A new study in Nature Sustainability maps the energy, water, and climate footprint of AI servers and identifies strategies to reach net-zero targets.
Artificial intelligence (AI) is reshaping industries from healthcare to clean energy, but a new study in Nature Sustainability forecasts that, by 2030, AI servers in the United States could consume 220–532 terawatt-hours (TWh) of electricity annually — up to 10% of the nation’s current usage — while generating 24–44 million tonnes of CO₂ emissions and a water footprint of 731–1,125 million cubic metres per year.
The research was conducted by an international team including The Environmental Change Institute (ECI) at the University of Oxford; KTH Royal Institute of Technology, Sweden; Cornell University, USA; Concordia University, Canada; the RFF-CMCC European Institute on Economics and the Environment, Italy; and Politecnico di Milano, Italy. The team provides a detailed assessment of how efficiency measures, server location, and renewable energy penetration could reduce these impacts.
Co-author Dr Francesco Fuso Nerini, Honorary Research Associate at the ECI and Professor at KTH Royal Institute of Technology, said:
Artificial intelligence enables many societal applications — from improving healthcare to accelerating clean energy transitions — but it also has a real energy, water and climate footprint.
If AI is powered by fossil fuels instead of renewables, its emissions could soar. Our research shows that smart planning, efficiency, and clean energy are essential for AI to truly support a sustainable future.”
The paper identifies pathways for achieving net-zero carbon and water footprints by 2030. Efficiency measures such as advanced liquid cooling and optimised server utilisation could reduce energy and water use by up to 12% and 32%, respectively. Strategic siting of servers in Midwestern states such as Texas, Nebraska, Montana, and South Dakota could take advantage of abundant renewable energy and lower water stress.
Even under best-practice scenarios, residual carbon emissions may require offsets equivalent to 28 GW of wind or 43 GW of solar power. Researchers warn that without careful planning, AI server growth could substantially increase environmental pressures, highlighting the urgent need for collaboration between governments, tech companies, and utilities to align AI expansion with sustainability goals.
The findings were also covered by New Scientist, emphasising that the AI sector’s net-zero ambitions for 2030 are unlikely to be met without coordinated action on energy efficiency, renewable deployment, and transparent impact reporting.
Read the full paper in Nature Sustainability: Environmental impact and net-zero pathways for sustainable artificial intelligence servers in the USA