Jevons Paradox

In the revised 1866 edition of The Coal Question, William Stanley Jevons wrote: “It is wholly a confusion of ideas to suppose that the economical use of fuel is equivalent to diminished consumption. The very contrary is the truth.”¹

Jevons observed that James Watt’s steam engine was vastly more efficient than Thomas Newcomen’s earlier design. Less coal per unit of work. Intuition says: less coal consumed overall.

British coal consumption rose from about 93 million tons in 1865 to about 225 million in 1900, roughly 2.4 times.² The trend fits Jevons’s warning. It does not show that steam-engine efficiency caused the whole increase; industrial growth, prices, trade, and new uses also mattered.

The mechanism: efficiency lowers the cost of using a resource per unit. Lower cost increases demand. If demand increases faster than efficiency gains, total consumption rises.

Jevons argued that Watt’s engine made steam power economical in more settings.¹ That is one route to rebound: lower cost per unit can increase use at the intensive margin, the extensive margin, or both.

This is the rebound effect. A rebound below 100% gives back part of the expected saving. Rebound above 100%, often called backfire or the Jevons paradox, means total resource use rises after the efficiency gain.³

Modern examples:

Outdoor lighting: Satellite observations found that Earth’s artificially lit outdoor area grew 2.2% a year and its measured radiance grew 1.8% a year from 2012 through 2016.⁴ The authors said the result did not fit a global fall in lighting energy use. The study did not isolate LEDs as the sole cause.

Air conditioning: Global cooling demand rises with heat, income, urban growth, access, building design, and equipment choice. Better efficiency restrains demand, but it does not erase those other forces.⁵

Computing: Avoid comparing one AI training run with “a small city” unless the run, hardware, place, and time are named. The International Energy Agency estimates that data centers used about 415 terawatt-hours in 2024, around 1.5% of global electricity use.⁶ That aggregate includes more than AI.

Efficiency can lower cost, emissions, or resource use per service. Whether it lowers the total depends on prices, demand, the system boundary, and the time span.

If the goal is a fixed total, policy may also use prices, caps, standards, or allocation. The paradox does not choose among those tools. It warns against treating an engineering ratio as a forecast of total demand.

Go Deeper

Books

The Coal Question by William Stanley Jevons (1865) — The original. Available free online. Surprisingly readable.
The Efficiency Paradox by Edward Tenner — Modern treatment of how efficiency gains create unexpected consequences.
The Jevons Paradox and the Myth of Resource Efficiency Improvements edited by John M. Polimeni et al. — Academic collection on rebound effects.

Essays

Amory Lovins’s responses to Jevons — Lovins argues that with proper policy, efficiency can reduce total consumption. The debate continues.
Steve Sorrell and colleagues’ review — Direct rebound for household energy services in OECD countries is often below 30%; wider economy effects remain harder to estimate.⁷

Sources

William Stanley Jevons, The Coal Question, “Of the Economy of Fuel”, 2nd ed. (1866), 122–123. ↩ ↩²
UK Department for Energy Security and Net Zero, “Historical coal data: coal production, availability and consumption, 1853 to 2024”. ↩
Paul E. Brockway and colleagues, “Energy efficiency and economy-wide rebound effects: A review of the evidence and its implications”, Renewable and Sustainable Energy Reviews 141 (2021), 110781. ↩
Christopher C. M. Kyba and colleagues, “Artificially lit surface of Earth at night increasing in radiance and extent”, Science Advances 3, no. 11 (2017), e1701528. ↩
International Energy Agency, The Future of Cooling, 2018. ↩
International Energy Agency, “Energy demand from AI”, 2025. ↩
Steve Sorrell, John Dimitropoulos, and Matt Sommerville, “Empirical estimates of the direct rebound effect: A review”, Energy Policy 37, no. 4 (2009), 1356–1371. ↩