Amazon's data centers pulled 2.5 billion gallons of water globally in 2025 — a number that evaporates next to the 117 trillion gallons withdrawn in the US alone back in 2015.
That 2.5 billion figure comes straight from Amazon's own blog post, and it's the kind of headline number that sends AI-skeptic corners of the internet into a frenzy. But scale matters. Stack it against 3.3 trillion gallons used annually on US lawns and landscaping, or the 1.3 trillion gallons that California almond orchards drink every year, and AI data centers start looking like a leaky faucet in a reservoir.
The Numbers That Put AI's Thirst in Perspective
Amazon is a latecomer to reporting water usage, but it's not alone. Google's data centers withdrew 6.1 billion gallons in 2024. Microsoft pulled 2.75 billion gallons that same year. Meta? Another 1.4 billion gallons. Add it all up and you get roughly 12.75 billion gallons across four of the biggest operators — still less than the 531 billion gallons the US pumps into golf courses annually.
These are withdrawal numbers, not consumption. A good chunk of that water returns to the source after cooling cycles, though evaporative cooling does lose some to the atmosphere. Still, the aggregate footprint is dwarfed by agriculture, which accounts for roughly 80% of US water consumption.
Individual Stress vs. Aggregate Drops
None of this means local water stress is imaginary. A single hyperscale data center in a drought-prone region can strain municipal supplies — just ask residents of The Dalles, Oregon, or Mesa, Arizona. But the global narrative that AI is somehow a primary driver of water scarcity doesn't hold up to arithmetic.
Amazon's blog post is careful to note location-specific mitigation strategies, including recycled water and adiabatic cooling that cuts evaporation. Google and Microsoft have similar programs. The real story isn't the total volume — it's where those gallons come from and who else is drawing from the same aquifer.
Why This Matters for Local Communities
If you're building a power-hungry AI cluster, the water conversation should start with your site selection and cooling design, not with a global guilt trip. The numbers prove AI data centers are a rounding error in the US water budget, but a single ill-placed facility can be a crisis for a small town.
Amazon, Google, Microsoft, and Meta are now reporting their water withdrawals with enough granularity to make those local impacts visible. The next step is transparent reporting on consumption versus withdrawal, and binding commitments to use reclaimed or brackish water in water-stressed regions.
AI's water footprint isn't the problem. The problem is where AI puts its feet.
Source: When it comes to total water use, AI data centers are a drop in the bucket
Domain: arstechnica.com
Comments load interactively on the live page.