Who Makes Money in AI: A Field Guide for Founders, Operators, and Investors

Nine trillion dollars went to the companies that make the chips. Two hundred and forty-six billion went to the ones everybody said would win.

Who Makes Money in AI is written for founders picking a layer of the AI stack, operators defending one and investors pricing it. Between the end of 2022 and September 2026 the money moved upstream and stayed there, and this book traces where it landed.

Twelve positions carry the AI stack. Each one sets four numbers before anyone at the company touches them: gross margin, the capital a dollar of revenue requires, when the cash arrives, and how fast the asset has to be rebuilt. A system of record keeps fifty-two cents of a revenue dollar, and forty-nine after the capital charge, on twenty cents of capital. A model owner keeps sixty, and minus a hundred and forty after the same charge. Execution moves a company inside its position, and changing position takes years.

The mispricings

• The application layer is priced with too little dispersion. A real system of record and a wrapper differ by more than fifty times on the compounding test, and the venture market prices them inside a factor of two.
• Power and land look cheap against the constraint they gate. The layer added $323 billion while gating a position that added $8.7 trillion, and capital cannot clear an interconnection queue.
• Compute landlords look expensive against their credit quality. They keep about a fifth of what they bill, against counterparties funded by the same market that funds them.

The book closes on one rule for anyone writing a check: underwrite the position before the company, because being right about the team and wrong about the layer is the most expensive mistake in this market.

Format

Paperback by Nitin Kumar, 343 pages, 6 by 9 inches. Printed to order and shipped from the US, UK or Canada.