Carlyle Warns on AI Hype and Credit Risk Ahead

Carlyle Co-President Mark Jenkins warned that private-credit exposure to AI data centers and hyperscalers is becoming concentrated in a single high-growth theme. He cited parallels with the prior "SaaS apocalypse," cautioning that apparent diversification can break down during volatility and that complex data-center contracts may introduce hidden credit risk. The warning favors traditional diversification over AI-driven lending hype.

Analysis

The investable issue is not AI demand but underwriting correlation: power availability, GPU residual values, tenant credit quality, and refinancing capacity can all deteriorate simultaneously if hyperscaler capex normalizes. Private-credit structures may appear diversified across separate facilities while retaining the same economic beta to a small group of cloud tenants and a narrow set of AI workloads. That creates a delayed mark-to-market problem for alternative-asset managers: realized losses may lag public-equity weakness by 2-4 quarters, but fundraising multiples can compress immediately if investors begin to question asset-level valuation marks.

CG is more exposed to a sentiment-driven multiple de-rating than to an identifiable near-term earnings hit absent disclosure of its data-center loan book, leverage, and covenant protections. Public data-center REITs EQIX and DLR have more transparent cash flows and generally stronger tenant diversification, but remain vulnerable if new capacity commitments are deferred; their key transmission channel is development yield compression rather than direct credit losses. Equipment suppliers VRT and ETN face a different second-order risk: a credit pullback could postpone smaller developer projects even if hyperscaler spending remains robust, widening the gap between premium hyperscaler-linked orders and the broader backlog.

Consensus is likely too broad in treating all AI infrastructure financing as interchangeable. Projects backed by investment-grade hyperscalers, contracted power, and meaningful sponsor equity should remain financeable; the stress point is merchant-like capacity, aggressive utilization assumptions, and structures dependent on residual GPU values. Over the next 1-3 months, any disclosure of rising payment-in-kind income, covenant amendments, extended fund duration, or lower private-credit realizations would be a more actionable warning than generic caution around AI concentration; over 6-18 months, a reduction in hyperscaler capex guidance would test the entire financing chain.

Key Decisions for Investors

• Maintain a cautious relative-value stance: long EQIX / short CG over 3-6 months. EQIX offers contracted infrastructure cash flows, while CG has greater sensitivity to private-market fundraising and valuation-mark scrutiny; target a 10-15% relative move, with a stop if CG discloses immaterial AI/data-center exposure and EQIX reports material leasing cancellations.
• Do not initiate a directional short in CG solely on this signal. Create an alert ahead of quarterly reporting for private-credit AUM growth, PIK income, realized-loss commentary, and concentration disclosures; a deterioration in any two metrics would support a 3-6 month CG underweight or put-spread position.
• Trim high-multiple AI-infrastructure exposure in VRT and ETN only if hyperscalers signal capex deferrals or data-center developers report financing delays. Until then, avoid using private-credit headlines as a short catalyst because large, well-capitalized customers can displace weaker developers and preserve supplier demand.
• For credit portfolios, favor senior, asset-backed exposure with contracted hyperscaler counterparties over subordinated data-center development debt for the next 6-18 months. Avoid structures whose debt service depends on uncontracted utilization or assumed GPU resale values; those assumptions are the likely first source of covenant stress in a demand reset.