
Focus · AI and its hardware
Recycling and reuse
The lifespan of devices and the electronic waste that AI could add.
By Projet Celsius
Focus areaSemiconductorsThe cost of manufacturing AI chips and servers.Open the focus area →
Focus areaMetals and resourcesThe hardware share of AI in published life cycle assessments, and the materials it requires.Open the focus area →
You are hereRecycling and reuseThe lifespan of devices and the electronic waste that AI could add.Charts and maps of the focus area
Charts and maps →Focus area
Chart2020 to 2030Electronic waste and AIWorld, 202262 MtWorld, 203082 MtGenerative AI, cumulative 2020-20301.2 to 5 MtITU and UNITAR, 2024; Wang et al., 2024ChartImpact CO2 values consulted in October 2026The footprint of digital devicesTelevision370 kgLaptop190 kgTablet87 kgInternet router81 kgSmartphone80 kgADEME, Impact CO2
Analyses and articles
Analyses
Analysis
Hardware14 November 2026Why does AI need graphics cards?To write each word, a dense language model performs one multiplication and one addition for each of its billions of parameters, and has to read them all from memory. A graphics card performs these operations by the thousand at the same time; the H100 SXM, installed 8 at a time in AI servers, reads its memory at 3.35 terabytes per second. One hour of work by this card uses about 0.99 kWh, including its share of the server and the data centre, according to the AI Impact Database estimate; manufacturing it emitted 164 to 235 kgCO2e according to estimates by Nvidia and ADEME.
Analysis
Hardware14 November 2026What is AI made of?An AI runs on servers whose computing chips are fabricated in Taiwan, whose memory is made in South Korea and Japan, and several of whose semiconductor raw materials are produced mainly in China. On the HGX board that carries 8 H100 chips, memory accounts for 42% of manufacturing emissions according to Nvidia, and 49% on the next generation. With the world average electricity mix, manufacturing the server accounts for about 7% of the emissions of one hour of computing; with French electricity, close to half, according to our estimate based on the AI Impact Database.On the Projet Celsius blog ↗
Across the hub
AI Impact DatabaseAll the figures, their methodology and their sources.
AI Ranking10 assistants rated out of 100: performance, ethics and environment.
AnalysesThe signed pieces: analyses, viewpoints, "Where does this figure come from?".
Charts and mapsThe reference visuals, with their data and sources.
ToolsOpen calculations, your assumptions on the left, the result on the right.



A question about a figure?
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.
Write to us →
Projet Celsius, which runs this hub, also helps companies measure and reduce the footprint of their AI use.
Accounting for AI in your Bilan Carbone® ↗
Water








