Amazon's global data center operations consumed 2.5 billion gallons of water in 2025 at a rate of 0.12 liters per kilowatt-hour of electricity, the company disclosed for the first time. Usage fell 2% from 2024 even as Amazon expanded its fleet, and the company claims its data centers run seven times more water-efficiently than the industry average. The disclosure lands days after Seattle enacted a one-year moratorium on new data centers, a measure some Amazon employees had publicly backed.
The 0.12 L/kWh figure is the metric Amazon wants the debate to center on. In a graphic accompanying the report, Amazon compared itself to Microsoft, Google, and Meta and showed each using more water per kilowatt-hour over recent years. Google posted the highest number on the chart by a wide margin.
There is a caveat embedded in that comparison. The Google data Amazon cites appears to be specific to Gemini AI data centers, while Amazon's own number covers all of its operations — a denominator difference that flatters the AWS-wide figure against a workload-specific rival number. Microsoft, Google, and Meta have not responded to the framing.
Key facts
- 01Amazon's data centers used 2.5 billion gallons of water in 2025, the company's first such disclosure.
- 02Water intensity hit 0.12 liters per kilowatt-hour, a 2% drop from 2024 even as Amazon expanded operations.
- 03Amazon says air cooling handles about 90% of operating hours, with evaporative cooling reserved for the hottest days.
- 04The company claims its data centers are 7x more water-efficient than the industry average.
- 05The disclosure follows Seattle's one-year data center moratorium, which some Amazon employees backed.
Cooling design explains most of the gap Amazon is claiming. Amazon says its data centers use air cooling roughly 90% of the time and switch to evaporative water cooling only during the hottest hours of the hottest days. The company has also raised the temperature tolerances of its servers, which lets racks run hotter before water-based cooling kicks in.
“about 90 percent of the time”— Amazon, company statement
The 7x efficiency claim is sourced to an adjusted figure from a peer-reviewed research paper released last year, not an independent audit of Amazon's 2025 operations. That makes it a useful benchmark, not a verified one. Amazon has not published the underlying site-by-site data that would let outside researchers reproduce the 2.5-billion-gallon total or the 0.12 L/kWh intensity figure.
The disclosure also has known blind spots. Amazon's number does not include indirect water consumption at the power plants generating electricity for its data centers, which for thermoelectric generation can dwarf on-site cooling water. It also excludes water used during the construction of new facilities, a non-trivial figure given the pace of AWS buildout.
Context matters for why Amazon is publishing this now. Local opposition to hyperscale data center siting has hardened over the past year, from Seattle's moratorium to disputes in places like Shelbyville, Indiana. Water use, alongside grid load, has become the specific number residents and council members ask about, and until this week Amazon was the largest US hyperscaler without a public figure to offer.
The competitive subtext is that AI training and inference workloads are the marginal driver of new water demand, and Amazon is positioning AWS as the lower-impact place to run them. That argument lands differently depending on whether a reader trusts the methodology. The 2% year-over-year decline in absolute water use, achieved while capacity grew, is the strongest single data point in Amazon's favor — efficiency gains are outpacing growth, at least on this metric.
Skeptics will note what the report does not address. Indirect water from upstream power generation can swing total footprint by an order of magnitude depending on grid mix, and Amazon's heavy reliance on evaporative cooling during peak heat means the water-use spike correlates with exactly the drought conditions communities worry about. A single annual average smooths over that seasonal concentration.
For the AI infrastructure market, the disclosure sets a new floor for what hyperscalers are expected to publish. Once one major cloud provider releases a fleet-wide water number, the others face pressure to match the format or explain why theirs looks worse. Expect Microsoft, Google, and Meta to respond with their own disclosures structured to make their numbers competitive — and expect the next round of local siting fights to be argued in liters per kilowatt-hour rather than in gallons.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
