AI's Hidden Cost: Data Centers Strain Power, Water and Land Globally

Every time someone types a question into ChatGPT, an answer materializes almost instantly on their screen. What feels weightless and immediate is anything but. Somewhere in the world—in a sprawling warehouse in Virginia, perhaps, or São Paulo, or Hong Kong—a data center the size of a city block is burning through electricity to process that single query, along with 2.5 billion others like it every day. The servers never stop running. The cooling systems never stop either. And the bill for all that processing power is being paid not in dollars but in energy, water, and land.

If data centers were ranked as a country, they would be the world's eleventh-largest consumer of electricity. That ranking comes from researchers at the United Nations University, and it carries weight: data centers now consume as much power as entire nations. The International Energy Agency projects that consumption will double by 2030, reaching roughly 945 terawatt-hours annually—equivalent to everything Japan uses in a year. The growth is accelerating because the demand for AI processing is accelerating, and there is no sign of it slowing.

The electricity has to come from somewhere, and in most cases, it comes from fossil fuels. More than half of America's data centers run on coal, gas, or oil, producing over 105 million tons of carbon dioxide annually. To offset that emissions load would require more than 31,000 wind turbines running continuously for a full year. The problem is so acute that utilities across the United States have delayed shuttering fifteen coal-fired power plants to keep pace with data center demand. Those plants alone released nearly 65 million tons of greenhouse gases in 2023. Meanwhile, more than 100 gigawatts of new gas-fired capacity is already under construction or announced. The carbon footprint of that new infrastructure would equal burning 50.4 million kilograms of coal. In Frankfurt, Germany, the electrical grid is so saturated that no new data center connections will be possible until at least the 2030s. In Dublin, data centers consume 82 percent of the city's electricity.

Water is the second hidden cost. A single large data center can draw 5 million gallons of water daily—roughly what a town of 10,000 to 50,000 people consumes. The water serves one purpose: cooling. Servers generate constant, intense heat, and water is the most efficient medium for absorbing and dispersing it through evaporation. About 80 percent of the water withdrawn simply evaporates into the air; the remaining 20 percent becomes wastewater that often overwhelms treatment systems never designed to handle such volume. In northern Virginia, where nearly 300 data centers cluster together to supply roughly a third of the world's internet traffic, facilities collectively consumed nearly 2 billion gallons of water in 2023 alone. More efficient cooling methods exist—immersion cooling, where servers are bathed in specialized coolant, or closed-loop systems that recycle water and can reduce freshwater consumption by up to 70 percent—but these technologies remain less common than traditional water-intensive approaches.

The heat that data centers expel doesn't vanish; it radiates outward into the surrounding environment. Researchers studying Phoenix, Arizona found that a single large facility can emit as much waste heat as tens of thousands or even hundreds of thousands of homes. That heat raises land surface temperatures by up to 9 degrees Celsius within half a kilometer of the facility and air temperatures by up to 2 degrees Celsius. In cities already 0.5 to 4 degrees Celsius warmer than surrounding rural areas due to the urban heat island effect, this additional warming increases heat stress on residents and drives up air conditioning demand—which requires more electricity, which requires more power plants, which emit more heat. Some data centers have begun capturing this waste heat and piping it into nearby homes and businesses through thermal energy networks, a practice that can reduce a facility's power demand by up to 30 percent.

Air quality suffers on two fronts. Coal and gas plants that power data centers release toxic particulate matter linked to respiratory disease, heart disease, and asthma. On-site diesel generators, deployed as backup power when the grid fails, are far worse: they emit 200 to 600 times more nitrogen oxides than natural gas plants. A single estimate found that when Virginia's data center backup generators operated at just 10 percent of their permitted emission levels, they could contribute to 14,000 cases of asthma symptoms, more than a dozen deaths, and up to $300 million in annual public health costs.

Land itself is being consumed. Data center construction requires large plots, often repurposed from agricultural or natural areas. In Hattersheim, Germany, an 18-acre parcel designated for agriculture and groundwater protection—a local green space known for rapeseed blossoms—was rezoned in 2024 to accommodate a five-building data center campus. The damage extends beyond the construction site itself. Gas pipelines and power plants built to supply these facilities fragment habitats and erode soil along their routes, fragmenting ecosystems that took decades or centuries to establish.

Renewable energy can power data centers, and pairing facilities with on-site solar or wind can ease pressure on local grids and even return surplus power to the network. But every kilowatt of green electricity directed toward data centers is electricity unavailable for electrifying transportation or decarbonizing other emissions-heavy industries. The choice is not between dirty and clean power; it is between competing claims on limited clean power. As AI demand grows, that competition will only intensify.