S&P: hyperscalers' AI obligations may be bigger than investors think

what happened

In a CNBC interview, Mark Habib of S&P Global Ratings, who leads the digital infrastructure sector, argued that the largest cloud operators may be holding more AI-linked obligations than the market sees. Data center leases and chip orders are not counted as debt until they begin operating, so these risks stay hidden.

S&P plans to bring such commitments onto balance sheets so investors can see what companies are really signing up for, and so the agency can factor them into its ratings. Habib said firms often use these structures to manage cash flow: rather than borrowing, they lease capacity or place orders while someone else funds the build. The effect is already visible. Oracle was cut to BBB- after breaching a rating trigger, one notch above speculative grade, while other hyperscalers remain within limits but with less headroom than a year or two ago.

Habib also pointed to Anthropic's recent S-1 filing and earlier leaked documents that referred to commitments above $500 billion. He questioned whether its revenue growth would prove durable enough to justify such spending.

why it matters

For credit markets, the key question is whether spreads and ratings for AI-heavy borrowers reprice once these obligations are added to balance sheets. Bondholders and equity investors may reassess leverage across the sector, which could weigh on valuations.

market impact

Hyperscaler bonds and AI stocks: hidden debt could widen credit spreads and pressure sector valuations

tags

instruments