Jevons Paradox – Definition and Explanation

The Jevons Paradox is an economic theory stating that an increase in efficiency in the use of a resource does not lead to a decrease in its consumption. Instead, because the resource becomes cheaper and more effective to use, the total demand for it actually increases, often offsetting any initial savings.

Jevons Paradox Explained

The paradox follows this logic chain.

• Technological Progress: A new invention makes a resource (like coal, electricity, or data) more efficient to use.
• Price Drop: Because you need less of the resource to do the same amount of work, the “effective price” of that work falls.
• Increased Demand: Lower costs encourage people to use the resource more often or find entirely new ways to use it.
• Net Increase: The surge in new usage outweighs the savings gained from efficiency, leading to higher total consumption than before the improvement existed.

The Jevons Paradox only applies when demand is price elastic. In this example the costs fall by 40%, but demand rises 100%. Not every technological innovation will lead to the Jevons Paradox.

Importance of Elasticity

For the Jevons Paradox to happen, the demand must be price elastic. This just means that when the price drops a little, people want a bigger percentage increase in demand.

• Elastic (Jevons happens): Software, Travel, Entertainment.

• Inelastic (Jevons unlikely): Salt. (Even if salt becomes 90% cheaper, you probably won’t start eating 10x more of it).

Historical example of coal

In 1865, William Stanley Jevons published The Coal Question, documenting a counterintuitive trend: as steam engines became more efficient, Britain’s coal consumption didn’t drop—it exploded. By lowering the cost of power, efficiency encouraged people to use more. This phenomenon, now known as Jevons Paradox, suggests that when we make a resource more economical to use, we don’t save it; we find more ways to use it.

Jevons Paradox and AI

The reduction of Marginal Costs (marginal cost is cost of last unit used)

• Large Language Models (LLMs) have changed the traditional cost structure of software development. By automating the “heavy lifting” and time intensive nature of coding —AI has reduced the time-to-output for code.
• However, firms aren’t taking these gains as a reason to do the same output Instead, the elasticity of demand for software is proving to be very elastic. There was latent demand for coding.
• Increased competition. Because of the reduction in costs, existing firms now have more competition, so they can’t stand still but have to make use of the improved technology. If you stand still, you will go out of business.
• Viability of the “Uneconomic”: Projects previously not started due  high engineering costs are now going ahead.
• The Micro-SaaS Explosion: Small teams are deploying enterprise-grade automation that once required a dedicated IT department.
• Embedded Intelligence: Established firms are no longer just “using software”; they are embedding it into every internal process, from logistics to HR.

The Infrastructure Tax

The paradox is that “cheaper” code carries a hidden, higher resource cost. Every line of AI-generated code, every automated query, and every new cloud-hosted application creates a higher demand across our physical infrastructure.

It is not straightforward

• Instead of a simple line of code, we use energy-intensive model inference to “reason” through tasks.
• Volume outpaces efficiency: Even if the energy cost per AI query drops by 50%, a 500% increase in total queries results in a massive net gain in electricity demand.
• Physical sprawl: More software necessitates more silicon, more cooling, and more data centers.

We often think efficiency means “saving stuff.” We assume that if a machine is 50% more productive, we’ll use 50% less energy.

But in the real world, efficiency acts like a catalyst—it’s a spark that makes things grow much bigger. When AI makes coding “frictionless” (easy and fast), we don’t just keep at previous levels of software output. Instead, the amount of software we create increases by a bigger proportion.

• Doing “Everything with More”: Instead of using AI to do our old tasks with less electricity, we are using AI to do thousands of new tasks that we never even tried before.

• The “Footprint” Problem: Even if one AI query uses very little power, we are now doing billions of queries a day. The total “footprint” (the land, energy, and water needed for data centers) keeps getting bigger.

• A Change in Scale: Think of it like a dam. If you poke a tiny hole, a little water comes out. If you take the dam away (make it “frictionless”), the entire lake rushes through. AI is taking away the “dam” that used to slow down how fast we could write code.

Other examples of Jevons

The “Paperless Office” (Printing Efficiency)

In the 1980s, people predicted that computers and email would lead to the “paperless office.” Since sending a digital file is 100% more efficient than mailing a physical letter, we thought paper use would fall.

• The Paradox: Because computers made it so easy to write, edit, and format documents, people started producing more documents than ever before.

• The Result: Instead of saving paper, global paper consumption tripled between 1980 and 2000. It became so “cheap” to print a draft that people printed every single version of a document just to proofread it.

About The Author

Tejvan Pettinger

Tejvan Pettinger. Author of “Cracking Economics” “What Would Keynes Do?” – studied PPE Oxford University (1995-99) and works as an economics teacher and writer.

See author's posts