What Is Jevons’ Paradox?

When making something more efficient can lead people to use more of it overall.

The short answer

Jevons’ Paradox is the possibility that an efficiency improvement increases total consumption instead of reducing it.

The basic pattern is simple:

1. A resource becomes cheaper, faster, or easier to use.
2. People find more reasons to use it.
3. The increase in use partly or completely cancels the original savings.
4. If the extra use is large enough, total consumption rises.

That last step is the strict version of the paradox. A smaller increase is usually called a rebound effect.

The important phrase is “if the extra use is large enough.” Efficiency does not automatically produce higher consumption. It creates a possibility that depends on prices, demand, available alternatives, and the limits of the system.

Where the idea came from

The idea is named after William Stanley Jevons, a nineteenth-century British economist. In The Coal Question, first published in 1865, Jevons examined Britain’s growing dependence on coal.

His argument was not simply that a more efficient steam engine used less coal for each unit of work. That part was true. His point was that the improvement could make coal-powered activity cheaper and more attractive. Businesses could use more efficient engines in more places, expand production, and create new demand for coal.

In other words, using less coal per unit of output did not guarantee using less coal in total.

That was the surprising part. Efficiency was not the same thing as conservation.

A small example

Imagine that 100 tasks each require two units of a resource:

100 tasks × 2 units = 200 total units

Now an improvement cuts the resource needed per task in half:

100 tasks × 1 unit = 100 total units

If the number of tasks stays at 100, total use falls. The efficiency improvement worked as expected.

But suppose the lower cost makes many more tasks worthwhile:

300 tasks × 1 unit = 300 total units

Each task is still twice as efficient as before. Total use is now 50 percent higher than it was at the start.

That is the mechanism behind Jevons’ Paradox: the savings per task can be overwhelmed by the number of tasks.

Rebound is not always the full paradox

It helps to separate three outcomes.

1. No rebound

The technology becomes more efficient, but people do not change how much they use it. Total use falls by roughly the amount saved per unit.

2. Partial rebound

People use more because the resource is cheaper or more useful, but not enough to erase all the savings. Total use still falls, just by less than expected.

3. A strict Jevons rebound

The increase in use is greater than the efficiency savings. Total consumption rises above its starting point.

People often use “Jevons’ Paradox” for the whole family of effects. Strictly speaking, the paradox is the third case. When discussing a real example, it is worth saying which case the evidence actually shows.

Why does the rebound happen?

There are several possible reasons.

Lower prices. If a task becomes cheaper, people may buy or perform more of it.

New uses. A resource that was previously too expensive may become practical for new purposes.

Expanded production. Businesses may use the savings to produce more goods or services rather than simply spend less.

Latent demand. People may have wanted the output all along but could not afford it, could not find the necessary expertise, or could not wait long enough for it.

A shift in the bottleneck. Once one input becomes cheaper, another input may become the limiting factor. The system can continue expanding until it meets a constraint such as electricity, materials, skilled review, regulation, or customer demand.

The opposite can also happen. Demand may reach a natural limit. People may have enough hot water, enough household lighting, enough reports, or enough of some other specialised output. In that case, efficiency is more likely to reduce total use.

What does this have to do with AI?

The concept applies to AI when someone claims that AI has become more efficient.

But “more efficient” is incomplete unless we know the unit being measured. It might mean:

• less energy per model response;
• less compute per image or code task;
• less money per customer interaction;
• less labour time per report;
• fewer emissions per dollar of company output.

Those are per-unit measures. They do not tell us what happened to total use.

A lower cost per AI response could lead to more responses. A cheaper coding tool could make it worthwhile to build software that was previously too expensive. A company might produce more reports, images, campaigns, or automated interactions because each one costs less.

That is a plausible mechanism, not proof that AI-wide rebound has occurred.

A 2025 analysis by Alexandra Sasha Luccioni, Emma Strubell, and Kate Crawford argues that AI efficiency should be studied alongside wider economic and social effects, rather than assuming that lower direct energy use automatically means lower environmental harm. Their peer-reviewed paper is also available as an open manuscript. It is an analysis of the problem, not an economy-wide measurement of AI’s rebound rate.

A study by Dongyang Yu and Bingjie Xu examined Chinese A-share listed companies from 2010 to 2022. The authors reported an association between AI adoption, lower carbon emissions relative to economic output, and higher total carbon emissions in that study population. You can read the paper through its DOI. The finding is important but bounded: it is not proof that every country, industry, company, or AI system will behave the same way.

Shane Greenstein has also applied the framework directly to AI in Artificial Intelligence and the Jevons Paradox. For a more accessible explanation, Hank Green’s video “Unfortunately, You Need to Know What the Jevons Paradox is” walks through efficiency, latent demand, software, and shifting constraints. The video is useful commentary and explanation, not empirical proof.

What Jevons’ Paradox does not tell us

Jevons’ Paradox does not prove that:

• every efficiency improvement increases total use;
• AI will inevitably increase energy consumption;
• more software output will create more software jobs;
• productivity gains will be shared fairly;
• environmental rules or capacity limits cannot change the result.

It also does not answer who benefits from the additional activity. A company may produce more with fewer workers. Consumers may receive lower prices. Owners may capture more of the gain. Review, security, and governance work may expand. These are separate questions.

The practical test

When someone says that AI, a machine, or a process has become more efficient, ask:

1. Efficient per what unit?
2. Did the amount of use change?
3. What happened to total energy, spending, labour, materials, or emissions?

Those three questions prevent a common mistake: treating a lower cost per task as proof of a lower total cost.

For the AI application of this idea, read More Efficient AI Does Not Guarantee Lower Total Use.

Bottom line

Jevons’ Paradox is not a claim that efficiency is bad. It is a warning that efficiency changes the economics of use.

Sometimes the result is lower total consumption. Sometimes the savings are partly cancelled. Sometimes demand expands enough that total use rises.

The only reliable way to know which outcome occurred is to measure both sides of the equation: how much resource each unit requires, and how many units people use afterward.

Research cutoff: 2026-07-27. This explainer describes the mechanism and the limits of applying it to AI; it does not provide an economy-wide estimate of AI rebound.