If tokens are basically free, why is the frontier still rationed?

Why the standard “Jevons Paradox” explanation of AI economics, though correct, hides the more important half of the story

A simple way to sort any AI expense into “commodity tier” or “frontier tier,” and why that split matters more than the sticker price

Why labs give intelligence away for free and ration it at the same time, and why that isn’t a contradiction

Where competitive moats and investment returns will actually concentrate as the bottom of the market commoditizes

A durable mental model for correctly reading every future “AI costs are collapsing” headline

Per-token AI prices have fallen roughly 1,000x in three years, yet enterprise AI spending has roughly tripled over a similar window, the pattern most people now correctly label the Jevons Paradox.

That label is true but incomplete. It treats “AI” as one commodity on one price curve, when the market has split into two curves moving in opposite directions.

Yesterday’s frontier capability is racing toward zero cost. This year’s frontier barely deflates and is often rationed outright rather than discounted.

This bifurcation, not the aggregate price collapse, is what actually determines where value and advantage accumulate in AI markets.

Most organizations are budgeting, building, and investing as if AI has one price. It has two.