You’ve probably seen the headline: one AI-generated email costs a whole bottle of water. It’s a stat that gets clicks. But the real picture of AI water consumption is more nuanced, and the question of what you can do about it deserves a straight answer.

The Big Picture Is Alarming

U.S. data centers consumed roughly 21.2 billion liters of water in 2014. By 2023, that tripled to approximately 66 billion liters, according to the Environmental Law Institute. Projections suggest that figure could double or quadruple by 2028. At the global level, a peer-reviewed study published in late 2025 estimated AI’s water footprint reached 312.5 to 764.6 billion liters that year, comparable to the world’s total annual bottled water consumption.

The real-world impact is concrete. The Lincoln Institute of Land Policy reports a single Meta data center in Newton County, Georgia consumes 500,000 gallons per day, 10% of the county’s entire water supply. Bloomberg analysis shows more than 160 new AI data centers were built in high water-stress areas in just three years.

But Your Individual Queries? Not the Villain

Here’s the rundown. Google’s own measurement found each Gemini text prompt consumes about 0.26 milliliters of water, roughly five drops. A single ChatGPT query uses about the same energy as watching TV for three minutes. And producing one hamburger requires over 600 gallons of water, per Warp News analysis, while data centers use about 500 ml per 300 AI queries. A study in Scientific Reports even found that the carbon emissions of writing and illustrating are lower for AI than for a human completing the same task, once you factor in commuting, office energy, and overhead.

What You Can Actually Do

The fight is at the infrastructure level. Google reported a 33x reduction in energy per Gemini prompt over just 12 months through efficiency improvements. Microsoft’s new closed-loop cooling design eliminates water evaporation entirely, saving over 125 million liters per data center annually. Although the trade-off is it’s a higher energy consumption model to create a closed-loop cooling design.

As individual users, the most impactful moves are systemic. Choose AI platforms committed to renewable energy and published sustainability metrics, as recommended by Schellman’s sustainability practice. Use AI intentionally, ask whether each query produces value proportional to its cost. And push for transparency: the UNEP recommends standardized environmental disclosures from AI companies. Consumer pressure matters.