UsesLevel 2 of 6 machine operationMachine operation, level 2 of 6Used to calculate 8 entries, levels 2 and 3Calculated with 1 entry, level 1Database structure 
6Territories
5Organisations
4Occupations and activities
3Uses
1Calculation factorsWrite 1,000 tokens
The server electricity needed to write 1,000 tokens.
- Selected reference: Medium model, Watershed default factor.
- CO2e is calculated from electricity and the worldwide grid; it is not a direct measurement of carbon.
- Out of scope: data centre cooling, counted separately (Google's measurement, from which Watershed derives the factor, already includes about 8%), server manufacturing and model training.
Sentence to citeWrite 1,000 tokens: 0.44 gCO2e per 1,000 tokens, for 0.96 Wh of electricity, servers worldwide (458 gCO2e/kWh).
- Calculated with
- Used to calculate
Origin of the value
The answer, the code, the draft of a model that reasons. Writing costs 3 times more than reading, the ratio selected by Watershed. The medium model keeps Watershed's default factor for a large operator (2026); the small and large models are derived from it using the gaps measured on the ML.ENERGY benchmark.
Comparison with the other machine operations
Characteristics
| Unit | 1,000 tokens |
|---|---|
| Default location | world, 458 gCO2e/kWh (Ember 2025, life cycle); 11 other locations to choose from |
| Counted | the computation on the server to write these words |
| Not counted | data centre cooling, counted separately (Google's measurement, from which Watershed derives the factor, already includes about 8%), server manufacturing and model training |
| Status | published, 1 reading of the same scope, 5 other published values |
All the characteristics
| Identifier | IA-0011, projetcelsius.com/en/ai/database/IA-0011/ |
|---|---|
| Version | 0.4 of 12 November 2026 |
| Category | Machine operations, what the server computes |
| Class | Small model, about 8 billion parameters (0.073 gCO2e), Medium model, Watershed default factor (0.44 gCO2e), Large frontier model (1.8 gCO2e) |
| Choice of the selected value | value taken as it stands from its source, without calculation |
| Sources | Watershed, 2026 |
| Licence | CC BY 4.0 |
Result for the chosen case
This block is made of no other: its value is read from its sources.
Entries that use it (8)
Synthesise one minute, Quick question, Deep research, Voice mode, 1 minute, Summarise a meeting, Code suggestion, Coding agent hour, Write, with reasoning
Sources of the calculation
2 publications, for this entry and those used to calculate it, with the chosen parameters.
Energy and factors
The references of the selected values and of the entries used in this calculation.
Watershed · Aug. 2026
Estimating GHG Emissions from AI Use: Framework for Corporate-Level Measurement ↗ (new tab)
Role and limits : Watershed
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 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
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
6 figures for the same quantity, each in its own scope.
| Source | Date | Electricity | What the figure measures | Scope | Status |
|---|---|---|---|---|---|
| Write 1,000 tokensselected by the database, v0.4 | 2026 | 0.96 Wh | the value selected by the database | same scope | published |
| Watershed ↗Estimating GHG Emissions from AI Use: Framework for Corporate-Level Measurement | August 2026 | 0.96 Wh | 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) | same scope | computation |
| ML.ENERGY, benchmark v3 ↗The ML.ENERGY Leaderboard v3, données par tâche | Feb. 2026 | 0.0329 Wh | Llama 3.1 8B, one H100 GPU, 64 requests at a time, GPU only, per 1,000 tokens written, reading of the question included | other scope | measured |
| ML.ENERGY, benchmark v3 ↗The ML.ENERGY Leaderboard v3, données par tâche | Feb. 2026 | 0.221 Wh | Llama 3.1 70B, 4 H100 GPUs, 64 requests at a time, GPU only, per 1,000 tokens written | other scope | measured |
| ML.ENERGY, benchmark v3 ↗The ML.ENERGY Leaderboard v3, données par tâche | Feb. 2026 | 0.665 Wh | Qwen 3 235B, of which 22 billion active, 8 H100 GPUs, 64 requests at a time, GPU only, per 1,000 tokens written | other scope | measured |
| ML.ENERGY, benchmark v3 ↗The ML.ENERGY Leaderboard v3, données par tâche | Feb. 2026 | 0.811 Wh | DeepSeek V3.1, 671 billion parameters of which 37 active, 8 B200 GPUs, 64 requests at a time, GPU only, per 1,000 tokens written | other scope | measured |
| ML.ENERGY, benchmark v3 ↗The ML.ENERGY Leaderboard v3, données par tâche | Feb. 2026 | 0.856 Wh | Llama 3.1 405B in 8 bits, 8 H100 GPUs, 64 requests at a time, GPU only, per 1,000 tokens written | other scope | measured |
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?
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 |
| Calculate a case without an entry | The formula, the default values and a step-by-step example |
Uses in the hub
Charts, figures, analyses and articles that cite this entry
History
Entry in version 0.4 of 12 November 2026, in the database in version 0.4
- 12 November 2026Status : working value becomes published.Source check of 12 November 2026: 0.96 Wh consistent with the factor published by Watershed (2026), 3 times the reading; same scope gap as for the reading of 1,000 tokens.
- 11 November 2026Watershed reading, nature : A default value, calibrated on non-production services becomes A default value that Watershed derives from Google's measurement.Watershed derives its per-token factors from the median query measured by Google (0.24 Wh); the value does not change.
- 2 November 2026Designation : 1,000 tokens written becomes 1,000 output tokens.More precise designation: the object first, the specification after the comma, the quantity or period at the end; the former title remains searchable.
- 2 November 2026Designation : 1,000 output tokens becomes Write 1,000 tokens.In people's own words.
- 2 November 2026Short title : Output tokens becomes Write 1,000 tokens.In people's own words.
- 2 November 2026Address : tokens-en-sortie becomes ecrire-1-000-tokens.Address in people's own words; the old one redirected.
- 2 November 2026Selected value (0.96).The value 0.96 moves to the default row (medium class), unchanged.
- 31 October 2026Creation of the database, in its first version (0.1).
