A study estimates between 731 and 1,125 million m3 the annual water requirement of the IA server fleet in the United States, and 24 to 44 Mt CO2-equivalent associated emissions.
Published in Nature Sustainability, this study tackles a dimension that is often absent from the debate on the footprint of artificial intelligence: water consumption, which is inseparable from the cooling of facilities.
The authors estimate that the deployment of dedicated AI servers in the United States would represent, between 2024 and 2030, an annual water footprint of between 731 and 1,125 million cubic meters, with additional emissions of 24 to 44 million tons of CO2 equivalent per year. The wide range reflects uncertainty about the pace of deployment and the cooling technologies chosen.
Beyond the diagnosis, the work explores the trajectories for achieving carbon neutrality for these infrastructures. It emphasizes that the choice of geographical location — and therefore the local electricity mix and water availability — weighs as much as the efficiency of the equipment itself.
Read the full article on the original source
Go to the source →