DoubleLine, Oaktree buy debt ahead of potential AI credit risks

DoubleLine and Oaktree are becoming more selective in AI-related credit as Bloomberg Intelligence estimates about $5 trillion of AI capex over the next five years, much of it debt-funded. Portfolio managers warned that data center financing faces frothy valuations, long-dated obsolescence risk, and overbuilding risk as hyperscalers have already issued more than $155 billion of unsecured bonds globally, up over 45% from last year. The comments are a cautionary signal for AI debt markets rather than a direct company-specific catalyst.

Analysis

The message from the credit market is not just that AI capex is large; it is that duration mismatch is becoming the central risk. Debt financing for assets with 10-20 year useful lives against a technology stack that can commoditize in 3-5 years creates a classic left-tail refinancing problem: even if near-term cash flows look fine, the second-order loss is on the takeout market and terminal asset values. That tends to show up first in unsecured paper and sponsor financing, then in the equity of firms with the weakest land/power/tenant concentration and the longest build cycles.

The likely winners are not the obvious semiconductor leaders, but the toll collectors around them: lenders with disciplined underwriting, contractors with contracted backlog, and balance-sheet-heavy operators that can self-fund buildouts. The losers are late-cycle data center developers and capital-light promoters relying on perpetual lease-up assumptions, because oversupply usually appears after rate of change in demand peaks, not when headlines are most bullish. A key second-order effect is that power, transmission, and cooling supply chains may remain tight even if financing conditions worsen, which means the market could punish speculative infrastructure while continuing to bid up constrained utility-adjacent names.

The contrarian point is that this is not an immediate bubble-top signal for the entire AI complex; it is an early warning that financing terms will bifurcate hard. In the next 3-12 months, you can still see equity capex optimism coexist with a credit repricing, especially if public markets keep subsidizing growth while private lenders tighten covenants. The real catalyst for a broader de-rating would be evidence of slower lease absorption, longer payback periods, or a few high-profile projects missing power/interconnect milestones, which would expose the fragility of the growth narrative.

Key Decisions for Investors

• Underweight speculative data-center developers and levered infrastructure equities; favor balance-sheet-heavy operators with contracted cash flow over the next 6-12 months, as financing dispersion should widen materially if credit spreads continue to back up.
• Go long cash-rich hyperscaler suppliers with pricing power and short-duration revenue visibility; avoid names dependent on multi-year project financing, which face the highest mark-to-market risk if underwriting standards tighten.
• Pair trade: long select utilities / power infrastructure beneficiaries versus short high-leverage data center growth names, targeting a 3-6 month window where power scarcity remains supportive but financing conditions deteriorate.
• Buy downside protection on the most leveraged AI-infrastructure proxies via put spreads 6-12 months out; the risk/reward is favorable because the market is underpricing refinancing and terminal-value risk rather than near-term revenue risk.
• If credit spreads on AI-linked issuance widen another 50-75 bps, add to shorts in capital-light promoters and reduce exposure to long-duration 'AI picks and shovels' names that rely on perpetual access to cheap debt.