HeadlinesBriefing favicon HeadlinesBriefing.com

How Much of a Problem Is AI's Water Use?

Ars Technica •
×

Protests against data center expansion are rising from Texas to Arizona over water use and energy consumption. Data centers use water evaporation to cool processors reaching 176° Fahrenheit, but consumption data from tech companies remains inconsistent. Estimates range from 500 ml per chatbot query to Google's 2025 claim of just five drops for a Gemini query. Researchers calculated data center cooling consumed 66 billion liters in 2023—under 1% of U.S. total—but Cornell's Fengqi You predicts this could hit 731–1,125 billion liters annually by 2030, roughly matching New York City's drinking supply.

While agriculture and manufacturing use far more water overall, local stress is acute in drought-prone regions like New Mexico and Arizona. Experts including UC Riverside's Shaolei Ren note AI efficiency gains are rapidly outdated; Ren's 2024 GPT-4 estimate is already obsolete. A 2025 Nature Sustainability study projects that shifting to renewables, improving processor efficiency, and strategic siting could significantly curb growth. You adds that replenishment efforts could eventually achieve net-zero water for on-site cooling.