Overview
- Research by Alex de Vries-Gao estimates AI systems in 2025 generated 32.6–79.7 million tons of CO2 and used 312.5–764.6 billion liters of water, with power demand projected around 23 GW that exceeds 2024 Bitcoin mining.
- The paper, published in Patterns, relies on public filings, earnings calls, and analyst estimates because companies rarely break out AI-specific energy or water use, a gap experts say likely makes the totals conservative.
- Most of a facility’s water footprint stems from electricity generation as well as cooling, and roughly 45% of data centers operate in river basins already facing high water stress, according to the reporting.
- More than 230 environmental groups urged Congress on December 8 to impose a national moratorium on new data-center construction, citing pollution, grid strain, and local water impacts tied to the AI buildout.
- U.S. grid analyses from DOE and NERC warn of resource adequacy risks as data-center load accelerates, while operators pursue firm power through on-site generation and backup, including Tesla’s gas turbines near Memphis and a planned nuclear restart in Iowa to supply Google.