AI Data Centers Consumed 222 Billion Liters of Water in 2025 as Tech Giants Deploy New Cooling Tech

AI Water Consumption Reaches 222 Billion Liters Amid Growing Public Backlash

AI data centers consumed 222 billion liters of water worldwide for cooling in 2025, according to Rystad Energy2

2

3

3

. Without adaptive measures, that figure could nearly triple to 644 billion liters by 20302

2

. This surge in AI's water footprint has triggered more than 130 protests across dozens of US states, with demonstrators carrying signs declaring "Water for people not AI" and "Don't mess with our water"1

1

. A Heatmap survey found that three-quarters of Americans now oppose new data center construction in their communities4

4

. The issue has become a talking point in upcoming midterm elections as activists challenge AI's environmental impact.

Tech Giants Deploy Closed-Loop Cooling to Address Water Crisis

Nvidia claims its DSX system can eliminate AI water consumption almost entirely at some facilities through closed-loop cooling2

2

3

3

. The system flows liquid directly through servers at 45°C, warmer than the typical 32°C used in most closed-loop systems in 20242

2

. "Simple fans circulating the air" are often sufficient, though extreme climates may still require evaporative cooling, said Josh Parker, Nvidia's head of sustainability2

2

. Microsoft, Amazon Web Services, and Meta also use closed-loop cooling systems involving no net water loss3

3

. Between 2022 and 2025, Microsoft improved water-use efficiency by 25% while AWS achieved 37% improvement2

2

, despite both hyperscalers using more water overall as they expanded operations.

The Trade-Off Between Water and Electricity Consumption

"There's a pretty direct trade-off between how much water is used and how much energy is used" for temperature control, said Andy Masley, an independent researcher covering AI data centers2

2

. Liquid cooling reduces water consumption but requires more electricity to cool the sealed pipes, typically by blowing air over them. Data centers consumed roughly 4.4% of all electricity in the United States in 2023, a figure projected to triple by 20284

4

. Globally, data center power consumption is on track to double by 2030, reaching an amount equivalent to Japan's entire annual electricity consumption4

4

. A single large data center can consume up to 5 million gallons of water daily, equivalent to a city of 50,000 people4

4

.

Renewable Energy and Strategic Siting Could Cut AI's Water Footprint by 86%

Shifting AI data centers from coal or gas to renewable energy could significantly reduce AI water consumption, according to Fengqi You of Cornell University1

1

. Much of AI electricity consumption comes from burning fossil fuels, with water used to cool steam after spinning electricity-generating turbines. Solar and wind power require little to no water to operate. Strategic siting in water-rich regions with ample renewable energy sources—like Montana, Nebraska, parts of Texas, and South Dakota—instead of water-stressed areas like Arizona, New Mexico, and Southern California could reduce AI's future water footprint by up to 86%1

1

. You's group predicted that by 2030, US AI data centers could consume 731 billion to 1,125 billion liters annually—the latter roughly equivalent to New York City's annual drinking water supply1

1

.

Per-Query Water Use Varies Wildly Across AI Applications

Google's 2025 study found that a median text prompt to Google Gemini used 0.24 watt-hours of electricity and 0.26 milliliters of water—roughly five drops5

5

. OpenAI boss Sam Altman claimed an average ChatGPT query uses 0.34 watt-hours and 0.32 milliliters of water5

5

. However, these figures mask enormous variation. Processing 100,000 tokens could require almost 40 watt-hours before generating a response5

5

. Image generation uses around 60 times more energy than text generation on average5

5

. Shaolei Ren of UC Riverside estimated in 2024 that drafting a short email with GPT-4 consumed 500 milliliters of water, though this figure is already outdated as AI models have become more efficient1

1

.

Hidden Water Costs and the Challenge of Measuring AI's Environmental Impact

Water consumed directly by data center cooling systems represents only part of AI's total water footprint. In the United States, indirect water use for generating electricity and manufacturing chips and servers can be twice the amount AI data centers consume directly2

2

3

3

. There is no industry-wide standard governing how companies report environmental, social, and governance efforts, making comparisons difficult2

2

. Elon Musk's SpaceX, now a major player in AI data centers after acquiring xAI, has never published an ESG report and received MSCI's lowest ESG score in June3

3

. "Because water is generally much cheaper than electricity," companies have less incentive to cut water use on cost grounds alone, said Ren2

2

. Incentives are driven more by public relations amid growing backlash.

Why Generative AI Defies Traditional Software Economics

Generative AI has shattered the traditional economics of software, forcing the technology sector to cope with new challenges in thermodynamics and resource scarcity4

4

. "Traditional software has very low marginal cost because computation happens primarily on the user's device or cheaper multitenant infrastructure," said Andrew Marshall of Yugabyte. "AI inference incurs a real computational cost for every interaction"4

4

. A single query to ChatGPT requires up to 10 times more electricity than a traditional Google search4

4

. Generating a five-second video using generative AI models consumes as much electricity as running a household microwave oven nonstop for over an hour4

4

. The energy and resource demands of AI are experiencing Jevons Paradox: even as processing becomes more efficient, demand increases to the point that total usage outweighs per-unit savings. Fitch Group estimates the top five hyperscalers will spend $750 billion on data center construction this year, with three-quarters earmarked for AI4

4

.