AI’s Thirst for Water

Artificial intelligence offers both benefits and risks, but its growing demand for one fundamental resource, water, can create serious consequences for nearby communities.

In May 2026, Alexandria Ocasio-Cortez, also known as AOC, presented an EPA official with jars of brown water that she said had been collected from Morgan County, Georgia, following the construction of a Meta data center. This was more than a scare tactic. It was a powerful visual that revealed how little many people understand about the environmental costs associated with artificial intelligence.

Residents have raised concerns about whether nearby data center development contributed to low water pressure and discolored water. Meta has disputed that connection, and the EPA has said that gathering information about the complaints does not constitute a formal investigation. Still, the controversy points to a larger issue. While many people see AI as an exciting technological innovation, the infrastructure behind it can place pressure on one of the most fundamental resources for life: water.

The Water Behind AI

Training and operating large AI models requires dense computing infrastructure. When AI systems run inside data centers, the equipment generates heat. Cooling systems are needed to prevent overheating, and many of these systems require water.

According to the Environmental and Energy Study Institute, a medium-sized data center can consume up to roughly 110 million gallons of water per year, an amount equivalent to the annual water use of approximately 1,000 households. This figure is astronomical, especially in areas already experiencing water scarcity.

As CSU Global puts it, “[T]he benefits of AI are widely distributed. The costs are often local.” Some data centers use closed-loop systems that allow water to be reused multiple times. However, these systems do not eliminate water use entirely, and the effect of a data center depends heavily on its design, location, climate, and water source. Communities already facing water scarcity may not have the capacity to lose even a portion of their available water.

When Infrastructure Competes With Communities

The water presented by AOC is one example of the concerns surrounding the effects of data centers on local water systems. Although the cause of the discoloration in Morgan County has not been conclusively established, communities across the country are asking whether the expansion of AI infrastructure could threaten their access to water.

Historically, many data centers have been constructed in urban areas, although new development is increasingly shifting toward rural communities. Locating data centers near users can reduce latency, but placing water-intensive infrastructure in populated or water-stressed areas can create competition between digital infrastructure and traditional water users.

According to UN-Water, rapidly growing urban communities already face challenges involving water access, sanitation, population growth, and infrastructure. Large-scale water use by data centers can amplify these pressures. Communities that already need assistance should not face additional barriers to obtaining a fundamental resource.

According to the Pew Research Center, 87 percent of currently operating U.S. data centers are in urban areas. However, 67 percent of planned data centers are located in rural areas. This shift means the debate over water use is no longer limited to cities.

A $3 Billion Project in Virginia

In Botetourt County, part of Virginia’s Roanoke Valley, Google plans to invest $3 billion in a new data center campus by 2030. According to agreements with the Western Virginia Water Authority, the campus could eventually receive the capacity to use up to 8 million gallons of water per day. The authority initially expects to provide capacity of up to 2 million gallons per day beginning in 2028.

These numbers describe the maximum amount of water the system could provide, not necessarily the amount the data center will consume each day. Even so, the scale is difficult to conceptualize and has caused concern among some residents. Local officials have said the regional system currently has excess water capacity, while opponents worry that dedicating so much capacity to one project could affect residents in the future.

The project is expected to create jobs and generate tax revenue, but communities should not have to accept major infrastructure projects without clear information about their possible effects. Access to water is a basic human need, and residents deserve transparency about how much water a data center will actually withdraw, consume, reuse, and return.

The Cost of a Single Query

Every time someone uses artificial intelligence for work, school, or a hobby, the computing hardware behind that request consumes energy. One query may seem harmless, but millions of daily interactions collectively create substantial demand.

According to MIT News, researchers have estimated that a ChatGPT query consumes about five times more electricity than a simple web search. The exact amount varies depending on the model, the length and complexity of the request, the hardware involved, and the source of electricity.

Data centers also need water for cooling. MIT reports an estimate of approximately two liters of cooling water for every kilowatt-hour of energy a data center consumes. A kilowatt-hour is the amount of energy used by 1,000 watts operating for one hour.

One analysis estimates that ChatGPT once processed approximately 200 million queries per day, with each query using about 0.0029 kilowatt-hours of electricity. Based on those assumptions, daily use would require roughly 621.4 megawatt-hours of electricity. That is approximately equal to the electricity 52 average American households use over an entire year.

However, these figures should be treated as estimates rather than fixed measurements. AI companies generally do not disclose enough data to calculate the precise energy and water cost of every query. Different models and data centers can have very different levels of efficiency.

Still, the broader point remains. A single AI request may use only a small amount of energy and water, but billions of requests, combined with model training and continued data center construction, can create a significant environmental footprint.

Can We Reduce the Damage?

Immense amounts of infrastructure have already been built, but future harm can still be reduced. AI has been integrated into many daily routines, sometimes without people even realizing it. Digital assistants such as Siri and Alexa use artificial intelligence. Social media platforms also rely on AI-powered algorithms.

To be frank, I believe people should avoid using AI when it is unnecessary. Its risks extend beyond the environment and include concerns about cyberattacks, employment disruption, misinformation, and other harms. We need technological progress, but we also need clean, safe drinking water and communities that can thrive.

Since AI has already been embedded into many daily habits, eliminating it entirely may not be realistic. However, people can minimize unnecessary use. One place to start is by reducing reliance on ChatGPT and other large language models for simple questions that a traditional search engine, trusted source, or book could answer. Users can also avoid AI overviews when they are unnecessary and reduce their use of digital assistants.

Individual choices alone will not solve the problem. Technology companies and governments also need to improve transparency, choose locations carefully, use reclaimed or non-potable water, adopt more efficient cooling systems, and power data centers with energy sources that require less water.

These are only a few ways to reduce AI’s environmental impact. Convenience comes with a cost, and communities deserve to know who is paying it.