The same smart money that learned "diversified SaaS exposure" was a mirage is now making an even bigger bet on AI infrastructure—and pretending concentration risk doesn't count if the spreadsheet says "data center" instead of "enterprise software."
The Summary
- Carlyle's Co-President Mark Jenkins warns private credit firms financing AI buildouts face the same concentration trap that hit SaaS lenders during the software downturn
- The rush to fund data centers and hyperscalers creates portfolios that look diversified on paper but collapse together when the theme cracks
- Complex data-center financing contracts can fail in ways lenders haven't stress-tested, per Jenkins
- The parallel: everything looked uncorrelated in SaaS portfolios until revenue multiples compressed and every "different" software company suddenly had the same problem
The Signal
Private credit has a new favorite customer, and Carlyle is flashing yellow. Firms are lining up to finance the AI infrastructure boom—data centers, GPU clusters, hyperscaler expansion—with the same enthusiasm they once brought to SaaS lending. Mark Jenkins, Carlyle's Co-President, sees the pattern repeating: portfolios that look diversified by name, geography, and contract structure, but are actually one correlated bet on a single macro theme.
The SaaS parallel is precise. In 2021, private credit portfolios held dozens of "different" software companies across verticals, customer segments, and revenue models. On paper, uncorrelated. In practice, they all lived or died by the same thing: investor appetite for recurring revenue at any multiple. When rates rose and growth-at-all-costs died, the entire book moved as one. Jenkins warns the AI buildout has the same structure, just bigger checks and longer payback periods.
"Concentration in a single hot theme can mirror the SaaS apocalypse—everything looked diversified until volatility hit."
The specific risk in AI credit is structural complexity meeting demand uncertainty. Data-center financing involves:
- Multi-year capacity commitments from hyperscalers who may reprice AI spend if model economics shift
- Power purchase agreements tied to grid reliability and regulatory approval timelines
- Technology obsolescence risk as chip generations turn over faster than lease terms
Any one of those variables can break a deal. If three break at once because the AI investment cycle stalls, lenders holding "diversified" portfolios of data-center debt will learn they own the same exposure with different counterparty names on the docs.
The deeper issue is that private credit grew up in a world where diversification meant industry and geography. Jenkins is saying AI infrastructure breaks that logic. A data center in Virginia and a GPU co-location facility in Texas aren't uncorrelated bets. They're two ways to be long the same trade: that AI capex keeps compounding at 40% annually and utilization stays high enough to service debt. If that assumption cracks, the whole book reprice at once, regardless of how the risk committee sliced the portfolio by sector code.
The Implication
If you're allocating to private credit funds, ask what percentage of the book has AI infrastructure exposure under any definition—data centers, power contracts, fiber buildouts, chip financing, hyperscaler mezzanine debt. If the answer is north of 20% and the pitch is "but they're all different deals," you're holding concentrated thematic risk with a diversification label.
For operators building in the agent economy, this is a financing window that won't stay open forever. The smart play is to lock long-term capacity agreements and debt while lenders are still underwriting on hype. When the SaaS correction hit, the companies with term sheets signed in 2021 survived. The ones pitching in 2022 got ghosted. Same pattern, faster cycle.