The AI Bet in Credit Is Bigger Than It Looks

What's New

Credit investors hold more AI risk than their concentration limits show, according to KKR Credit & Markets. Christopher Sheldon and Tal Reback lay out the case in Let There Be Compute!, the firm's latest quarterly credit review. Measured by direct bonds, AI-related debt is about 6.3% of the ICE US IG index. Add data center bonds backed by hyperscaler lease guarantees, and technology becomes the largest sector in JPMorgan's JULI index. Its weight is 11.6%, ahead of US banks. Concentration limits should count guaranteed lease debt alongside direct bonds.

Why It Matters

Spreading capital across IG, high yield, asset-based finance, and infrastructure looks diversified on paper. KKR argues the same short list of counterparties stands behind all of them: power, chips, land, leases, and financing route back to a handful of companies. Passive and index-oriented holders are most exposed, since they may own the theme without ever deciding to. Historical correlations offer little comfort because this capex cycle has no history to test against.

By The Numbers

• $1.1 trillion in hyperscaler leases has not yet commenced and does not show up as reported debt.

• $7.6 to 8 trillion is the projected AI infrastructure capex through 2030. Public IG capacity for the largest hyperscalers is about $1.7 trillion.

• 160 basis points was the relative move of JPMorgan's US high yield AI basket against the broad index. It took 13 weeks, and the broad index barely changed.

• 82% of the high yield exposure in the AI bond complex is secured project finance.

Zoom In

KKR sets out how to measure and manage the exposure.

• Hard limits on lease-backed debt. KKR recommends setting concentration limits that include data center bonds backed by hyperscaler lease guarantees. Under stress, a bond with a Meta or Alphabet guarantee behaves more like corporate risk than project finance.

• Monitoring for off-balance-sheet commitments. Leases not yet commenced and other commitments belong in portfolio construction reviews as a tracking tool.

• Higher credit quality than past buildouts. 13 of 31 issuers in the AI bond complex are investment grade. Those issuers carry about 90% of the outstanding balance.

• Tags by dependency. KKR sorts credits into AI-dependent names such as hyperscalers and data centers, AI-sensitive names such as semiconductors and software, and beneficiaries such as utilities and construction.

The Wrap

Allocators should measure AI exposure by what drives repayment. Obligor limits are enough while hyperscaler free cash flow keeps pace with capex. They fall short if growth slows and spending has to stretch over a longer horizon. In that case a repricing could move across equity, credit, and infrastructure at once, as software already did on a smaller base. KKR expects the next 2 to 3 years to start separating winners from losers.