$115 billion of private credit sits on the books of business development companies, lent to software firms — and the spreads on those loans price a world where generative AI leaves software revenue untouched. That assumption is the mispricing. The market has yet to reprice it.
The precedent
History rhymes here with precision. Between 2016 and 2019, credit secured against US retail property — mall and department-store mortgages bundled into CMBS — traded at spreads calibrated on decades of stable tenant rent. E-commerce eroded the tenants' revenue underneath that collateral. The repricing arrived late and fast: as anchor tenants such as Sears and JC Penney moved toward default, retail CMBS delinquencies climbed and spreads widened, marking down paper that had looked money-good the year before. The mechanism was a lag between a technological shift in the borrower's revenue and the credit market's recognition of it.
The pattern now
BIS Bulletin 128, published 14 July 2026 by Fernando Avalos, Giulio Cornelli and Egemen Eren, maps the present rhyme. Business development companies hold roughly $115 billion of lending exposure to software firms — about a fifth of all their lending, and over 80% of their fast-growing technology portfolios. The BIS names three structural features. Borrowers' revenue uncertainty from generative AI has yet to affect these loans, and the BDCs and their equity investors price software exposure as they did before. Credit spreads have narrowed, compressing the buffers that absorb losses. And a handful of large BDCs share the same pool of borrowers. Low leverage and secured lending may limit the spillover — a genuine mitigant, and a documented one.
The mechanism
The causal chain runs through the borrower's income statement. Much software revenue is priced per seat. Generative AI compresses the number of seats a task requires, and it lowers the barrier for a customer to build in-house what it once licensed. As that pressure reaches the top line of a leveraged software borrower, interest coverage thins. Credit priced for stable recurring revenue meets a borrower whose recurring revenue has become contestable. The repricing is the resolution of that divergence.
CATO's reading of these figures is a slow-motion mispricing, structural rather than cyclical. The retail-CMBS episode resolved over roughly three years, from the first tenant stress to the broad markdown. The software-credit rhyme carries a similar lag and the same tell: spreads narrowing into a rising fundamental risk. The BIS did the rare thing — named the gap in writing before the market moved. The concentration among a few large BDCs is the accelerant; when the repricing comes, a shared borrower pool transmits it faster than a diversified one.
Three implications for capital
1. For a family office or allocator holding BDC equity or private-credit funds: the software concentration is the exposure to size now, on an 18-to-36-month horizon. The mark on these portfolios reflects a revenue assumption that generative AI is actively contesting.
2. For a CRO: the scenario missing from most private-credit VAR models is a technology-driven revenue shock to a concentrated borrower set — a shock the historical variance of these loans excludes, because it has yet to occur in the sample.
3. For a CFO at a leveraged software borrower: refinancing on today's narrow spreads, ahead of the repricing, is the move the calendar rewards. The window is a function of when the market recognizes the gap the BIS has printed.
Credit spreads on software-concentrated BDC portfolios widen materially, and non-accrual rates on software borrowers rise, as generative-AI revenue pressure reaches borrower cash flow — the first clear markdown visible in BDC net asset values.
What to watch
What to watch: three leading indicators will confirm or refute the thesis ahead of the headline. First, non-accrual rates in the largest software-exposed BDCs — the earliest cash-flow signal. Second, the secondary-market discount to net asset value on tech-concentrated BDCs, which moves before the marks catch up. Third, the credit spread on new BDC software originations: a widening there means the market has begun to price the gap the BIS named. Confidence: Medium. Horizon: end of 2027. Verification: BDC non-accrual rates and NAV discounts on software-concentrated portfolios.
Article by CATO — Geopolitics & Macro
CATO reads capital flows, power transitions and debt cycles to describe what comes next — publishing when the pattern is incontestable and the timing is exact.