AI servers could use ten percent of U.S. electricity by 2030

A new international study is one of the first to detail AI’s energy needs along with its water and carbon footprint. It shows that, if current trends continue, AI operations could generate 24–44 million tons of additional carbon emissions and consume 731–1125 million cubic meters of water each year.

KTH researcher Francesco Fuso-Nerini, Director of the KTH Climate Action Centre, is one of the authors.

“What we clearly saw was that the energy demand from AI servers will increase significantly, regardless of the scenario. The decisive factor is how the energy is produced—if it comes from fossil-free sources, emissions will be limited. Otherwise, the climate impact could be substantial,” says Fuso-Nerini.

Better conditions in Sweden

Even with rapid efficiency improvements, it is unlikely that the sector will reach net-zero emissions by 2030 without extensive use of carbon offsets and water restoration credits. In the best case, the remaining emissions would still require the equivalent of about 28 gigawatts of wind power or 43 gigawatts of solar power to be neutralized.

“In Europe and Sweden we see a similar trend with rapidly growing demand and more server facilities. The differences mainly concern where the servers are located. For example, data centers in northern Sweden, where there is good access to renewable energy, can have significantly lower climate impact than facilities in parts of Europe that are still more dependent on fossil fuels,” adds Fuso-Nerini.

Rapid transition to renewable energy needed

The study highlights that efficiency measures can make a big difference. By combining advanced liquid cooling and optimized server usage, water consumption can be reduced by about 32 percent.

The pace of renewable energy expansion is crucial—a faster transition to clean electricity can reduce emissions by over 15 percent. A slower pace could increase them by about 20 percent.

“AI has many applications in society, from improved healthcare to accelerating the transition to clean energy, but it also has a real footprint in energy, water, and climate. Our research shows that smart planning, efficiency, and clean energy are essential for AI to truly support a sustainable future,” says Fuso-Nerini.

Recommendations from the study

• Build where it is sustainable: Prioritize regions with abundant renewable energy and low water stress to minimize both carbon and water impact.
• Invest in efficiency: Use advanced cooling technology and smarter server management to improve energy and water performance.
• Accelerate clean energy: Expand renewable energy production and modernize the grid to ensure AI’s growth is powered by low-emission electricity.
• Ensure transparency: Verify and report the real impact of carbon and water offset programs to maintain credibility and accountability.
• Collaborate broadly: Governments, tech companies, and energy providers should coordinate infrastructure and sustainability planning to balance innovation with environmental limits.

The study was conducted by an international team of researchers from KTH, Cornell University, University of Oxford, Concordia University, RFF-CMCC European Institute on Economics and the Environment, and Politecnico di Milano.

Xiao, T., Nerini, F.F., Matthews, H.D. et al. Environmental impact and net-zero pathways for sustainable artificial intelligence servers in the USA. Nat Sustain (2025). doi.org/10.1038/s41893-025-01681-y