Environmental impact and net-zero pathways for sustainable artificial intelligence servers in the USA

This peer-reviewed study models the combined energy, water, and carbon impacts of AI server deployment across the United States from 2024 to 2030. It projects annual water footprints of 731–1,125 million m³ and additional carbon emissions of 24–44 Mt CO₂-equivalent, finding that the AI industry is unlikely to meet net-zero goals by 2030 without substantial reliance on carbon offsets and water restoration. The authors recommend concentrating deployment in Midwestern states with abundant renewables and low water stress, and call for coordinated regulatory and industry action.