Data Center Shells Beat GPUs on Risk-Adjusted Returns

What's New

Private lenders chasing GPU financing are taking technology risk the extra yield does not cover. Mike Arougheti, CEO at Ares Management, argues this in a fireside chat at the Barclays 24th Annual Global Financial Services Conference. He ranks AI opportunities by risk-adjusted return: data center shells first, adjacent infrastructure second, chips last. He says no one can yet describe the depreciation curve for GPUs. Lenders should size AI credit to what they can underwrite and treat the largest pool of demand as the riskiest.

Why It Matters

Most of the AI capital need is in chips, so the market pays a premium to close that gap. Arougheti's view runs against much of the industry's push to make compute a financeable asset class. It also challenges the idea that the biggest addressable market is the best place to deploy. Ares targets 16% to 20% annual growth in fee-related earnings, and his framing lets it reach that goal from shells and infrastructure alone.

By The Numbers

• $5 trillion: the capital Arougheti says the AI build-out needs by the end of the decade.

• $800 billion: the share for data center shells alone.

• $3 trillion to $3.5 trillion: the share for chips, the largest slice.

• 100 to 200 basis points: the extra return the market offers for GPU financing.

The Other Side

The standard industry view is that scale follows the largest need. Chips are the biggest slice of AI capex. Lenders who structure around investment-grade offtakers argue they can take the premium with limited loss. Lenders who stay out may give up share in one of the fastest-growing parts of private credit.

Memorable Quotes

• "The capital requirements in the AI space right now are mind-blowingly large." Arougheti accepts the demand case in full.

• "No one could really articulate, at least to me, what the depreciation curve looks like for that technology." His objection to GPUs rests on this one unknown.

• "One of the reasons why we focused on sub-investment grade ABF versus investment grade is the fee rates are eight times higher." The admission shows how fee economics shape where Ares chooses to play.

• "You have to be leading with your risk appetite and not your appetite to deploy." This is the discipline behind the ranking.

The Wrap

The thesis holds if GPU values fall faster than lenders assumed, or if offtakers push back on terms as chips age. Lenders who took the premium win if chips keep their value long enough to repay on schedule. Ares also needs a steady flow of shell and infrastructure deals to meet its growth targets without chips. The first GPU-backed loans to season before 2030 will settle the argument.