Jevons paradox is the observation that making a resource more efficient to use can increase, not decrease, total consumption of it. The English economist William Stanley Jevons described it in 1865: when James Watt’s improved steam engine made coal-powered work far more efficient, England’s total coal consumption rose, because the cheaper, more useful technology was adopted far more widely. Efficiency lowered the unit cost, and demand expanded to more than fill the gap.
The idea returned to prominence on January 27, 2025, the same day a cheap Chinese model triggered a major selloff in AI chip stocks. Microsoft CEO Satya Nadella posted: “Jevons paradox strikes again! As AI gets more efficient and accessible, we will see its use skyrocket, turning it into a commodity we just can’t get enough of.” His point was a direct rebuttal to the market’s fear that cheaper AI training and inference would shrink demand for GPUs and data centers. In the Jevons view, the opposite happens: when a unit of intelligence gets cheaper, people find vastly more uses for it, and aggregate compute demand grows.
This framing became one of the central arguments of the AI-buildout bulls. It reconciles two facts that otherwise seem contradictory - that inference costs are collapsing (see LLMflation) and that compute spending keeps rising. If Jevons is right for AI, falling per-token costs are not a threat to the buildout but the very mechanism that drives it.
Why a business reader should care: whether Jevons paradox holds for AI is one of the load-bearing assumptions behind hundreds of billions of dollars of infrastructure spending. If cheaper AI really does expand total demand, the capex makes sense; if demand is more bounded than the optimists assume, the same efficiency gains could leave a lot of expensive hardware underused.