https://www.eesi.org/articles/view/data-centers-and-water-consumption
How AI Data Centers Consume Water and Electricity
For simple tasks, such as rewording an email, OpenAI’s popular GPT-5 model can use as much electricity as a three-watt LED light bulb consumes in around 20 minutes, along with slightly less than a teaspoon of water. For one person sending one prompt, that amount may not seem significant.
OpenAI Receives 25 Billion Queries Every Day
OpenAI says it receives 25 billion queries every day. When we add up the resources required to process those requests, the total becomes enormous. The water consumption could meet the daily drinking needs of everyone in Monterrey, Mexico, while the electricity consumption could power more than 3,000 years of Netflix binge-watching.
Experts expect the number of AI users to roughly double over the next five years. As more people use AI, companies must build additional data centers to handle the growing demand.
Traditional Cooling Methods Keep Data Centers Running
Cooling has always played a central role in data center operations. Even some of the earliest computers in the 1940s used vacuum tubes that required fans and ventilation to prevent overheating.
Modern data centers look surprisingly similar in principle, despite offering computing power trillions of times greater than that of early computers. However, today’s systems consume far more energy, and cooling can account for nearly half of a data center’s total electricity use.
Hyperscale Data Centers Require More Cooling
Google’s data center in Oregon can perform the massive computations that AI applications require. A typical data center covers approximately 9,000 square meters, while this facility occupies an area about 13 times larger.
The industry calls facilities of this scale hyperscale data centers. According to Professor Scanlan from the University of Wisconsin, who has researched the hidden environmental costs of AI data centers, hyperscale facilities operate on a much larger scale than the data centers that supported digital services over the past decade.
As operators pack more servers into these facilities, they must also provide significantly more cooling.
Data Centers Depend on Water for Cooling
According to public records obtained by Oregon’s largest newspaper, Google’s hyperscale data center in The Dalles used nearly a third of the city’s water supply in 2021.
Many data center operators rely on water to cool their equipment. This dependence helps explain why large clusters of data centers often appear near freshwater sources.
The Great Lakes Region Attracts Data Centers
The Great Lakes region of the United States contains nearly a tenth of the world’s data centers. It also holds approximately one-fifth of the world’s surface freshwater, much of which stays naturally cold.
This combination makes the region attractive to companies that want access to large water supplies for cooling their computing infrastructure.
Evaporative Cooling Saves Electricity but Uses More Water
Many facilities use evaporative cooling to reduce their electricity consumption. This method allows operators to use fewer fans and less power, but it requires more water.
The Great Lakes cannot provide unlimited water. Less than one percent of their water supply gets replenished annually, and ecosystems, cities, and industries depend on relatively stable lake levels.
Communities must therefore balance their water consumption carefully and ensure that data centers do not place excessive pressure on local supplies.
Air Cooling Saves Water but Consumes More Electricity
Google’s data center in Mesa, Arizona, operates in one of the driest regions of the United States. The company uses air cooling there, with large fans pushing air across the servers.
This approach reduces the need for cooling water but increases electricity consumption.
Ireland faces a similar challenge with energy demand. Environmental activist Rosie Leonard explains that data centers consume a significant share of the country’s electricity, putting additional pressure on the energy system.
When data centers increase electricity demand, the public may experience higher energy bills. Research suggests that evaporative cooling can reduce the electricity needed for cooling by approximately 30 to 60 percent compared with air cooling, depending on the system and operating conditions.
Direct-to-Chip Cooling Transfers Heat More Efficiently
Direct-to-chip cooling uses a liquid coolant to remove heat directly from processors. Liquids transfer heat more efficiently than air because they have a much higher density and can carry away more heat within the same volume.
Maxie Reynolds from data center company Subsea Cloud explains that liquid cooling helps operators remove heat from powerful computing components more effectively.
Some systems circulate coolant through a sealed loop, which can eliminate the need for continuous on-site water consumption for cooling. However, the overall water footprint still depends on the facility’s electricity supply and other operational requirements.
Direct-to-Chip Cooling Is Becoming More Common
Direct-to-chip cooling is gaining popularity among large data center operators.
Microsoft is building a powerful AI data center in Wisconsin, in the Great Lakes region. The company plans to install hundreds of thousands of processors and use direct-to-chip cooling to reduce water consumption.
Industry research from McKinsey and Future Market Insights suggests that direct-to-chip cooling could account for approximately one-third of the data center cooling sector’s buildout by the end of 2030.
Immersion Cooling Submerges Servers in Special Liquids
What if operators could cool an entire server by placing it directly into a liquid bath?
Immersion cooling makes this possible. Although ordinary water could damage electronic components, data center operators use specially formulated, nonconductive liquids that do not conduct electricity or easily cause metal corrosion.
Taiwanese data center services provider Gigabyte says that immersing servers in this type of cooling liquid allows operators to arrange computing equipment more densely.
Immersion Cooling Can Reduce Energy Consumption
Immersion cooling can reduce the electricity that data centers spend on cooling, allowing them to direct more energy toward computing tasks.
One operator in Athens, Greece, has reported that its immersion-cooled facility directs more than 97 percent of its energy toward computing rather than cooling.
Industry researchers valued the immersion-cooled data center sector at approximately $17 billion in 2025, representing around one-third of the wider liquid-cooled data center market. These figures reflect reported industry estimates.
Subsea Cloud Is Taking Data Center Cooling Underwater
In Denmark’s cold coastal waters, one operator is taking immersion cooling further by placing data centers beneath the sea.
Subsea Cloud says its underwater approach eliminates the need for electrically driven cooling systems and can reduce power consumption by approximately 40 percent, with a similar reduction in carbon emissions under the company’s reported estimates.
Underwater Data Centers Use Nonconductive Coolant
Subsea Cloud’s container-like capsules weigh approximately 46 tons. Inside each capsule, high-density server racks hold more than 1,100 processors immersed in a nonconductive liquid coolant.
The coolant absorbs heat from the computing components and transfers it outward into the cold surrounding seawater.
The company says its design eliminates the need for fans, lights, and pumps to circulate the coolant, allowing approximately 99 percent of the electricity to go directly toward computing.
Operators could also connect these modules to offshore wind turbines to power their computing operations with renewable energy.
Underwater Data Centers Face Maintenance Challenges
Despite their cooling advantages, underwater data centers present significant maintenance challenges.
Maxie Reynolds explains that GPUs and other hardware can fail for many reasons. Technicians can replace or repair faulty components relatively easily in a land-based facility, but underwater installations make access and repairs much more difficult.
Data centers also face reliability problems regardless of their location. Surveys indicate that more than half of facilities have experienced at least one outage over a three-year period.
Space-Based Data Centers Could Use Solar Energy
Beyond traditional and underwater facilities, companies are exploring another frontier: space.
The concept involves launching data centers into orbit, where solar panels could provide electricity for computing. Some proponents also hope to use the space environment to help manage heat, although engineers must overcome significant thermal challenges.
In 2026, Elon Musk’s rocket company SpaceX merged with his artificial intelligence company xAI, with the goal of launching AI data centers into space.
Space Cannot Remove Heat as Easily as Many People Expect
Astronomer Samantha Lawler, who studies space debris, warns that cooling computers in space presents serious engineering challenges.
Although space is extremely cold in many respects, spacecraft cannot rely on surrounding air to carry heat away. Instead, they must radiate heat into space, which can be difficult when powerful equipment generates large amounts of thermal energy.
Objects exposed to direct sunlight can reach temperatures of approximately 120 degrees Celsius. Without effective thermal management systems, solar radiation and internal computing activity can cause equipment to overheat.
Space Debris Raises Concerns About More Satellites
Space already contains thousands of satellites, and adding large numbers of data center satellites could make low Earth orbit even more crowded.
Lawler argues that companies and governments should carefully prioritize satellites that provide the greatest benefits to society. Satellites already support life-saving services, help farmers plan crop rotations, and allow researchers to monitor climate change.
Expanding space infrastructure therefore requires careful consideration of the environmental, scientific, and practical benefits of each new launch.
Beyond Cooling: Reducing AI’s Energy Demand
Improving cooling technology is only one way to reduce the environmental impact of AI. Another approach involves reducing the amount of computing power that AI systems need in the first place.
Some researchers and developers support the idea of using smaller AI models to perform specific tasks efficiently. This approach, sometimes described as “small data, big tasks,” can reduce unnecessary computation and lower the energy and cooling demands of data centers.
AI Users Can Help Reduce Resource Consumption
AI companies can improve efficiency by developing better hardware, more efficient models, and advanced cooling systems. However, users can also make thoughtful choices about when they need AI.
People do not necessarily need a chatbot to draft every email, answer every minor question, or automate every small task in daily life.
Using AI where it provides meaningful value can help limit unnecessary computing demand. Every AI response relies on physical infrastructure that consumes resources, including water and electricity, and can place pressure on the communities surrounding data centers.
The Environmental Cost of AI Deserves More Attention
AI offers valuable capabilities, but its rapid expansion also creates environmental challenges. Traditional cooling systems can consume substantial amounts of water or electricity, while direct-to-chip cooling and immersion cooling offer alternative ways to manage heat.
Underwater and space-based data centers may provide additional possibilities, but they introduce their own technical, financial, and environmental challenges.
Companies must evaluate these technologies carefully and balance computing performance with energy efficiency, water conservation, reliability, and community needs.
What Do You Think About AI’s Water and Energy Use?
As AI becomes more common in everyday life, its demand for electricity, water, and computing infrastructure will continue to raise important questions.
Are you concerned about how much water and energy AI data centers consume? Do you think companies should focus more on reducing AI’s environmental impact before expanding these systems further?
