Amazon has, for the first time in a level of detail that’s likely to matter to regulators and communities, put a number on how much water its data centers used last year. In a new disclosure tied to its sustainability reporting, the company says its global data center operations consumed 2.5 billion gallons of water in 2025. The figure comes with a second metric—water use per unit of electricity—intended to help readers compare efficiency across time and, potentially, across companies.
The timing is hard to ignore. Seattle recently enacted a one-year moratorium on new data centers, a move that was supported by some local officials and pushed for by employees inside Amazon itself. That pause is part of a broader debate that has been intensifying as AI workloads drive rapid expansion of cloud infrastructure. In that debate, water has become a central concern alongside energy demand: not just because data centers require cooling, but because the water they use can be drawn from local systems that are already under stress, especially during hot seasons or drought conditions.
Amazon’s disclosure attempts to address both the scale of the problem and the question of whether the industry is improving. According to the company, its data centers used water at a rate of 0.12 liters per kilowatt-hour of electricity. It also claims that this represents a two percent decrease compared with 2024, even as it expanded operations. In other words, Amazon is arguing that growth in capacity did not translate into proportional growth in water intensity.
That “water intensity” framing is important, because raw totals can mislead. A company can increase total water use simply by adding more servers, more regions, or more facilities. But if water per unit of electricity declines, it suggests improvements in cooling systems, operational practices, or the mix of technologies and locations. Amazon’s report is essentially asking readers to look past the headline number and focus on the efficiency trend.
Still, the headline number is enormous. Two and a half billion gallons is the kind of quantity that forces a different conversation than one about incremental efficiency. It’s the difference between “a metric” and “a resource.” For communities near data center sites, the question isn’t only whether a company is slightly better than it was the year before; it’s whether the local water supply can handle the demand, and whether the benefits of hosting large-scale infrastructure outweigh the environmental and public-service costs.
To understand why this disclosure lands now, it helps to look at what’s driving data center growth. AI training and inference workloads have increased the appetite for compute, and compute requires power. Power, in turn, creates heat. Heat must be managed, and cooling is where water often enters the picture—either directly through evaporative cooling systems or indirectly through processes that still depend on water availability. Even when a facility uses air cooling rather than water-intensive methods, the broader ecosystem of equipment manufacturing, maintenance, and regional utility operations can still connect back to water use somewhere in the chain. That’s why water metrics are increasingly treated as a proxy for the physical footprint of digital services.
Amazon’s reported rate—0.12 liters per kilowatt-hour—also signals that the company is trying to align with how regulators and researchers increasingly evaluate sustainability claims. Electricity-based normalization is a way to compare across different operating conditions. If two data centers consume different amounts of electricity for different workloads, comparing total water use alone doesn’t tell you much. But water per kilowatt-hour offers a common denominator: it ties water consumption to the energy required to run the compute.
Even so, the metric doesn’t eliminate all uncertainty. Water use can vary widely depending on climate, local water sources, cooling design, and how much of the facility’s cooling system relies on evaporation versus recirculation. It can also vary based on how much of the year the facility runs at higher cooling loads. A single annual average can smooth out seasonal spikes that might be most relevant to local water managers. That’s one reason moratoriums and permitting debates often focus on worst-case or peak-demand scenarios rather than annual averages.
Amazon’s disclosure also raises a practical question: what does “global data center operations” include? Companies sometimes define their water accounting differently—whether it covers only owned and operated facilities, includes leased sites, counts certain types of water withdrawals but not others, or includes water used for specific processes beyond cooling. Amazon’s statement, as reported, is framed as a global total and an efficiency rate, but the details of boundaries and methodology are what determine how comparable the numbers are to other companies’ disclosures.
The Verge’s reporting around the disclosure points to Amazon’s own sustainability materials and notes that the company is presenting the information in a way that suggests it wants to be benchmarked. Amazon also claims it is using water more efficiently than some Big Tech rivals, referencing a graphic in its report. That kind of comparison is likely to be scrutinized, because benchmarking can be persuasive—or misleading—depending on whether the underlying assumptions match. If one company’s metric includes more facilities, uses different cooling technologies, or counts water differently, the comparison may reflect accounting choices as much as operational performance.
But even with those caveats, the disclosure is still a meaningful step. For years, water use has been discussed in the abstract—cooling towers, drought concerns, and the tension between digital growth and environmental limits. Yet many companies have been reluctant to publish consistent, detailed water metrics at the same cadence as energy metrics. Energy reporting is now relatively standardized across the industry, and investors and policymakers have built frameworks around it. Water reporting has lagged behind, partly because it’s harder to measure consistently across diverse geographies and partly because water is politically sensitive in ways that energy sometimes isn’t.
Amazon’s decision to publish a number like 2.5 billion gallons for 2025 suggests it recognizes that the conversation is no longer theoretical. When cities consider moratoriums, they need data. When communities ask for transparency, they need more than assurances. And when regulators weigh permits, they need to understand not only whether a facility will use water, but how much and how efficiently.
There’s another layer to this story: the relationship between water and energy is becoming a policy issue, not just an engineering one. Cooling systems can be designed to reduce water use, but those designs may increase energy consumption. Conversely, designs that minimize energy might rely more heavily on water. That tradeoff means that sustainability strategies can’t treat water and energy as separate problems. They have to be optimized together, and the optimization depends on local conditions—grid mix, ambient temperatures, water availability, and regulatory constraints.
Amazon’s reported decline of two percent in water intensity from 2024 to 2025, despite expansion, is therefore a signal about how it is managing that tradeoff. If the company is indeed improving water efficiency while scaling up, it suggests that new facilities or upgrades are incorporating better cooling approaches, improved controls, or more efficient water management systems. It could also mean that the company’s growth is shifting toward locations or designs that are less water-intensive per unit of electricity.
However, the decline is modest. A two percent improvement over a year is progress, but it doesn’t automatically resolve the underlying concern that total water use remains extremely high. In other words, even if water intensity improves, the absolute demand can still rise as capacity expands. That’s why moratoriums and permitting restrictions often focus on total impact and local thresholds rather than on efficiency alone.
This is where the “unique take” on the story becomes important: the real issue may not be whether Amazon is improving, but whether the industry’s improvement pace matches the pace of expansion—and whether local governance can keep up. Data centers are being built quickly, and water systems are not always able to expand at the same speed. Even if a company is more efficient than before, the cumulative effect of multiple projects in a region can overwhelm local infrastructure or create political backlash that slows future development.
Seattle’s moratorium is a case study in how governance responds when the math becomes uncomfortable. When a city pauses new data center approvals, it’s often because it needs time to assess impacts, update planning models, and ensure that utilities can handle additional load. Water is part of that assessment, but so is energy demand, grid capacity, wastewater management, and the broader strain on municipal services. Amazon’s disclosure doesn’t directly change Seattle’s decision, but it provides context for why such decisions are happening and what kinds of metrics communities will likely demand going forward.
For policymakers, Amazon’s numbers offer a starting point for questions that go beyond the company. If one major operator can report a water intensity metric, then other operators can be asked to do the same. That could lead to more standardized reporting requirements, similar to how energy and emissions reporting have evolved. Over time, that standardization could make it easier to compare proposals during permitting: not just “how much water will this facility use,” but “how does that compare to best-in-class performance under similar conditions.”
For communities, the disclosure may also shift the tone of public debate. Instead of relying on estimates or worst-case assumptions, residents can point to a disclosed baseline. That doesn’t mean the debate ends—because local conditions still matter—but it changes the conversation from speculation to accountability. It also gives advocates a way to ask follow-up questions: Are the water sources renewable or reclaimed? Is the water withdrawn from stressed aquifers? How much is returned to the system, and in what quality? What happens during drought restrictions? How does the facility manage peak cooling demand?
Amazon’s disclosure, as described, doesn’t answer all of those questions in the excerpt available here. But by publishing a global total and an efficiency rate, Amazon has created a framework that invites those questions. That’s often the first step toward more detailed reporting: once a company publishes a metric, stakeholders can demand the next layer of transparency.
There’s also a strategic dimension. Publishing water use data can be seen as proactive risk management. As AI infrastructure expands, companies face increasing scrutiny from regulators, investors, and the public. Transparency can reduce the likelihood of sudden policy shocks—like moratoriums—by demonstrating that