# The Jevons paradox: AI got 33× more efficient, so why will its power use double? > Google's Gemini uses 33× less energy per prompt than a year before, yet data centres' electricity is set to more than double by 2030. A coal book from 1865 explains the pattern. A 53-second explainer, with the numbers on both sides. - URL: https://oliverz.dev/videos/jevons-paradox - Author: Oliver Zhang (https://oliverz.dev) - Language: en - Published: 2026-10-31 - Tags: ai, energy, economics, explainer - Video: https://www.youtube.com/shorts/7BQIKw5YqC0 > **In short:** per prompt, AI really is getting far more efficient: Google says the median Gemini text prompt used 33× less energy over 12 months. Yet data centres' electricity is set to more than double by 2030, and the IEA names AI as the main driver. That pattern has a name and a 160-year-old proof. In the 1860s, Scotland's iron furnaces cut their coal per ton of iron to under a third, and Scotland burned ten times the coal. W. S. Jevons wrote it down in 1865; today it's called the Jevons paradox. "More efficient" doesn't settle the energy question: how much we use does. ## 33× less energy per prompt In August 2025, Google published how much energy its Gemini assistant uses to answer. The headline: "over a recent 12 month period, the energy and total carbon footprint of the median Gemini Apps text prompt dropped by 33x and 44x, respectively, all while delivering higher quality responses." In absolute terms, Google estimates "the median Gemini Apps text prompt uses 0.24 watt-hours (Wh) of energy". In its own comparison: "The per-prompt energy impact is equivalent to watching TV for less than nine seconds." That number is about one company's model, Gemini, and its median text prompt. It isn't "AI" as a whole. ## Yet the total is set to double Now the other number. The International Energy Agency (IEA) says: "Data centres accounted for around 1.5% of the world's electricity consumption in 2024, or 415 terawatt-hours (TWh)." And it expects that to grow: "Data centre electricity consumption is set to more than double to around 945 TWh by 2030. This is slightly more than Japan's total electricity consumption today. AI is the most important driver of this growth, alongside growing demand for other digital services." So: much less per use, much more in total. That's not a contradiction once you remember that the total is energy per prompt × the number of prompts, and the second number is the one that explodes. ## It happened before, with coal In 1865, W. S. Jevons published *The Coal Question*. In it he made a claim that sounds backwards: "It is wholly a confusion of ideas to suppose that the economical use of fuel is equivalent to a diminished consumption. The very contrary is the truth." His evidence was Scotland's iron industry. Furnaces there had become far more efficient, and yet: "the reduction of the consumption of coal, per ton of iron, to less than one-third of its former amount, has been followed, in Scotland, by a ten-fold total consumption". Less than a third of the coal per ton. Ten times the coal in total. In his words, "the greater number of furnaces will more than make up for the diminished consumption of each." His explanation is that the saving itself drives the use: "It is the very economy of its use which leads to its extensive consumption. It has been so in the past, and it will be so in the future." ## The Jevons paradox The idea now carries his name. Yale's Energy History project: "His insight, often called Jevons' Paradox, is today used to explain why increased energy efficiency does not necessarily reduce present-day demand for electricity or gasoline." Applied to AI: cheaper prompts mean more prompts. And the use is already huge. "ChatGPT receives 2.5 billion prompts from global users every day, OpenAI told Axios," TechCrunch reported in July 2025. That's one product. Microsoft's CEO said it about AI. In January 2025, as quoted by Fortune, Satya Nadella wrote: "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." Researchers have taken it up too. A 2025 paper by Luccioni, Strubell and Crawford "examines how the problem of Jevons' Paradox applies to AI, whereby efficiency gains may paradoxically spur increased consumption." A careful note: none of this says Gemini's 33× *caused* the doubling. Jevons is the mechanism the sources name for why efficiency and rising totals go together, not a measurement of this case. ## For scale Data centres are not the whole story of the world's electricity. The IEA again: "Data centres account for around one-tenth of global electricity demand growth to 2030, less than the share from industrial motors, air conditioning in homes and offices, or electric vehicles." ## Takeaway 1. Per prompt, AI really is getting far more efficient (Gemini: 33× in a year). 2. Total use still climbs, because cheaper use means much more use. Jevons saw it with coal in 1865. 3. So "more efficient" doesn't settle the energy question; how much we use does. (And data centres are still about a tenth of the world's demand growth.) ## Sources - Google Cloud (Amin Vahdat, Jeff Dean), [Measuring the environmental impact of AI inference](https://cloud.google.com/blog/products/infrastructure/measuring-the-environmental-impact-of-ai-inference), 21 Aug 2025 - IEA, [Energy and AI, executive summary](https://www.iea.org/reports/energy-and-ai/executive-summary), Apr 2025 - W. S. Jevons, [The Coal Question](https://archive.org/stream/bub_gb_nVtJUsGgLcoC/bub_gb_nVtJUsGgLcoC_djvu.txt), 2nd ed., Macmillan, 1866, ch. VII "Of the Economy of Fuel", pp. 123–125 and 134–135 - Yale University, Energy History, [W. Stanley Jevons, The Coal Question (1865)](https://energyhistory.yale.edu/w-stanley-jevons-the-coal-question-1865/) - TechCrunch, [ChatGPT users send 2.5 billion prompts a day](https://techcrunch.com/2025/07/21/chatgpt-users-send-2-5-billion-prompts-a-day/), 21 Jul 2025 - Fortune, [Microsoft CEO Satya Nadella on DeepSeek and the Jevons paradox](https://fortune.com/2025/01/27/microsoft-ceo-satya-nadella-deepseek-optimism-jevons-paradox/), 27 Jan 2025 - Luccioni, Strubell, Crawford, [From Efficiency Gains to Rebound Effects: The Problem of Jevons' Paradox in AI's Polarized Environmental Debate](https://arxiv.org/abs/2501.16548), arXiv, Jan 2025 The video is animated with code and the script was written with AI assistance; every figure above is checked against the sources. Illustrative: the two-line chart (its curves show the shape, not data; the sourced numbers sit on the bars and counters), the furnaces and coal blocks, and the grid of prompt bubbles. Made with [Shotloop](https://oliverz.dev/products/shotloop), my free Mac app for code-made explainers, fixed one shot at a time.