Concentrated AI Credit Has to Pay for Yield and Safety

What's New

Concentrated lending, which the AI buildout now dominates, requires investors to demand both higher compensation and added safety, since investment grade credit offers no upside to absorb a loss. Marc Rowan, CEO at Apollo, argues this in an interview on The Inside Brief from The National. Hyperscaler borrowing is moving the top of US investment grade issuance from the 5 global banks to the 5 largest cloud companies. Credit investors should lend secured, against chips or energy, with an investment grade counterparty on the offtake.

Why It Matters

The claim tells allocators to treat AI credit as project finance with collateral, and to discount unsecured paper even when the issuer is investment grade. Neoclouds are on the other side. They carry equity valuations above $40 billion. Their debt pays mid-teens returns, and Rowan says one of the two markets is wrong. Apollo's origination model and its new stake in Mubadala Capital both depend on demand for complex, long-dated private credit, so the thesis serves its book.

By The Numbers

• 10 stocks make up half of the S&P 500, all levered to the same trend. Equity can absorb that concentration; credit cannot.

• Half of the roughly $1 trillion Apollo manages is its own money, which Rowan cites as the reason it only sells what it would own.

• 2.5 years: the average life of a chip financing. Terms run 5 years. Repayment is 20% a year, so supply can be underwritten against fabs already under construction.

• $1 trillion: what he expects the Gulf to spend hardening infrastructure and building alternative routes after the Iran conflict.

Memorable Quotes

• "We do not sell anything to a client that we are not prepared to own ourselves." The principal lens is his basis for refusing unpaid concentration.

• "If someone wants you to be concentrated, they have to give you two things, not one thing." Compensation alone cannot repair a loss in investment grade credit.

• "Whether this can continue to be done at the scale we've seen, at the current pace we've seen, I don't know." The secured hyperscaler trade has no precedent at this volume.

• "When you have your equity hat on, make sure you have a seat belt." He expects valuations to correct even if the technology delivers.

The Wrap

The secured approach works as long as compute stays scarce, energy demand rises with cheaper compute and hyperscalers remain investment grade regardless of AI returns. It fails if chip supply catches demand within the 2.5-year life of those loans and collateral loses value. It also fails if regulation after an AI incident stalls data center buildout and strands powered assets. Dispersion in below investment grade software credit over the next 2 years will show whether the bear case has arrived.