Who Gets Paid When a Model Gets Smarter: A Follow-the-Invoice Map

Every dollar a customer spends on a smarter model flows through the same four hands: the chipmaker, the power company, the model lab, and the wrapper that sells the result.¹ Nvidia sits in the first seat, OpenAI or Anthropic in the third, Perplexity in the fourth. Follow the invoice through all four and one uncomfortable pattern appears: the margin does not sit where the attention does.

The wrapper: most exposed, least protected

Start at the top of the stack, where the wrappers live. They own the customer and the least of the cost, and they are the most exposed: a wrapper's entire margin is a line item on the lab's price list, and the lab can move it any morning.⁵

The lab: brand, and a brutal cost structure

The labs own the brand and a brutal cost structure.² They capture the demand and hand most of it straight to the two players beneath them.

The utility: scarce, and pricing like it

Underneath sits power, newly scarce and newly pricing like it.⁴ The utility does not care whose model wins; it bills the winner and the loser identically, and the meter only turns one way.

The chipmaker: paid first, every time

At the bottom, or the top depending on how you draw it, is the chipmaker, the one hand that gets paid before a single token is served, in advance, with a deposit.³

Draw the map and the conclusion is old news dressed in new acronyms⁶: in a gold rush, sell the shovels, rent the picks, and meter the water.