AI Borrowing Scale Pays Lenders Extra Spread Without Extra Risk
What's New
AI debt pays lenders a premium over same-rated industrial debt because of the size of the borrowing, while the borrowers stay strong credits. Marc Rowan, CEO at Apollo, argues this in a conversation with CNBC's Sara Eisen at TomorrowX Summit. Lenders earn only interest and principal, so they resist concentration. An issuer that wants concentrated lenders must offer both more return and more safety. Rowan says that is happening now. Allocators who want AI exposure should compare secured, investment-grade AI credit with the equity.
Why It Matters
Rowan's view challenges the idea that lenders carry the most fragile part of the AI trade. He points out that most of the spending comes from a small group of companies with large, successful businesses outside AI. If AI revenue falls short, he expects low returns and stranded capex, with the borrowers still able to pay. Apollo needs investment-grade assets with yield to back its retirement products, so the thesis fits its own business model.
By The Numbers
6 of 10: the top bond issuers Rowan expects to be tech companies. He puts the shift 3 to 4 years out. Today the 10 largest banks lead issuance.
15%: the yield on bonds of some neocloud companies. Those companies have market caps of $50 to 60 billion.
2.5 years: Rowan's effective horizon on loans secured by the latest chips. The chips depreciate to zero over 5 years.
$1 trillion: roughly the size of Apollo's business. Half of it is Apollo's own money through insurance.
Memorable Quotes
"If an issuer wants you to be concentrated, they have to give you not just more return, but more safety." Rowan treats lender concentration limits as a source of pricing power.
"A company that size, that scale can offer you return, but it can't offer you safety." He separates the neoclouds from the largest borrowers and keeps the thesis to the strongest credits.
"There will be low returns. There will be stranded capex." Rowan concedes that much of the AI spending may not pay off, and he puts that loss on equity holders.
"The spread over treasuries for AI companies is multiples of the spread over treasuries of the same rating for industrial companies." This is the claim the thesis rests on, and the bond market can test it directly.
The Wrap
The premium is real as long as AI borrowers keep large profitable businesses outside AI and lenders hold secured claims on hard assets. It erodes in two ways. Weaker borrowers like the neoclouds could take a larger share of issuance, or chip collateral could lose value faster than 5 years. Spreads could also converge as investors absorb the supply. The bond market's issuer mix over the next 3 to 4 years will settle it.



Comments