Our Take
Amazon's comparison to Google and Meta omits indirect water use from power plants and construction—the real efficiency story remains incomplete.
Why it matters
Water consumption is now a regulatory flashpoint for data center expansion, especially in water-stressed regions. Seattle's moratorium and Amazon employee activism signal that infrastructure operators can no longer defer environmental disclosure.
Do this week
Infrastructure teams: request full Scope 3 water accounting (power generation + construction) from your cloud providers before committing to multi-year capacity contracts.
Amazon reports 2.5 billion gallons of water use across data centers
Amazon disclosed its global data center water consumption for 2025, reportedly for the first time, after Seattle enacted a one-year data center moratorium that some of the company's own employees supported. The company consumed 2.5 billion gallons of water at a rate of 0.12 liters per kilowatt-hour of electricity, a 2 percent decline from 2024 despite expanding operations (per Amazon's disclosure).
The timing is deliberate. Water-use scrutiny has intensified alongside AI data center buildout debates, forcing major cloud operators to publish efficiency metrics. Amazon claims its facilities use air cooling approximately 90 percent of the time and deploy evaporative cooling only during peak heat hours, while raising server heat tolerances. The company states it operates at seven times greater water efficiency than the industry average, citing an adjusted figure from a peer-reviewed research paper released in 2024 (company-reported).
Amazon also benchmarked itself against Microsoft, Google, and Meta in its report, showing each rival consuming more water per kilowatt-hour over recent years. Google consumed the most in the comparison graphic—though Amazon's data covers all its data center operations while the cited Google figure appears focused specifically on Gemini AI infrastructure (per The Verge's analysis of Amazon's report).
The efficiency claim omits a material cost category
Amazon's 0.12 liters-per-kilowatt-hour metric excludes indirect water use at the power plants supplying electricity to its data centers, as well as water consumed during new facility construction. These are not trivial. Power plant cooling accounts for the majority of water footprint in data center operations at many hyperscalers, and new AI facility buildout requires substantial water for concrete mixing and site preparation.
The peer-reviewed baseline Amazon cites (the "seven times more efficient" claim) used an adjusted denominator that may not be directly comparable to current fleet composition or construction-inclusive accounting. Without independent verification of Amazon's efficiency metric or full-scope accounting, the comparison to rivals rests on incomplete data sets.
For regulators and community groups, this is the gap. Amazon can claim progress on operational cooling while downstream water impact—and construction impact—remains outside the reported scope. Seattle's moratorium and employee pushback suggest that disclosure alone, without full accounting, will face ongoing pressure.
Request full water accounting before capacity commitments
If your team is evaluating multi-year data center contracts or planning infrastructure expansion, ask your cloud provider for Scope 3 water data: power generation plus construction. Single-metric efficiency claims (liters per kilowatt-hour) may mask the total water cost. Compare vendors using the same scope boundaries, not cherry-picked metrics. In water-stressed regions, this distinction now affects permit timelines and community approval, making it a business-risk variable, not just an environmental metric.