AI is unpriceable. Capital is deploying anyway.

The market has begun deploying capital into a category that no one can price, and the price-setter for the last five credit cycles is publicly saying so.

Howard Marks, who built his reputation pricing distressed debt through every crisis since 1978, is on record calling the AI IPO market unpriceable. His phrasing, that the decision amounts to a "thumb suck" rather than a calculation, is the admission. It is not a bubble call. Bubble calls are directional, a claim that prices are too high. Marks's claim is epistemic, a claim that the framework to call any price does not yet exist.

That gap has structural consequences. Roughly $1.7 trillion sits in private credit, the lending market underwriting much of the AI capex stack, and only about 3 percent of its managers have ever operated through a downturn. Marks's Oaktree runs inside that pool. The counter is that private rounds are clearing at real multiples, which is itself a price. The falsifier is whether those multiples survive a cycle no one has run.

When the standard discount mechanism fails, the market substitutes momentum. The repeatable mechanism is this: earnings twelve years out become unknowable, the underwriters of the underwriters have no prior downturn to test against, and pricing becomes a referendum on belief rather than a discount on cash flow. Marks's wager that no forecaster will land within 50 percent of the 2036 numbers is almost certainly correct. The exposure is real. The frame to size it is not.