ChatGPT vs Google search: energy use
A text AI prompt and a Google search use about the same energy today: roughly 0.3 Wh each. The figure that says otherwise, ten times more for AI, comes from a 2023 estimate built on hardware assumptions, not a measurement of a real service. Here is where it comes from, and what still holds up.
Today, a text prompt uses about as much as a Google search: roughly 0.3 Wh on both sides. The famous “10 times more” comes from a single estimate published in 2023 (2.9 Wh per query), repeated as-is by the IEA in 2024, and never measured on a production service. Honest caveat: reasoning or multimodal queries remain far heavier.
Figures updated on June 30, 2026
How much does an AI prompt use today, compared with a Google search?
Roughly 0.3 Wh on both sides. Google measured 0.24 Wh for the median Gemini Apps text prompt, on 21 August 2025, across its own fleet. OpenAI announced 0.34 Wh for an average query, on 10 June 2025. A peer-reviewed study published in April 2026 in the journal Joule simulates a median of 0.31 Wh on open-weight models. Three different methods, a measurement, a statement, a simulation, one single order of magnitude. On the search side, 0.3 Wh is the figure Google has quoted since 2009 for an average query: nobody has published a revision since, even though data centers have grown more efficient too.
Where does the famous “10 times more than a Google search” come from?
From a commentary published in October 2023 in the journal Joule by researcher Alex de Vries. His estimate: 2.9 Wh per ChatGPT query against 0.3 Wh for a Google search, a factor of ten. The calculation borrows a SemiAnalysis assessment: 3,617 Nvidia HGX A100 servers (28,936 GPUs in total) to run ChatGPT, drawing 564 megawatt-hours a day, spread over roughly 195 million daily queries. So this was never a measurement of ChatGPT in production: it's a reconstruction from the hardware judged necessary, at a time when OpenAI published no figures of its own. The IEA repeated this calculation as-is in its “Electricity 2024” report, published in January 2024, which gave it the authority of an official number even though its source stayed a third party's estimate.
Why is the 2.9 Wh figure outdated today?
Because it predates the providers' own published numbers, and it rests on hardware and a scale-up that don't reflect an optimized service. The authors of the April 2026 Joule study go further: they judge the most-cited estimates overstated by a factor of 4 to 20, because they assume out-of-production conditions, without batching or fully loaded hardware. On this specific point, the facts back them up. Google (0.24 Wh), OpenAI (0.34 Wh) and the Joule study itself (0.31 Wh) all fall between 0.24 and 0.34 Wh, far below the 2.9 Wh from 2023.
Does this hold for every type of AI query?
No, only for a standard text prompt. A reasoning query, the kind that shows a thinking step before the answer, costs about 20 Wh, more than 65 times more. Generating an image costs around 9 Wh, a bracket built from measurements on open models (1.4 to 11.5 Wh depending on the model), for lack of a published figure from a commercial service. Comparing a Google search to ChatGPT only makes sense for the simplest, most common use: typing a question and reading a text answer.
Should you change how you search for information to save energy?
Not for a simple question: swapping a search for an AI prompt changes nothing about the energy use, and adds an error risk a search engine doesn't have. The trade-off makes sense the other way round, when a single answer avoids several searches and several page visits: there, the math can tip in AI's favor. That's not what a line-by-line comparison of one prompt against one search measures, and this site doesn't claim to know which of the two uses is more common.
Sources
- Epoch AI : énergie par requête ChatGPT
- Google Cloud (21 août 2025) : impact environnemental de l'inférence IA, 0,24 Wh, 0,26 mL d'eau et 0,03 g CO₂e pour le prompt texte médian de Gemini Apps
- de Vries : « The growing energy footprint of artificial intelligence », Joule (oct. 2023, DOI 10.1016/j.joule.2023.09.004) : 2,9 Wh par requête ChatGPT contre 0,3 Wh pour une recherche Google, hypothèse fondée sur un parc de GPU Nvidia A100
- IEA : « Electricity 2024 » (janv. 2024), reprend l'estimation de 2,9 Wh par requête ChatGPT contre 0,3 Wh par recherche Google
- Oviedo et al. : « Energy use of AI inference, efficiency pathways, and test-time scaling », Joule (avril 2026, DOI 10.1016/j.joule.2026.102430 ; préprint arXiv 2509.20241) : médiane de 0,31 Wh par requête, intervalle interquartile 0,16 à 0,60 Wh
- « How Hungry is AI? » (arXiv 2505.09598)
- Luccioni, Jernite & Strubell : « Power Hungry Processing: Watts Driving the Cost of AI Deployment? » (arXiv 2311.16863, nov. 2023 ; ACM FAccT, juin 2024) : 1,38 kWh pour 1 000 images avec Stable Diffusion v1.5, 11,49 kWh avec Stable Diffusion XL, le modèle le moins efficace du panel
Frequently asked
Does ChatGPT use more than a Google search?+
For an average text prompt, no: both sit around 0.3 Wh, measured or announced by Google and OpenAI, and confirmed by a peer-reviewed study in 2026. The gap shows up mainly with reasoning, image or video queries.
Where does the “10x more” figure come from?+
From a 2023 estimate by researcher Alex de Vries in the journal Joule (2.9 Wh per ChatGPT query), repeated as-is by the IEA in 2024. It predated the providers' own public figures and rested on hardware assumptions, not a measurement of a real service.
Can one AI answer replace several searches?+
Often yes: a single answer synthesizes what would take several searches and page visits. In practice, the balance can even tip in AI's favor, as long as you don't repeat the same question.