The Most Powerful Data Centers Use as Much Electricity as an Entire State

A new study reveals the environmental costs of hyperscale data centers—most of which are powered by fossil fuels such as coal and natural gas.

A photo illustration featuring a black and white images of up close data servers with cords plugged in, with a red map of the United States on the top left corner.
A study recently published in the journal Nature Sustainability estimates that all U.S. hyperscale data centers use as much electricity as the state of Michigan or Tennessee.
Photo Illustration, NGM Staff; Source photos: mikroman6, Getty Images; Erik Isakson, Tetra Images, Getty Images
ByMeryl Davids Landau
Published November 18, 2026

Across the country, large data centers have become a charged political issue. But basic information, including how much energy they use and how much pollution they emit, has been largely lacking.

A paper recently published in the journal Nature Sustainability documents the consumption and output of the largest and most powerful data centers across the United States. Those roughly 400 hyperscale centers use almost as much electricity in a year as massive, mature industries like steel and aluminum and even as much as all the collective homes, businesses, and facilities in some individual states.

“It's pretty extraordinary—a few hundred buildings are consuming as much as an entire state,” says Gianluca Guidi, a data scientist at Harvard’s T.H. Chan School of Public Health and the study’s lead author.

Most of the electricity comes from burning fossil fuels, causing the combined centers to emit more than 25 million metric tons of carbon dioxide (CO2) a year, the study found. Carbon dioxide is a greenhouse gas largely responsible for climate change.

(Solar panels that plug into your wall are about to be everywhere.)

Hundreds more hyperscale data centers, each of which contain tens of thousands of computer servers that power the artificial intelligence industry and cloud computing, are planned or under construction, so their electricity demands are poised to grow significantly, Guidi says.  

“Communities and policymakers need reliable, independent data on what these facilities mean for our air, health, and the climate. This study is an important step,” Gina McCarthy, the former administrator of the U.S. Environmental Protection Agency, said in a statement.

The researchers calculated the number of hyperscale centers using a database from a commercial research company along with satellite imagery documenting their size. Then, using the research company’s information along with their own calculations of size and operating hours, they estimated annual electricity consumption for each facility. Finally, data from the Environmental Protection Agency revealed the mix of electricity sources used by utilities serving each center. The study covers the year between May of 2024 and April 2025.

Collectively, the centers required between 68 and 99 terawatt-hours of electricity (a terawatt is a trillion watts) during the study year. Roughly half, 54 percent, came from fossil fuels, including natural gas (38 percent) and coal (15 percent). Another 21 percent used nuclear energy, and a quarter employed hydropower, solar, wind, and other renewables.

Prior to this calculation, the last estimate quantifying hyperscale centers’ energy consumption published in a scientific journal appeared five years ago, using data from 2018. It calculated cumulative energy use then at 23 terawatt hours. “This was before the AI boom rapidly expanded the number of these centers. Our data shows it’s now at least three time higher,” says Francesca Dominici, a Harvard data scientist and coauthor of the paper. The earlier research also found carbon emissions were roughly three times lower than they are now.

The implications of this much electricity consumption go beyond the strain many electric grids are currently experiencing. The study authors say their future research will study how much these data centers take a toll on human health. Research clearly links emissions from fossil-fuel-based energy generation with changes to the climate, as well as health harms like lung and heart disease.

Even low levels of chronic exposure to pollution, including from fossil-fuel power plants, damages the heart, according to a study published in July in the journal Radiology. Cardiac computed tomography (CT) scans that had been performed in more than 11,000 Canadians concluded those living in areas with higher fine particulate pollution have more calcium plaque and artery obstruction. This occurred even in places where the particulate matter fell below allowable government standards.

“There was no lower level at which we didn't see an association, which indicates there may be really no safe level of air pollution exposure,” says Kate Hanneman, a cardiac radiologist at the University of Toronto, who conducted the research.

(Debunking three myths about air pollution.)

Exposure to emissions from coal-powered plants—some 15 percent of the hyperscale centers’ energy consumption—may be particularly harmful, says Corwin Zigler, an environmental scientist at Brown University. Older people exposed to its sulfur-dioxide-laden pollutants have especially high death rates, Zigler reported in Science in 2023. In recent years, coal plants have installed technologies reducing their emissions, but “still, the burden is not zero,” Zigler says.

The Nature Sustainability study found very large data centers are concentrated in specific parts of the country, with four states—Virginia, Oregon, Ohio, and Iowa—consuming more than half of the electricity use reported.

This concentration means some locations bear more burden than others. Research published last year by the same authors, examining computer centers mining for bitcoins, found higher concentrations of particulate pollution in some regions of the country. Some of that pollution remained close to the data centers, but other particulates appeared hundreds of miles away, thanks to wind and atmospheric patterns that disperse them.

Similarly, coal power pollution impacts health far from the plant locations. “If you live in Boston, some of that pollution you're breathing came from Ohio or Kentucky or Missouri or Indiana,” Zigler says of his research.

Communities with a concentration of data centers can also struggle to get the power they need, especially on hot days when air conditioners are running, Shaolei Ren, an electrical and computer engineer at the University of California, Riverside, said in an email. His ongoing research, not yet published, also finds health issues associated with living near data centers “can be very uneven across communities. In the most affected counties, the per-household health burden can be about seven times the national average,” his email stated.

Data centers can also deplete another resource, water. Some is used to keep computers from overheating, but as much as three-quarters of the usage is linked to electricity generation, according to early research by the Harvard researchers that has not yet been peer-reviewed by other scientists.

The rapid growth of data centers “should be a wake-up call” for communities and regulators, who are only starting to grapple with how to reliably and safely power this fast-growing industry while not straining the grid, polluting the air, harming the climate, or raising prices for other energy users, says Kasparas Spokas, who directs the electricity program at the research and advocacy nonprofit Clean Air Task Force. “How the demand for those data centers will be met will shape power sector emissions for decades,” he says.

Of nearly two dozen planned hyperscale data centers, only about half intend to source electricity from grids adding new renewable energy to their mix, according to an analysis in September by the nonprofit EnergyTag.

Some states are exploring options to demand that new, large data centers consume more clean energy, such as by charging them “large-load” tariffs or mandating that a high percentage of their power comes from renewables, Spokas says. “It's a difficult balance that communities and policymakers have to strike between promoting economic development while still prioritizing the economic and environmental concerns of local communities,” he says.

“AI may feel very digital, but the infrastructure powering it is very physical,” Guidi says. “This has real consequences for our environment and our communities.”