In short
Track total AI energy as well as energy per task, so growth is visible, and decide in advance what happens to the savings: retire freed capacity, keep a budget or cap, or reinvest deliberately. Pair efficiency with sufficiency by asking whether each new use is worth it. Without these, cheaper AI tends simply to be used more.
Steps
- Report total AI energy or emissions alongside energy per task in every review.
- Decide before an efficiency project what will happen to the capacity or budget it frees.
- Set a budget or cap on total compute, tokens or spend for each team or service.
- Retire servers, licences or reserved capacity that efficiency has made unnecessary.
- Ask whether each new AI use is needed before approving it.
Efficiency only becomes a saving when someone decides what happens to the capacity it frees. Otherwise cheaper AI gets used for more things, more often, and the total keeps rising.
Why does this happen?
The economist William Stanley Jevons observed in 1865 that more efficient steam engines led to more coal being burned, not less. Luccioni, Strubell and Crawford argue the same rebound dynamic applies to AI: efficiency gains lower costs, which can drive more use. The International Energy Agency projects data-centre electricity use to more than double between 2024 and 2030, with AI the most important driver, even as each request gets cheaper.
What can an organisation do?
- Make totals visible. Per-task figures alone will look like steady progress.
- Budget. Give teams a compute, token or spend budget, like any other resource.
- Bank the saving. When a model or a workflow becomes more efficient, shut down or return the capacity it no longer needs.
- Reinvest on purpose. Spending savings on new uses can be the right choice, but make it a decision rather than a drift.
- Practise sufficiency. Ask whether a task needs AI, or a model this large, at all; see decide whether you need AI.
Common mistakes
- Announcing a percentage efficiency gain as a cut in emissions.
- Leaving freed servers running "just in case".
- Approving every new AI feature because each one is cheap.
A 2026 UK green AI sandpit of mathematicians, computer scientists and industry listed avoiding the rebound effect as one of the main open questions for green AI. Read more in efficiency is not sufficiency.
Sources
- The Coal Question, W. Stanley Jevons, Macmillan (1865)
- From efficiency gains to rebound effects: the problem of Jevons' paradox in AI's polarized environmental debate, Luccioni, Strubell and Crawford, ACM FAccT (2025)
- Energy and AI, International Energy Agency (2025)
Related
Efficiency is not sufficiency: the rebound effect in AI
Why making AI more efficient does not guarantee it uses less energy overall, what the rebound or Jevons effect is, and what helps keep savings real.
How do I set targets for AI energy and emissions?
How to set AI energy and emissions targets that mean something: a measured baseline, intensity and absolute goals together, and a regular review.
How do I check an AI project is worth its environmental cost?
A simple way to weigh an AI project's own energy, carbon and hardware against the benefit it delivers, before you build it and once it runs.
How to write a green AI policy
A step-by-step guide to writing an organisation's green AI policy, with a template outline covering use, models, measurement, buying, claims and review.