ClimateLevel 6 of 6 territoryTerritory, level 6 of 6Calculated with 5 entries, levels 1, 3 and 6Database structure 
5Organisations
4Occupations and activities
3Uses
2Machine operations
1Calculation factorsAI assistants worldwide, people's requests, 1 day
What people ask AI assistants worldwide in one day. Projet Celsius estimate.
- "ChatGPT, per day" accounts for 55% of emissions in this scenario.
- 64% of the calculation rests on database assumptions; it does not measure actual use.
- Out of scope: uses by businesses and developers through programming interfaces, agents, model training and idle servers.
Sentence to citeAI assistants worldwide, people's requests, 1 day, Projet Celsius estimate: 1.1 ktCO2e per day, for 2.4 GWh of electricity, servers worldwide (458 gCO2e/kWh).
How the calculation is built
Less than 1% of the electricity of AI in data centres (0.5 to 0.9%). Not counted are businesses and developers through programming interfaces, agents or model training. Counted are ChatGPT, Gemini, Le Chat, Google's AI-generated search overviews and the other assistants. A sum of readings and assumptions, 1.6 times below the calculation based on populations, to be compared with the electricity of AI in data centres, never added to it.
Comparison with other territories
Characteristics
| Unit | 1 day |
|---|---|
| Default location | world, 458 gCO2e/kWh (Ember 2025, life cycle); 11 other locations to choose from |
| Counted | the conversations, images and AI-generated overviews that people request, data centre included |
| Not counted | uses by businesses and developers through programming interfaces, agents, model training and idle servers |
| Status | estimated |
All the characteristics
| Identifier | IA-0519, projetcelsius.com/en/ai/database/IA-0519/ |
|---|---|
| Version | 0.1 of 2 November 2026 |
| Category | Territories, cities, territories, global services |
| Choice of the selected value | database calculation from the entries this one is made of, at the quantities of the unit |
| Sources | calculation of the database |
| Licence | CC BY 4.0 |
Usage parameters
Advanced settings
Result for the chosen case
150,000,000 images per week, servers worldwide: 1.1 ktCO2e, 2.4 GWh.
Breakdown by item
Quantity, unit value, contribution and nature of the data for each item
| Item | Quantity | Unit value | Contribution | Share | Nature of the data |
|---|---|---|---|---|---|
| ChatGPT, per day | × 1 | 880 MWh | 880 MWh | 36 % | recorded |
| Gemini app, per day | × 1 | 650 MWh | 650 MWh | 27 % | assumption |
| AI overview of a search | × 3,430,000,000 | 0.13 Wh | 450 MWh | 19 % | assumption |
| ChatGPT, per day | × 0.5 | 880 MWh | 440 MWh | 18 % | assumption |
| Vibe, per day | × 1 | 450 kWh | 450 kWh | 0 % | assumption |
Entries that use it
No other entry in the database is calculated from this one.
Sources of the calculation
16 publications, for this entry and those used to calculate it, with the chosen parameters.
The result is calculated by Projet Celsius. The sources below support its uses, its factors and its conversion; they do not all measure this total.
Uses and assumptions
These publications shed light on AI tasks and uses. They do not measure the carbon footprint of the case calculated.
Arcom · June 2026
Baromètre de l'audience des services d'intelligence artificielle, 1re édition ↗ (new tab)
Role and limits : Arcom
Author. Arcom, based on Médiamétrie Netratings
What the publication covers. unique visitors and time spent per AI service, France, 2023 to 2026
What it does not cover. query volumes
Reading by the database. A panel-based audience measurement, service by service
Vibe, per day → : Quick question →
Le Chat answers over one day, counted as a question to an average model with an answer of 400 tokens, that of the Mistral AI analysis, which publishes only carbon and water; assumed volume, Mistral AI does not publish it
Scope of the figure used. computation of the response, data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
AI assistants worldwide, per day → : ChatGPT, per day →
The other assistants (Meta AI, Copilot, Claude, Grok, Perplexity, Chinese assistants), counted as a share of ChatGPT
Scope of the figure used. queries, based on the average announced by OpenAI
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Epoch AI · February 2025
Role and limits : Epoch AI
What the publication covers. the calculation from the hardware and the assumed size of the model
What it does not cover. actual measurements
Reading by the database. A public, transparent calculation
Conversation of 10 exchanges → : Re-read from cache →
The context re-read from cache at each turn: the previous exchanges, and the document already in the conversation if there is one
Scope of the figure used. server, convention C4
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Reference of the selected value
Scope of the figure used. GPT-4o, 500 tokens written, full scope
Limit. A public calculation, from the hardware and the assumed size of the model
Role and limits : Google
What the publication covers. monthly users of the Gemini app, images generated per day
What it does not cover. the number of questions
Reading by the database. A vendor announcement, without a methodology
Gemini app, per day → : Question to Gemini →
The questions asked in one day by the billion monthly users, 0.7 per user
Scope of the figure used. full service, data centre included, measured
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Gemini app, per day → : Create an image →
More than 150 million images generated every day, according to Google; the electricity of a Gemini image is not published, that of an open generator of 2023 stands in for it
Scope of the figure used. server and data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
OpenAI · September 2025
How People Use ChatGPT (NBER Working Paper 34255) ↗ (new tab)
Role and limits : OpenAI
Author. Aaron Chatterji, Tom Cunningham, David Deming, Zoë Hitzig, Christopher Ong, Carl Shan, Kevin Wadman (OpenAI, Duke, Harvard)
What the publication covers. exact message volumes of the consumer plans; classification of a sample of about 1.1 million conversations by topic, intent and link to work
What it does not cover. enterprise plans, under-18s, logged-out users
Reading by the database. The only published measurement of the volumes and uses of a major assistant; internal data, automatic classification
ChatGPT, per day → : Conversation of 10 exchanges →
76% of messages: an ordinary exchange, one tenth of a ten-exchange conversation
Scope of the figure used. messages, context re-read from cache, responses, data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
ChatGPT, per day → : Conversation of 10 exchanges →
10% of messages with reasoning
Scope of the figure used. messages, context re-read from cache, responses, data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
ChatGPT, per day → : Web search with an AI →
10% of messages that launch a web search
Scope of the figure used. searches, pages read, response, data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
ChatGPT, per day → : Create an image →
4% of messages that create an image, out of the 7% of "multimedia" messages
Scope of the figure used. server and data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Projet Celsius · November 2026
Tokens per word, per line of code and per cell, measured with 7 public tokenisers
Role and limits : Projet Celsius
What the publication covers. 479,888 words from 196 blog articles and their English translation, code from the site and the Python library, a table of measurements; tokenisers of GPT-4o, GPT-4, Llama 3, Gemini (via Gemma 3), Mistral, DeepSeek and Qwen
What it does not cover. the Claude tokeniser, not public
Reading by the database. A measurement on technical texts, figures included; to be repeated if publishers change their tokeniser
Projet Celsius measurement, repeated on request with the public tokenisers
Conversation of 10 exchanges → : Re-read from cache →
Length convention: one page
Scope of the figure used. A report page, headings and paragraphs included: between the 1,500-character sheet (250 words) and a full A4 page in single spacing (about 550 words, at 6.1 characters per word measured on our articles)
Limit. This publication supports the quantity of text or content processed, not its carbon footprint.
The other publication in this section
Zhang and Nam · June 2026
HANSEL: Extracting Breadcrumbs from Web Agent Trajectories for Interactive Verification ↗ (new tab)
Role and limits
Author. Yujin Zhang and Daye Nam
What the publication covers. 45 web agent trajectories: 13.16 steps and 6.02 pages visited per task on average
What it does not cover. electricity
Reading by the database. A small sample, which gives the order of magnitude of an agent's steps
Web search with an AI → : Read 1,000 tokens →
The 5 pages found, read by the model, and the question: web agents visit 6 pages per task on average
Scope of the figure used. server, excluding the data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Energy and factors
The references of the selected values and of the entries used in this calculation.
ADEME · July 2026
Role and limits : ADEME
Reading by the database. A public calculation
AI assistants worldwide, per day → : ChatGPT, per day →
ChatGPT, the selected reading
Scope of the figure used. queries, based on the average announced by OpenAI
Limit. This figure covers part of the calculation; it does not measure the whole entry.
Reference of the selected value
Scope of the figure used. ChatGPT, one day, queries
Limit. A calculation, not a measurement published by OpenAI
Google · January 2009
Role and limits : Google
What the publication covers. a search, data centre
Reading by the database. Never remeasured
Reference of the selected value
Scope of the figure used. A search, data centre included
Limit. Never publicly remeasured; Google's AI answers are not published
Role and limits : Google
What the publication covers. chips, host, standby machines, data centre, on-site water
What it does not cover. manufacturing, training
Reading by the database. The only production measurement published by a provider
Reference of the selected value
Scope of the figure used. Median text question, May 2025; chips, host, idle reserve machines, data centre
Limit. Record R-0001, attached to the Gemini block
Reference of the selected value
Scope of the figure used. Median of a text prompt; chips, host, standby machines, data centre
Limit. The only production measurement published by a provider, not audited by a third party
Luccioni et al. · October 2024
Role and limits : Luccioni et al.
Author. Luccioni, Jernite and Strubell
What the publication covers. GPU, processor, memory
What it does not cover. data centre
Reading by the database. Open models only
Create an image → : Generate an image →
One image, of the chosen size: standard by default
Scope of the figure used. GPU, processor, memory; excluding the data centre
Limit. This figure covers part of the calculation; it does not measure the whole entry.
Reference for the variant "Standard image, from 512 to 1,024 pixels per side"
Scope of the figure used. Average of the open models tested, 1,000 images; median 1.35 Wh, maximum 11.49 Wh
Limit. Open models only; closed generators publish nothing
Microsoft Research · 2026
Estimation de l'énergie des requêtes en production (Joule) ↗ (new tab)
Role and limits : Microsoft Research
Author. Microsoft Research, Oviedo et al.
What the publication covers. H100 nodes, production
Reading by the database. Calibrated on production
Reference of the selected value
Scope of the figure used. Median, 500 tokens read and 300 written, H100 nodes
Limit. Not a measurement, but calibrated on production; judges public estimates to be 4 to 20 times too high
The 4 other publications in this section
ML.ENERGY · February 2026
Role and limits
Author. Chung et al., University of Michigan
What the publication covers. GPU energy per token, per response, per image, in steady state, according to the number of requests served at once
What it does not cover. processors, host memory, data centre cooling, manufacturing
Reading by the database. The best open benchmark: it serves to compare model sizes with each other, not as a production value
Support for the "Small model" setting
Scope of the figure used. A model of about 8 billion active parameters, like Llama 3.1 8B: about 6 times less electricity per token than the medium model, according to the ML.ENERGY benchmark.
Limit. The setting is a convention of the database supported by this publication.
Reference of the variant "Small model, about 8 billion parameters"
Scope of the figure used. Llama 3.1 8B, one H100 GPU, 64 requests at a time, GPU only, per 1,000 tokens written, reading of the question included
Limit. GPU only, on a benchmark: the database keeps only the ratio to the 70 billion model, 1 to 6.7
The measurement covers a different scope; it is not a direct measurement of this entry.
OpenAI · June 2025
Role and limits
Author. Sam Altman, OpenAI
What the publication covers. an average query
What it does not cover. method, scope
Reading by the database. One sentence, no method
Reference of the selected value
Scope of the figure used. "An average query", with no published method
Limit. No published scope; consistent with Google
Role and limits
Author. PEReN, published by Arcep
What the publication covers. GPU energy at inference, 22 open models, 6 question sets
What it does not cover. cooling, manufacturing, training
Reading by the database. Open models
Reference for the variant "With reasoning, medium effort"
Scope of the figure used. + 92% on average with reasoning, effect estimated by regression on 22 open models and 6 question sets; 95% interval from - 21% to + 348%
Limit. Open models, varied tasks
Watershed · Aug. 2026
Estimating GHG Emissions from AI Use: Framework for Corporate-Level Measurement ↗ (new tab)
Role and limits
Author. Bistline, Ulissi, Davis, Glidden, Joyce, Li, Mohsenin and Suh (Watershed)
What the publication covers. factors per 1,000 tokens read and written at the server before PUE (host and standby machines included), PUE separate (1.10 for a large operator, 1.56 for an unknown operator)
What it does not cover. clean measurements, the model and the hardware
Reading by the database. Default factors proposed for review (public data as of June 2026); for a large operator, taken from Google's measurement published in 2025 (Elsworth et al.), which already contains the data centre overhead; constant per token, they underestimate long contexts and agents according to the authors
Reference for the variant "Medium model, Watershed default factor"
Scope of the figure used. Default factor for 1,000 tokens read, medium model, large operator, at the server before PUE (Watershed table 2, public data as of June 2026)
Limit. A default value that Watershed derives from Google's measurement (0.24 Wh for the median text query of the Gemini app), with about 500 tokens per query and a written token costing 3 times a read token; Watershed does not give the split between read and written tokens. The 0.24 Wh already include the overhead of Google's data centre (0.02 Wh, 8%): with 375 tokens read and 125 written, the simplest reading, the factor contains it, and the PUE of 1.10 applied afterwards counts it a second time (about 9%). Values proposed for review, final at the next version of the framework
Reference for the variant "Medium model, Watershed default factor"
Scope of the figure used. Default factor for 1,000 tokens written, medium model, large operator, at the server before PUE (Watershed table 2, public data as of June 2026)
Limit. 3 times the reading factor, based on the ratio between writing and reading selected by Watershed (Patel et al., Splitwise, 2024). Same derivation and same caveat as for reading: Google's 0.24 Wh already include about 8% of data centre. Values proposed for review, final at the next version of the framework
Conversation of 10 exchanges → : Data centre PUE →
The data centre, convention C3
Scope of the figure used. building cooling and losses
Limit. This figure covers part of the calculation; it does not measure the whole entry.
Reference of the selected value
Scope of the figure used. Default factor for a large operator, from the server to the data centre; 1.56 for an unknown operator
Limit. A default value, not a measurement
Web search with an AI → : Data centre PUE →
The data centre, convention C3
Scope of the figure used. building cooling and losses
Limit. This figure covers part of the calculation; it does not measure the whole entry.
Create an image → : Data centre PUE →
The data centre, convention C3
Scope of the figure used. building cooling and losses
Limit. This figure covers part of the calculation; it does not measure the whole entry.
Quick question → : Data centre PUE →
The data centre, convention C3
Scope of the figure used. building cooling and losses
Limit. This figure covers part of the calculation; it does not measure the whole entry.
Reference factor: Medium model, Watershed default factor
Scope of the figure used. Default factor for 1,000 tokens read, medium model, large operator, at the server before PUE (Watershed table 2, public data as of June 2026)
Limit. This measurement concerns the reference variant, not a measurement of "Small model, about 8 billion parameters". A default value that Watershed derives from Google's measurement (0.24 Wh for the median text query of the Gemini app), with about 500 tokens per query and a written token costing 3 times a read token; Watershed does not give the split between read and written. The 0.24 Wh already include the overhead of Google's data centre (0.02 Wh, 8%): with 375 tokens read and 125 written, on the simplest reading, the factor contains it, and the PUE of 1.10 applied afterwards counts it a second time (about 9%). Values proposed for review, to be finalised in the next version of the framework
Reference factor: Medium model, Watershed default factor
Scope of the figure used. Default factor for 1,000 tokens written, medium model, large operator, at the server before PUE (Watershed table 2, public data as of June 2026)
Limit. This measurement concerns the reference variant, not a measurement of "Small model, about 8 billion parameters". 3 times the read factor, based on the write-to-read ratio adopted by Watershed (Patel et al., Splitwise, 2024). Same derivation and same caveat as for reading: Google's 0.24 Wh already include about 8% for the data centre. Values proposed for review, to be finalised in the next version of the framework
Conversion into carbon
The source of the electricity factor of the selected location.
Worldwide: 458 gCO2e/kWh, life cycle.
Role and limits : Ember
Author. Ember, Our World in Data
What the publication covers. electricity generation over the life cycle, by country
What it does not cover. guarantees of origin
Reading by the database. The same source for all countries
Conversion of electricity into carbon: worldwide
Scope of the figure used. the production of electricity over the life cycle, including power plant construction and fuels
Limit. An average for the chosen country, not the electricity contract or the data centre of a specific service.
Published values
1 figure for the same operation, each within its own scope.
| Source | Date | Electricity | What the figure measures | Scope | Status |
|---|---|---|---|---|---|
| AI assistants worldwide, per dayselected by the database, v0.1 | 2026 | 2.43 GWh | the value selected by the database | same scope | estimated |
| Database checks and calculations | |||||
| Projet Celsius, based on published estimatesL'électricité de l'IA dans le monde en 2025 | 2026 | 274 GWh to 477 GWh | AI in the world's data centres in 2025, from 100 to 174 TWh, scaled to one day | check | computation |
The published figures do not all count the same scope; the scope of each figure is written next to it. Why do the estimates diverge?
Nature of the data
Published readings, conventions, assumptions
- published readings 36 %
- Projet Celsius assumptions 64 %
Calculation assumptions
Quantities, nature and source of each assumption
ChatGPT, per day : counted 1 times
ChatGPT, the selected reading
Gemini app, per day : counted 1 times
Gemini
Vibe, per day : counted 1 times
Le Chat
AI overview of a search : × 3,430,000,000
The AI-generated overviews at the top of Google results, the search itself not counted
ChatGPT, per day : depending on the settings
The other assistants (Meta AI, Copilot, Claude, Grok, Perplexity, Chinese assistants), counted as a share of ChatGPT
Methodology conventions
With their alternatives
| The standard query | A query counts 100 tokens read and 400 written |
|---|---|
| The data centre | Server electricity is multiplied by 1.10 for the data centre |
| The cache | A token re-read from the cache costs 10% of a token read |
| The default location | Without other information, the servers are in the world, at the world average electricity |
| Electricity, location-based, on a life cycle basis | The kilowatt-hour is that of the servers' country, on a life cycle basis, from a single source for all countries |
| Two significant figures | A calculated value is rounded to two significant figures, a published value is reproduced with the precision of its source; the calculation keeps full precision |
| The formula | the sum of what makes up the entry, then CO2e: electricity (Wh) multiplied by the carbon intensity of the server location (gCO2e/kWh), divided by 1,000 |
| Calculate a case without an entry | The formula, the default values and a step-by-step example |
History
Entry in version 0.1 of 2 November 2026, in the database in version 0.4
- 2 November 2026Version 0.1 of the entry, as of the date of its current value.
- 12 November 2026Designation : All generative AI worldwide, people's use, 1 day becomes AI assistants worldwide, people's requests, 1 day.Decision of 12 November 2026: the title states the scope, narrower than the place or the service alone, so that a figure quoted on its own is not taken for all the AI of the place; the address and the value do not change; the old title remains searchable.
- 12 November 2026Short title : Generative AI worldwide, 1 day becomes AI assistants worldwide, per day.Decision of 12 November 2026: the title states the scope, narrower than the place or the service alone, so that a figure quoted on its own is not taken for all the AI of the place; the address and the value do not change; the old title remains searchable.
- 12 November 2026Description : first sentence without the scope becomes a first sentence that says what is not counted, and the share of AI electricity covered, based on the check reading.Decision of 12 November 2026: the title states the scope, narrower than the place or the service alone, so that a figure quoted on its own is not taken for all the AI of the place; the address and the value do not change; the old title remains searchable.
- 31 October 2026Creation of the database, in its first version (0.1).
