations were to collapse?
The authors use data primarily from FR Y-14Q Schedule H (loan-level regulatory filings from large US banks with $100bn+ in assets) and FR Y-9C for capital/asset figures, supplemented by third-party estimates from JPMorgan, MSCI Real Capital Analytics, CBRE, and S&P Global. Borrowers are sorted into “AI-adjacent” industries (software, energy, semiconductors) using NAICS codes, then exposure is measured as a share of total commitments and tier 1 capital, split further by credit rating and delinquency status. Rather than a formal quantitative stress test, the authors use qualitative tail-risk scenarios to reason through how losses could cascade across interconnected AI sectors, while explicitly excluding indirect exposure (e.g., via private credit funds) due to data limitations.
Findings
- Direct exposure is currently modest: outstanding commercial and industrial (C&I) loans to AI-adjacent industries average just 0.8% of total bank assets, with delinquency rates in line with banks’ overall portfolios.
- Committed exposure is much higher than outstanding: averaging 25% of tier 1 capital, meaning losses could spike quickly if stressed borrowers draw down credit lines before defaulting.
- Software lending is the weakest link: 26% of commitments to software companies (about $50 billion) are rated B or below, reflecting reliance on continued equity injections rather than cash flow to fund AI spending.
- Interconnection risk is significant: a pullback by software companies could cascade into semiconductor manufacturers, energy providers, and data centers simultaneously, causing correlated rather than isolated losses.
- Data centers carry added risks: high upfront capital costs, potential technological obsolescence, and difficulty finding replacement tenants given specialized property design.
- Analysis likely understates total risk: it excludes indirect exposure through lending to nonbank financial institutions like private credit funds that finance AI investments, a channel flagged as important but too hard to quantify with current data.
Tail Risk for Banks Posed by Investments in Generative Artificial Intelligence
Authors: Greg Cohen, Cooper Killen, Simon Lau
From: Federal Reserve Bank of Chicago
AI and private credit
This FSB report identifies private credit as playing a “critical role” in financing the AI infrastructure boom, particularly data center construction driven by generative AI (GenAI) demand for high-performance computing. As internal cash flows from technology companies have proven insufficient to cover the scale of capital required, asset-based finance (ABF) has become a key funding source, leveraging predictable, long-term revenue streams such as data centre leases.
- Scale of the financing gap: AI infrastructure capex is projected at $2.9 trillion between 2025 and 2028, with $1.5 trillion expected to come from external capital rather than internal company cash flow.
- Rising concentration: AI’s share of total private credit deals reached 34% in 2025, up sharply from an average of just 17% over the prior five years, indicating fast-growing sector concentration.
- Loan characteristics: Loans to AI companies broadly mirror terms seen in other sectors (similar maturities and rate spreads) but tend to be larger in scale given the capital intensity of data centre projects.
- Contained but growing exposure: The report notes that exposure to AI-related sectors currently remains relatively small for the average private credit fund, but “interest and activity in this area have been steadily growing over time”.
This sits within the report’s broader concentration-risk discussion: private credit loan originations remain focused in a handful of industries (technology, healthcare, services), and this specialization reduces diversification and increases exposure to sector-specific shocks, of which AI is now a prominent example. The report frames this as part of its wider warning that private credit “remains untested” at its current size and scope, meaning a downturn in AI-related revenue or valuations could expose leverage and credit-quality vulnerabilities that have not yet been tested through a full economic cycle.
Report on Vulnerabilities in Private Credit
From: Financial Stability Board
