Amazon says AI data centers used 2.5B gallons in 2025, but local strain is real

For years, AI data centers have been framed online as water monsters. But Amazon’s new reporting suggests the aggregate story is a lot smaller than the memes make it sound: the company says its data centers withdrew about 2.5 billion gallons globally in 2025.

That number is “incredibly large” to the brain’s spreadsheet instincts, but it drops into perspective fast. It sits tiny against the 117 trillion gallons of water withdrawn in the US alone in 2015, and it looks even smaller compared with water-intensive benchmarks like the 3.3 trillion gallons used annually on US lawns and landscaping, the 1.3 trillion gallons a year used in California almond orchards, and 531 billion gallons a year used just for US golf courses. In other words, if your debate is “is AI water usage a bucket-sized national problem,” Amazon’s 2025 figure argues the answer is no.

Now for the part that keeps this story from being a total victory lap: “small in aggregate” does not mean “no pressure anywhere.” Even the source acknowledges what operators often learn the hard way, a modest-looking national or global number can still create outsized local impact. Data centers do not exist everywhere. They cluster, they draw from specific municipal or regional systems, and they tend to expand in places where infrastructure, permitting, and water availability already have limits. So yes, Amazon’s 2.5 billion gallons can be a drop in the bucket overall while still contributing to strain in particular areas.

This is also why Amazon matters here beyond one datapoint. It is one company, and it is described as a relatively latecomer to reporting its data center water usage numbers. That timing matters because the political and regulatory framing is often driven by what is measurable. When reporting is sparse, worst-case assumptions fill the vacuum. When reporting arrives, even if it does not erase local concerns, it forces decision-makers to separate “global narrative” from “site-specific reality.”

Amazon is not the only player building its own water footprint disclosure. The source notes that Google data centers withdrew more than 6.1 billion gallons of water in 2024. It also says Microsoft withdrew about 2.75 billion gallons and Meta withdrew about 1.4 billion gallons in the same year. Put those side by side and you get a pattern that executives should recognize: major AI infrastructure providers are tracking and publishing these numbers, but not in a single harmonized way that makes comparisons painless.

That mismatch creates a board-level problem. Water use can drive permitting decisions, community opposition, and operational constraints, but it often arrives in the form of fragmented disclosures. Executives cannot manage what they cannot benchmark, and boards are increasingly asked to oversee ESG risks that include utilities, water, and cooling. Even if AI data centers are a “drop” compared to certain national baselines, the local water story can still become the trigger for delays, compliance costs, or redesigns, especially when expansions run through constrained regions.

There is also a strategic implication for how companies talk about cooling. Much of the public anxiety centers on evaporative cooling, and the source references that backdrop. But aggregate comparisons can shift the debate from “AI is inherently water-wasteful” toward “cooling method and siting determine real-world impact.” That reframe matters because it changes what governance teams and real estate teams should prioritize: not just whether water is used, but how much, where, under what scarcity conditions, and with what mitigation plan.

For executives, the stakes are straightforward. Your peers may be able to defend AI data centers as small relative to national withdrawal totals, yet still lose permitting fights if a single facility, expansion, or nearby community water demand becomes the point of leverage. Amazon’s 2025 figure gives you ammunition for the aggregate debate, but it does not replace the operational homework. If the industry is going to keep scaling, companies will need credible water accounting, transparent reporting, and serious attention to local supply, because that is where “small in the bucket” can still turn into “big on the ground.”