By Abbas Nazil
The rapid growth of artificial intelligence (AI) is driving unprecedented demand for large-scale data centers, raising serious environmental concerns.
Cornell University researchers have quantified the environmental impact of AI infrastructure on a state-by-state basis, revealing that by 2030, current trends could release between 24 and 44 million metric tons of carbon dioxide annually.
This level of emissions is equivalent to adding five to ten million cars to U.S. roadways.
The study also found that AI data centers could consume between 731 and 1,125 million cubic meters of water per year, equal to the annual household water usage of six to ten million Americans.
These projections suggest that without intervention, the AI industry’s net-zero targets will be out of reach.
The research team, led by Fengqi You, Professor in Energy Systems Engineering at Cornell, combined financial, manufacturing, and location-specific data with AI-assisted modeling to estimate energy, water, and carbon footprints.
The study emphasizes that location is critical, as clustering facilities in water-scarce regions like Nevada and Arizona could strain local resources, while states such as Texas, Montana, Nebraska, South Dakota, and New York offer more sustainable options.
The researchers propose a roadmap for sustainable AI growth, including smart siting, faster grid decarbonization, and operational efficiencies.
Implementing these measures could reduce carbon emissions by approximately 73 percent and water use by 86% compared to worst-case scenarios.
Even with aggressive decarbonization, residual emissions will remain, requiring deployment of renewable energy and energy-efficient technologies such as advanced cooling and improved server utilization to further lower environmental impacts.
Cornell researchers warn that this decade’s infrastructure choices will determine whether AI accelerates climate progress or becomes a new environmental burden.
The findings were published in Nature Sustainability and supported by the National Science Foundation and the Eric and Wendy Schmidt AI in Science program, with international collaborators from Sweden, Canada, and Italy contributing to the research.
