ClimateLevel 6 of 6 territoryTerritory, level 6 of 6Used to calculate 1 entry, level 6Calculated with 3 entries, levels 1 and 4Database structure 
5Organisations
4Occupations and activities
3Uses
2Machine operations
1Calculation factorsUnder 25 years in France, personal use, 1 year
Personal AI use by 12 to 24 year olds in France over one year, excluding studies. Projet Celsius estimate.
- "Free personal use, 1 year" accounts for 74% of emissions in this scenario.
- Light use: 370 tCO2e; heavy use: 11 ktCO2e, with the other settings unchanged.
- Out of scope: homework and studies, counted with pupils and students, work, the over-12s, server manufacturing and model training.
Sentence to citeUnder 25s in France, personal use, 1 year, Projet Celsius estimate: 2.2 ktCO2e per population per year, for 4.8 GWh of electricity, servers worldwide (458 gCO2e/kWh).
- Calculated with
- Used to calculate
How the calculation is built
10.7 million young people, nearly half of whom use it for themselves. Insee headcounts as at 1 January 2026, Eurostat user shares for 2025.
Comparison with other territories
Characteristics
| Unit | 1 population, 1 year |
|---|---|
| Default location | world, 458 gCO2e/kWh (Ember 2025, life cycle); 11 other locations to choose from |
| Counted | personal use by 12 to 24 year olds, free or by subscription, data centre included |
| Not counted | homework and studies, counted with pupils and students, work, the over-12s, server manufacturing and model training |
| Status | estimated |
All the characteristics
| Identifier | IA-0510, projetcelsius.com/en/ai/database/IA-0510/ |
|---|---|
| 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
365 days, 5.4 seconds per video, servers worldwide: 2.2 ktCO2e, 4.8 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 |
|---|---|---|---|---|---|
| Free personal use, 1 year | × 3,480,000 | 750 Wh | 2.6 GWh | 55 % | recorded |
| ChatGPT Plus, 1 year | × 303,000 | 3 kWh | 920 MWh | 19 % | assumption |
| Free personal use, 1 year | × 1,220,000 | 750 Wh | 910 MWh | 19 % | assumption |
| ChatGPT Plus, 1 year | × 106,000 | 3 kWh | 320 MWh | 7 % | assumption |
Entries that use it (1)
Sources of the calculation
21 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.
Anthropic · 2026
Role and limits : Anthropic
What the publication covers. tokens of an image according to its size: one patch of 28 × 28 pixels per token, 1,296 for a megapixel
What it does not cover. electricity
Reading by the database. The publisher's billing rule
Analyse an image → : Read 1,000 tokens →
The photo cut into about 1,100 image tokens, read as text
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.
Analyse an image → : Read 1,000 tokens →
Length convention: a photo sent to an AI
Scope of the figure used. A one-megapixel photo, cut into image patches: 765 tokens at OpenAI for GPT-4o, 1,032 at Google, 1,229 for recent OpenAI models, 1,296 at Anthropic
Limit. This publication supports the quantity of text or content processed, not its carbon footprint.
Anthropic · June 2025
Role and limits : Anthropic
What the publication covers. an agent uses about 4 times the tokens of a conversation, a multi-agent system about 15 times
What it does not cover. electricity
Reading by the database. Publisher orders of magnitude, with no absolute value
Deep research → : Re-read from cache →
The context re-read from cache at each of the 40 steps, 40,000 tokens on average
Scope of the figure used. server, convention C4
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Apec · June 2025
Role and limits : Apec
What the publication covers. frequency of generative AI use at work, 2,000 private-sector managers surveyed in March 2025
What it does not cover. non-managers and the number of requests
Reading by the database. Recent French survey on frequency of use at work; copy read on a specialist press site
Support for the "Typical use" setting
Scope of the figure used. The median user among those who use AI, end of 2026: they turn to it every working day for part of their questions, writing and reading, about 5 to 15 requests per day in an office job. In an organisation or a territory, it also sets the share of employees who use it (25%).
Limit. This setting remains an assumption of the database, not a measurement of the occupation.
Support for the "Typical use" setting
Scope of the figure used. The median user among those who use AI, end of 2026: they turn to it every working day for part of their questions, writing and reading, about 5 to 15 requests per day in an office job. In an organisation or a territory, it also sets the share of employees who use it (25%).
Limit. This setting remains an assumption of the database, not a measurement of the occupation.
Crédoc · 2026
Role and limits : Crédoc
Author. Crédoc for Arcep, Arcom, the CGE and the ANCT
What the publication covers. use of generative AI by people aged 12 and over in France in 2025: who, for what, how often, paid or not
What it does not cover. the number of requests
Reading by the database. The reference survey in France, 4,145 people
Published observation on uses
Scope of the figure used. Generative AI users aged 12 and over: information search with AI every day (28% 1 to 2 times a week, 24% several times a month)
Limit. Survey of 2025 published in 2026, 4,145 people; support for the frequencies and the use
Published observation on uses
Scope of the figure used. Users: writing, translating or improving text with AI every day (25% 1 to 2 times a week, 20% several times a month)
Limit. Survey of 2025 published in 2026, 4,145 people; support for the frequencies and the use
Published observation on uses
Scope of the figure used. Users: creating images, photos or videos every day (16% 1 to 2 times a week, 17% several times a month)
Limit. Survey of 2025 published in 2026, 4,145 people; support for the frequencies and the use
Published observation on uses
Scope of the figure used. Users: programming or data processing every day (12% 1 to 2 times a week, 10% several times a month)
Limit. Survey of 2025 published in 2026, 4,145 people; support for the frequencies and the use
Published observation on uses
Scope of the figure used. Users: discussions and interactions with AI, all frequencies
Limit. Survey of 2025 published in 2026, 4,145 people; support for the frequencies and the use
Free personal use, 1 year → : Quick question →
About 10 information needs per day; share done with AI based on the 21% of users who look for information with AI every day and the 28% who do so 1 to 2 times a week (Crédoc, Baromètre du numérique, edition 2026, data 2025).
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.
Free personal use, 1 year → : Write an email →
About 10 personal messages or emails per week, assumption; low share done with AI, based on the Crédoc Baromètre du numérique (edition 2026, data 2025).
Scope of the figure used. tokens read and written, data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Free personal use, 1 year → : Proofread a page →
About 1 important text to rework per week (letter, CV, message), assumption; share done with AI based on the 38% of users who write or improve a text with AI at least once a week (Crédoc, Baromètre du numérique, edition 2026, data 2025).
Scope of the figure used. tokens read and written, data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Free personal use, 1 year → : Translate a page →
About 1 page to translate per week, assumption; share done with AI based on the Crédoc Baromètre du numérique (edition 2026, data 2025).
Scope of the figure used. tokens read and written, data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Free personal use, 1 year → : Create an image →
About 2 image requests per week (map, illustration, portrait), assumption; share done with AI based on the 26% of users who create images or videos at least once a week (Crédoc, Baromètre du numérique, edition 2026, data 2025).
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.
Free personal use, 1 year → : Edit a photo →
About 2 photos to retouch per week, assumption; share done with AI based on the Crédoc Baromètre du numérique (edition 2026, data 2025).
Scope of the figure used. photo read, image redone, data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Free personal use, 1 year → : 5-second clip →
About 1 video request per week, assumption; low share done with AI, about 2 clips per year at the typical level, the free plan limiting the number, assumption.
Scope of the figure used. server and data centre; 7 models measured
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Free personal use, 1 year → : Voice mode, 1 minute →
About 5 minutes per week during which one could talk to an assistant, assumption; share done with AI based on the 41% of users who chat with AI (Crédoc, Baromètre du numérique, edition 2026, data 2025).
Scope of the figure used. voice heard, voice spoken, context re-read, data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Free personal use, 1 year → : Generate a function →
About 1 small program every 2 weeks, assumption; share done with AI based on the 20% of users who program with AI at least once a week (Crédoc, Baromètre du numérique, edition 2026, data 2025).
Scope of the figure used. tokens read and written, data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Under 25s → : ChatGPT Plus, 1 year →
12 to 15 years, the 8% of personal users who pay for a subscription
Scope of the figure used. tasks handed to the AI, data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Published observation on uses
Scope of the figure used. Users who pay for a generative AI service, about 9% of the population
Limit. Survey of 2025 published in 2026, 4,145 people; support for the frequencies and the use
ChatGPT Plus, 1 year → : Question to ChatGPT →
About 15 occasions to ask a question per day; use higher than on the free plan, paying users being the most intensive users (Crédoc, Baromètre du numérique, edition 2026, data 2025); usage cap of the plan not verified.
Scope of the figure used. not published
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
ChatGPT Plus, 1 year → : Proofread a page →
About 1 text to rework per day, assumption; half of daily users of AI for writing pay (Crédoc, Baromètre du numérique, edition 2026, data 2025).
Scope of the figure used. tokens read and written, data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
ChatGPT Plus, 1 year → : Create an image →
About 3 image requests per week, assumption; 55% of daily users of AI for creating visuals pay (Crédoc, Baromètre du numérique, edition 2026, data 2025).
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.
Under 25s → : ChatGPT Plus, 1 year →
16 to 24 years, the 8% of personal users who pay for a subscription
Scope of the figure used. tasks handed to the AI, data centre
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
EnronSent
Role and limits : EnronSent
Author. Will Styler
What the publication covers. 96,106 messages sent by Enron employees, 13,810,266 words
What it does not cover. the messages received
Reading by the database. Professional emails from 2000 and 2001: an order of magnitude, not a norm
Write an email → : Write, with reasoning →
The email written, 150 words
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.
Write an email → : Read 1,000 tokens →
Length convention: an email
Scope of the figure used. A typical email: the sent messages of the Enron corpus average about 144 words; figure read in an excerpt, notice not verified
Limit. This publication supports the quantity of text or content processed, not its carbon footprint.
Write an email → : Write, with reasoning →
Length convention: an email
Scope of the figure used. A typical email: the sent messages of the Enron corpus average about 144 words; figure read in an excerpt, notice not verified
Limit. This publication supports the quantity of text or content processed, not its carbon footprint.
The 7 other publications in this section
Epoch AI · February 2025
Role and limits
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
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
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.
Eurostat · June 2026
Individuals - use of generative AI tools (isoc_ai_iaiu) ↗ (new tab)
Role and limits
Author. Eurostat, survey on ICT use in households
What the publication covers. share of people aged 16 to 74 who used generative AI in the last 3 months, at work, privately, for studies, by country and category, 2025 data
What it does not cover. the frequency of use
Reading by the database. Harmonised European survey, first year
Under 25s → : Free personal use, 1 year →
12 to 15 years: 3.31 million, 40% use AI for themselves, 92% without a paid subscription
Scope of the figure used. queries, images and videos, at the server
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Published observation on uses
Scope of the figure used. France, 16 to 74 years, used generative AI privately in the last 3 months, 2025
Limit. Share of users, which depends on the organisation and the population; here, support for the gap between occupations
Published observation on uses
Scope of the figure used. France, 16 to 74 years, used generative AI in the last 3 months, 2025 (European Union: 32.7%)
Limit. Share of users, which depends on the organisation and the population; here, support for the gap between occupations
Under 25s → : Free personal use, 1 year →
16 to 24 years: 7.4 million, 51% use AI for themselves
Scope of the figure used. queries, images and videos, at the server
Limit. The quantity and the share entrusted to AI are database assumptions, not a measured footprint.
Google · 2026
Role and limits
What the publication covers. up to hundreds of sites browsed and pages processed per search
What it does not cover. electricity
Reading by the database. A presentation page: a ceiling, not an average
Deep research → : Read 1,000 tokens →
The web pages read, about a hundred by default: Google announces up to hundreds of sites
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.
OpenAI · September 2025
How People Use ChatGPT (NBER Working Paper 34255) ↗ (new tab)
Role and limits
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
Published observation on uses
Scope of the figure used. ChatGPT consumer plans, average over 7 days to 26 June 2025
Limit. Used to check the average across the population; enterprise plans are not counted
Published observation on uses
Scope of the figure used. 18 billion messages per week sent by 700 million users, July 2025
Limit. Or about 26 messages per weekly user per week, the benchmark for the check on the average
Published observation on uses
Scope of the figure used. Share of practical advice in all messages, stable over a year
Limit. Support for the advice conversations in personal use
Published observation on uses
Scope of the figure used. Share of multimedia requests (mostly images), a little over 7% in July 2025 against 2% a year earlier
Limit. Support for the low use of images for occupations that do not create
Free personal use, 1 year → : Conversation of 10 exchanges →
About 3 pieces of practical advice per week (health, cooking, travel, sport), assumption; share done with AI based on practical advice, about 29% of messages (OpenAI, 2025).
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 Plus, 1 year → : Conversation of 10 exchanges →
About 1 piece of practical advice per day, assumption; share done with AI based on OpenAI (2025).
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.
Role and limits
What the publication covers. audio tokens: one per 100 milliseconds of the user's speech, one per 50 milliseconds of the assistant's speech
What it does not cover. electricity
Reading by the database. The publisher's billing rule, read in the page excerpt
Voice mode, 1 minute → : Read 1,000 tokens →
What the user says, half of the minute, in audio tokens
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.
Voice mode, 1 minute → : Write 1,000 tokens →
What the AI answers, the other half, in audio tokens
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.
Voice mode, 1 minute → : Read 1,000 tokens →
Length convention: one minute of voice heard by an AI
Scope of the figure used. One audio token per 100 milliseconds of the user's speech at OpenAI; Gemini counts 32 per second, or 1,920 per minute
Limit. This publication supports the quantity of text or content processed, not its carbon footprint.
Voice mode, 1 minute → : Write 1,000 tokens →
Length convention: one minute of voice spoken by an AI
Scope of the figure used. One audio token per 50 milliseconds of the assistant's speech at OpenAI
Limit. This publication supports the quantity of text or content processed, not its carbon footprint.
Projet Celsius · November 2026
Tokens per word, per line of code and per cell, measured with 7 public tokenisers
Role and limits
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.
Proofread a page → : Read 1,000 tokens →
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.
Proofread a page → : Write, with reasoning →
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.
Translate a page → : Write, with reasoning →
The translated page, counted as a French page
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.
Translate a page → : Read 1,000 tokens →
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.
Translate a page → : Write, with reasoning →
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.
Query a PDF → : Read 1,000 tokens →
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.
Deep research → : Write 1,000 tokens →
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.
Write a page → : Write, with reasoning →
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.
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
Agent in the browser → : Analyse an image →
A screenshot read at each step, 15 steps: published web agents take 13 on average
Scope of the figure used. image reading on the server; outside 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.
Delavande et al. · September 2025
Role and limits : Delavande et al.
Author. Delavande, Pierrard and Luccioni
What the publication covers. GPU, processor, memory
What it does not cover. data centre
Reading by the database. Seven open models
5-second clip → : Second of video →
The seconds of the measured video, 5.4
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.
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 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
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
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 3 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
Reference of the selected value
Scope of the figure used. "An average query", with no published method
Limit. Record R-0004, attached to the ChatGPT block
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
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 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
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.
Write an email → : 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.
Proofread a page → : 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.
Translate a page → : 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.
Edit a photo → : 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.
5-second clip → : 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.
Voice mode, 1 minute → : 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.
Query a PDF → : 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.
Deep research → : 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.
Generate a function → : 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
Write a page → : 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.
Agent in the browser → : 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.
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.
Nature of the data
Published readings, conventions, assumptions
- Projet Celsius assumptions 100 %
Calculation assumptions
Quantities, nature and source of each assumption
Free personal use, 1 year : depending on the settings
12 to 15 years: 3.31 million, 40% use AI for themselves, 92% without a paid subscription
ChatGPT Plus, 1 year : depending on the settings
12 to 15 years, the 8% of personal users who pay for a subscription
Free personal use, 1 year : depending on the settings
16 to 24 years: 7.4 million, 51% use AI for themselves
ChatGPT Plus, 1 year : depending on the settings
16 to 24 years, the 8% of personal users who pay for a subscription
Methodology conventions
With their alternatives
| 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.
- 31 October 2026Creation of the database, in its first version (0.1).
