Amazon data centres used 2.5 billion gallons of water: AI's hidden water footprint

According to The Verge's analysis of Amazon's latest sustainability report, the company's global data centre network used about 2.5 billion US gallons — more than 9.5 billion litres — of water last year. The number has significantly raised attention to hyperscale data centre water use and put on the agenda the concern that, in the age of artificial intelligence, the figure will only grow.

Why do data centres use water? The first and largest use is cooling. In modern hyperscale facilities the heat produced by thousands of running servers is removed through evaporative cooling towers. Evaporated water carried by air-flow keeps equipment at safe temperatures. The second use is indirect water consumption in electricity generation — Amazon's sustainability report does not add that figure to direct consumption.

The Verge report puts the 2.5 billion gallon figure into context. Compared with an average American's yearly household use (around 33,000 gallons), the figure is roughly equivalent to 75,000 households' annual consumption. A single Amazon data centre campus can use between one and three million gallons a day, approaching half the daily consumption of some cities.

The geographic distribution matters too. Amazon Web Services' fastest-growing data centre regions include Virginia (especially Loudoun County), Oregon, Arizona and Ohio. Some of these regions — Arizona in particular — sit in a climate of high water stress. The Verge reports that local water authorities in Arizona have begun monitoring more closely the consumption commitments of Amazon and other hyperscale operators.

The artificial intelligence factor magnifies the question. GPU-intensive AI training workloads produce three to five times more heat per hour than traditional enterprise compute workloads. That means more water is needed to carry the cooling load. A GPT-class large language model training run can use thousands of gallons inside a data centre; per query, water use ranges from about 16 millilitres to 50 millilitres.

Amazon's response is multi-pronged. The company has committed to becoming "water positive" by 2030 — returning as much water to communities as it consumes. The target is met through rainwater harvesting, wastewater recycling, wetland restoration and local water infrastructure projects. Amazon is also accelerating the move to hardware techniques that lower water use, such as direct-to-chip liquid cooling.

What are the competitors doing? Google's March 2026 sustainability report disclosed about 6.1 billion gallons, while Microsoft reported about 1.7 billion gallons. All three use different measurement methodologies; The Verge notes that the industry has not yet agreed on a common standard.

The regulatory framework remains in early stages. In the United States, some states — Virginia and Oregon in particular — have started requiring water-use permits for new data centres. The European Union's Corporate Sustainability Reporting Directive CSRD requires large companies to report water consumption figures using a standard methodology starting in 2025. In Türkiye, the Ministry of Industry has begun a study on data centre water use, but reporting is not yet mandatory.

Long-term technical solutions exist. Liquid cooling removes heat from GPUs directly via liquid-contact plates and does not require evaporation. Free cooling — using outside air — provides large savings in cooler climates year-round. Geothermal heat pump integrations are being tested in some pilot projects. If these technologies become widespread, they could reduce data centre water intensity by 50 to 80 per cent.

The practical take-away for Vesper readers is that there is a water cost to using artificial intelligence and that cost, hidden behind any ChatGPT or Claude query, is real. For Türkiye's growing data centre investments — particularly in Istanbul and Ankara — water infrastructure planning should become a priority. The industry agreeing on a common methodology and reporting transparently is a successful starting point for the sustainability conversation.