Calculated with French electricity, one question to an AI emits 0.012 gram of CO2e; Mistral AI publishes 1.14 gram, more than 90 times as much. These figures are accurate; they do not cover the same scope.
Read the analysis →
Your AI use over one year3 kgCO2eequivalent to 21 km by carThe calculation toolsEstimate your usage, compare server countries, browse orders of magnitude.Open the tools →
10 assistants rated out of 100: performance, ethics and environment.
Viewpoint
SocietySurface AI and deep AIPublic debate focuses on the AI of chat windows, which accounts for less than 5% of AI's electricity. Most of the footprint is decided in AI that has no interface.Read the viewpoint →
Tool
ClimateCompare the footprint of an AI use by the country where the servers areOne task, 12 electricity grids: a difference of up to 21 times between Norway and Poland.
Tool
ElectricityBrowse the orders of magnitude of electricity use, from one question to the planet16 sourced benchmarks, on a single scale.
Tool
UsesMy AI footprintThe carbon footprint of your AI use over a year, calculated task by task, with the effect of your settings.Hover over a country to read its value; + and - to zoom.
Emissions from producing one kilowatt-hour, on a life cycle basis, 2025 or latest year published. The electricity for the same computation emits 14 times more CO2e in Poland than in France.
Blog article ↗What is the environmental impact of AI?Electricity, water, metals, emissions: the overview, from the query to data centres.
Blog article ↗ElectricityHow much electricity does AI consume?From chips to grids: what data centres consume and what drives the increase.
Blog article ↗UsesHow much electricity does a ChatGPT query use?The measurements published in 2025, what they count and what they leave out.
Blog article ↗
WaterHow much water does a ChatGPT query use?Cooling, electricity, manufacturing: why estimates range from 1 to 100.
Blog article ↗ClimateThe carbon footprint of AI: what a ChatGPT query costs and what it weighs in a Bilan Carbone®From electricity to grams of CO2e, depending on the grid and the model.
Blog article ↗
HardwareHow and why does AI pollute?The full life cycle: mines, chips, data centres, uses.UsesThe 2 boxed words can be changed: the length of the text and the model. EcoLogits estimates, at the world electricity mix.
The 16 models, at each length →
GPT-6.1 Sol
A 250-word response, with the global electricity mix
0.39 gCO2e
equivalent to 2.7 m by combustion-engine car
0.78 Wh of electricity
GPT-6.1 Sol emits 19 times more than Claude Haiku 5.5
0.3 Wh per question: the hub's reference value, bracketed by the only 2 values published by providers, 0.24 Wh measured by Google for Gemini and 0.34 Wh declared by OpenAI for ChatGPT, with no methodology.
Read the article ↗From 0.26 to 50 mL per question, depending on what is counted: only the water evaporated for cooling (0.26 mL measured by Google for Gemini), or also the water used by power stations (UC Riverside, GPT-3, 2023).
Read the article ↗From 0.012 to 1.14 gCO2e, a gap of more than 90 times between 2 scopes and 2 models: the electricity alone for one question on the French grid, or the full life cycle of one response, as estimated by Mistral AI.
Read the article ↗2.9 Wh for an image, 90 Wh for a 5-second video with the most downloaded open generator. A video uses as much as 31 images, or 300 questions; an image, as much as 9.7 questions.
See the chart →24 to 40% of the world's jobs are exposed to AI, depending on the study (ILO, May 2025, for generative AI; IMF, January 2024, for AI in general), and the studies measure exposed tasks, not jobs lost.
Read the article ↗The only 2 electricity values published by providers are close to each other (0.24 Wh per question at Google, 0.34 Wh declared by OpenAI), and 2 choices matter more than the provider: reasoning mode, which multiplies the energy of a question by 2 to 30, and the country where the servers are, with 14 times more CO2e in Poland than in France.
Read the article ↗A figure for every use of AI, from a quick question to a whole country: calculated step by step, linked to its sources, free to reuse.
The hub is run by Projet Celsius, an environmental strategy consultancy: Bilan Carbone®, life cycle assessment, climate regulation.
Another question? An error, a source to add, a methodology question, a press request: write to us.
Guillaume Pakula, Clément Reynaud, Sébastien Pierfederici and Alexandre Veilly will reply.
