AI data centers' water consumption raises concerns in drought-prone regions

Data centers for AI consume significant water for cooling, but overall usage is less than 1% of national consumption. However, rapid expansion could strain water resources in arid areas like New Mexico and Arizona. Engineers are exploring ways to reduce water use and potentially achieve net-zero water for cooling.

Data center cooling systems rely on evaporative processes similar to human perspiration, with processors reaching internal temperatures of 176°F. While 2023 consumption stood at 66 billion liters—under 1 percent of national totals—projections suggest that figure could reach 1,125 billion liters annually by 2030, roughly equivalent to New York City's yearly drinking water supply.

Efficiency gains are already emerging. Google estimates a median Gemini query now requires only five drops of water, a sharp decline from earlier GPT-4 estimates of 500 milliliters per email. Researchers also point to renewable energy adoption, processor improvements, and careful facility placement as measures that could substantially reduce future water demand.