Hyperscaler debt signals warning sign, Apollo cautions

Apollo warned that hyperscaler credit risk is rising as AI capital expenditures are increasingly debt-financed, with the CDS spread versus bank CDS widening to about 60bps from roughly zero in October 2025. The firm cited rising leverage, negative free cash flow and uncertain returns on rapidly depreciating AI assets; Alphabet, Amazon and Meta have forward free cash flow of negative $25.7B, negative $30B and negative $25.7B, respectively. Microsoft remains an outlier with $33.4B in positive forward free cash flow, while technology investors argue improving margins and capacity additions extending into 2027-28 make it premature to draw adverse conclusions.

Analysis

The relevant equity transmission is not default risk but a higher cost of capital imposed on an industry whose valuation assumes long-duration AI returns. A sustained 50-75bp widening in hyperscaler credit spreads would raise incremental funding costs modestly, but more importantly forces boards to prioritize capex ROI, slowing the willingness to pre-build capacity. That is negative first for the lowest-visibility AI infrastructure commitments—NVDA, VRT, ETN and data-center REITs—before it becomes material to cloud revenue growth.

MSFT should be relatively insulated in a risk-off rotation because enterprise software and Azure monetization provide a clearer path from AI spend to recurring revenue, while AMZN, META and GOOG have greater exposure to investment cycles where the revenue realization is less directly matched to the infrastructure outlay. The near-term catalyst is not quarterly capex itself—markets expect that—but evidence of weaker incremental cloud/advertising monetization per dollar of AI depreciation in the next two earnings cycles. A regulatory safe harbor for model developers could reduce litigation-tail risk, but it would not solve excess-capacity or asset-obsolescence risk.

Consensus may be too focused on aggregate leverage ratios. The more consequential risk is duration mismatch: servers and accelerators depreciate quickly while customer adoption, pricing power, and regulatory clarity develop more slowly. Conversely, this is not yet a broad technology-credit event; if operating margins continue expanding and cloud backlog converts into paid consumption through 2027, spread widening can reverse without meaningful equity downside. The thesis is falsified by stable-to-tightening hyperscaler CDS alongside accelerating cloud growth and rising FCF conversion over the next two reporting periods.

Key Decisions for Investors

• Initiate a 3-6 month quality pair: long MSFT / short equal-dollar basket of AMZN, META and GOOG. Target 8-12% relative return if credit differentiation persists; exit if the hyperscaler CDS-to-bank CDS gap compresses below 20bp or if AMZN/META/GOOG show two consecutive quarters of improving FCF conversion.
• Reduce tactical exposure to AI power-and-cooling beneficiaries VRT and ETN, and avoid chasing NVDA on infrastructure-demand headlines. These names face the highest second-order sensitivity to any 2027-28 capacity deferral; revisit after hyperscaler capex guidance and backlog disclosures confirm deployment schedules.
• Use a watch trigger rather than a directional credit trade: if the hyperscaler CDS premium widens another 25bp while cloud growth decelerates, buy 6-month QQQ puts or establish a QQQ/SPY downside hedge. The combination would signal that the issue is moving from isolated credit pricing into technology multiple compression.
• Maintain or add selective exposure to APO on credit-market volatility only if private-credit origination and fee-related earnings remain intact. Wider spreads can improve deployment economics, but a genuine hyperscaler funding retrenchment would reduce a potentially large financing opportunity; treat it as a volatility beneficiary, not a pure AI-credit long.