Amazon just disclosed how much water its data centers drink, and the timing is no accident. According to The Verge AI, the company says its global data center operations used 2.5 billion gallons of water in 2025, reportedly the first time it has shared that figure publicly. The disclosure landed right after Seattle passed a one-year data center moratorium that some of Amazon’s own employees had pushed for.
What stands out here is that a Big Tech giant is suddenly volunteering numbers it used to keep quiet. That tells you how much pressure the AI build-out is putting on local water and power debates.
The Numbers at a Glance
- Total: 2.5 billion gallons of water consumed in 2025.
- Efficiency rate: 0.12 liters per kilowatt-hour of electricity.
- Trend: down 2 percent from 2024, even as Amazon expanded operations.
- Cooling mix: air cooling about 90 percent of the time, with evaporative water cooling saved for the hottest hours of the hottest days.
- Amazon’s claim: its data centers are seven times more water-efficient than the industry average.
That seven-times figure comes from an adjusted number based on a peer-reviewed paper released last year, The Verge AI reports. Worth keeping in mind when you see it quoted.
The Comparison to Rivals Deserves a Closer Look
Amazon’s report includes a graphic positioning itself ahead of Microsoft, Google, and Meta, showing each rival using more water per kilowatt-hour over recent years. Google came out worst by a wide margin.
Here’s where the framing matters. The Google data appears to focus specifically on Gemini AI data centers, while Amazon is reporting across all of its operations. That’s not an apples-to-apples comparison. A targeted AI workload and a company’s entire fleet are different measurements, so the gap looks bigger than a clean side-by-side would show.
What the Figure Leaves Out
The 2.5 billion gallons covers direct water use at Amazon’s facilities. It does not capture:
- Indirect water use at the power plants generating the electricity these data centers run on.
- Water consumed during new data center construction.
Those are real costs that don’t show up in the headline number. For a sector expanding as fast as AI infrastructure, the indirect footprint can rival or exceed the direct one, which is exactly the blind spot critics point to.
Why This Matters
Water and energy have become the central friction points in nearly every new data center fight. Communities are pushing back, and Seattle’s moratorium shows the resistance is now coming from inside the companies too. This is significant because transparency is becoming a competitive and political necessity, not a nice-to-have.
Amazon’s move does two things at once. It gets ahead of regulators and activists by putting a number on the table, and it reframes the conversation around efficiency rather than total consumption. A 2 percent drop while growing sounds good. The absolute figure, 2.5 billion gallons, is still enormous.
For practitioners and operators, the signal is clear:
- Expect more public reporting. Once one hyperscaler discloses, the others face pressure to match it.
- Watch the methodology. How a company defines water use shapes the story as much as the data itself. Whole-fleet versus single-workload numbers aren’t comparable.
- Cooling strategy is now a public metric. Air cooling, higher heat tolerances, and limited evaporative use are becoming selling points, not just engineering choices.
What Comes Next
Microsoft, Google, and Meta now have to decide whether to respond with their own full-fleet numbers or let Amazon’s framing stand. I’d expect counter-disclosures, especially from Google, given how it was portrayed. Meanwhile, local moratoriums like Seattle’s could spread, and water reporting may shift from voluntary PR to something closer to a regulatory expectation.
The bigger question the disclosure doesn’t answer: what happens to these numbers as AI workloads keep scaling? Efficiency gains are real, but they’re racing against demand that’s climbing far faster. Full details are available at the original report from The Verge AI.