Hardwarepublished as isLevel 1 of 6 calculation factorCalculation factor, level 1 of 6Database structure 
6Territories
5Organisations
4Occupations and activities
3Uses
2Machine operationsManufacturing an AI server (8 GPUs)
The emissions from manufacturing the hardware that runs AI, from raw material extraction to the factory gate.
- Selected reference: The baseboard with 8 H100 GPUs, according to Nvidia.
- Counted: the manufacturing of the components and their assembly, up to the factory gate.
- Out of scope: the electricity consumed during use, the data centre building and the rest of the server when the row does not say so.
Sentence to citeManufacturing of an AI server (8 GPU): 1.31 tCO2e, figure published by Nvidia.
Origin of the value
Memory weighs the most: more than 40% for the baseboard with eight H100 GPUs, nearly half for the next generation. The default value is that of the HGX H100 baseboard with its 8 GPUs, according to Nvidia (1,312 kgCO2e), without the processors, the RAM, the storage or the server chassis; the whole server, according to GreenIT, weighs about 6 times more, with a simpler method.
Comparison with the other calculation factors
Characteristics
| Unit | 1 server : the baseboard with 8 H100 GPUs, according to Nvidia |
|---|---|
| Counted | the manufacturing of the components and their assembly, up to the factory gate |
| Not counted | the electricity consumed during use, the data centre building and the rest of the server when the row does not say so |
| Status | published, 1 reading of the same scope, 3 other published values |
All the characteristics
| Identifier | IA-0016, projetcelsius.com/en/ai/database/IA-0016/ |
|---|---|
| Version | 0.3 of 2 November 2026 |
| Category | Calculation factors, the foundation of the calculation |
| Component | The baseboard with 8 H100 GPUs, according to Nvidia (1.31 tCO2e), The baseboard with 8 B200 GPUs, according to Nvidia (2.27 tCO2e), A single H100 GPU, according to ADEME (235 kgCO2e), The whole server with 8 H100 GPUs, distribution and end of life included, according to GreenIT (8.34 tCO2e) |
| Choice of the selected value | value taken as it stands from its source, without calculation |
| Sources | Nvidia, 2025 |
| Licence | CC BY 4.0 |
Variants
Sources of the calculation
1 publication, 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.
Role and limits : Nvidia
What the publication covers. manufacturing of the baseboard and its components, up to the factory gate
What it does not cover. rest of the server, use, end of life
Reading by the database. The footprint published by the manufacturer, reviewed by a third party
Reference for the variant "The baseboard with 8 H100 GPUs, according to Nvidia"
Scope of the figure used. HGX H100 baseboard with 8 H100 GPUs, from cradle to factory gate, analysis reviewed by a third party
Limit. Without the server's processors, RAM or storage
Published values
4 figures for the same quantity, each in its own scope.
| Source | Date | CO2e | What the figure measures | Scope | Status |
|---|---|---|---|---|---|
| Nvidia ↗HGX H100 Product Carbon Footprint Summary | 2025 | 1.31 tCO2e | HGX H100 baseboard with 8 H100 GPUs, from cradle to factory gate, analysis reviewed by a third party | same scope | estimated |
| Nvidia ↗HGX B200 Product Carbon Footprint Summary | 2025 | 2.27 tCO2e | HGX B200 baseboard with 8 B200 GPUs, from cradle to factory gate | same scope | estimated |
| ADEME ↗Analyses de cycle de vie de GPU pour l'intelligence artificielle | Feb. 2026 | 235 kgCO2e | A single H100 GPU, without the baseboard, from cradle to factory gate, ADEME parametric model | other scope | estimated |
| GreenIT ↗Impacts environnementaux et sanitaires de l'intelligence artificielle dans le monde en 2025 et 2030 | Sept. 2026 | 8.34 tCO2e | Entry-level AI server with 8 H100 GPUs, manufacturing, distribution and end of life, excluding use | other scope | estimated |
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
- ChartManufacturing of an AI server, by component →
- Blog articleHow and why does AI pollute? ↗
- Blog articleWhat is the environmental impact of AI? ↗
History
Entry in version 0.3 of 2 November 2026, in the database in version 0.4
- 11 November 2026Description : The default value was not named becomes The default value is the HGX H100 baseboard with its 8 GPUs (Nvidia), without the rest of the server.The whole server (GreenIT, 8,341 kgCO2e) weighs about 6 times more; the default does not change.
- 2 November 2026Designation : The manufacturing of an AI server becomes Manufacturing of an AI server (8 GPU).More precise designation: the object first, the specification after the comma, the quantity or period at the end; the former title remains searchable.
- 1st November 2026Creation : version 0.1.Entry added: the manufacturing of an AI server, four published assessments (Nvidia, ADEME, GreenIT).
- 31 October 2026Creation of the database, in its first version (0.1).
