Amazon reveals data center water consumption for the first time: 2.5 billion gallons annually

Seattle approved a one-year moratorium on new data centers within its territory in early June, driven in part by Amazon employees who publicly pressured the company. Days later, Amazon published a figure it had previously kept away from public scrutiny: its global data center operations consumed 2.5 billion gallons of water last year.

The timing seems deliberate. The disclosure arrives just as debate over the environmental impact of AI infrastructure reaches unusual intensity in municipal forums, specialized media, and within tech companies themselves. According to The Verge, this would be the first time Amazon has shared this metric publicly.

What the number means

2.5 billion gallons equals just over 9.46 billion liters. For context: annual water consumption in a city like Madrid amounts to around 400 billion liters, so in absolute terms Amazon's figure represents approximately 2.3% of that urban demand. It is not a marginal volume, but it is also not straightforward to interpret without knowing what percentage goes to evaporative cooling, how much is reused, and which regions account for the heaviest consumption—information Amazon has not yet detailed.

What matters is not just the magnitude, but the pattern: major cloud providers have historically been opaque about these figures. Microsoft published its water metrics earlier, and Google has done so with varying levels of detail, but Amazon Web Services, the world's largest cloud infrastructure provider by market share, had until now avoided putting concrete numbers to this variable.

Why it matters now, and not before

Growth in computational demand associated with language models and massive inference workloads has intensified pressure on existing infrastructure. Modern data centers hosting high-density accelerators generate more heat per square meter than those from five years ago, and cooling them requires more water or more electricity, or both, depending on the cooling system used.

This context has led local administrations to treat new data centers like any other high-impact industrial facility: with environmental impact assessments, negotiations over water consumption, and in cases like Seattle, temporary moratoriums while regulatory frameworks are updated. That Amazon employees are among the drivers of this moratorium adds a layer of internal tension rarely seen so visibly in major technology companies.

Who should care about this figure

For engineering teams deploying proprietary infrastructure or evaluating cloud providers, this disclosure raises a practical question: how is that consumption distributed across regions? A global figure aggregates markets with abundant water and zones with severe water stress in ways that obscure real risks. A data center in the Pacific Northwest does not have the same impact as one in Arizona or southeastern Spain.

For those responsible for corporate sustainability policy, Amazon's disclosure can serve as a baseline reference: if the world's largest cloud operator had to publish its consumption under regulatory and social pressure, remaining providers, and their customers, will have harder justifications for opacity.

For municipal teams and regulators, the figure confirms that the industry's water transparency remains fragmented and that moratoriums like Seattle's can be an effective tool for forcing disclosures that would otherwise take years to materialize.

What remains unknown

Amazon has not specified what percentage of that water evaporates without possibility of recovery, which regions account for the heaviest consumption, or what water efficiency targets it has for coming years. Without these details, the figure functions more as a gesture of minimal transparency than as a real accountability tool.

From our perspective, the assessment is moderately positive: publishing the figure is better than not publishing it, but the pressure to obtain it should not have needed to come from a municipal moratorium. The industry has spent years building AI capacity without updating its environmental disclosure standards at the same pace.