It has been revealed that A study estimates that 403 major U.S. data centers used 82 terawatt-hours of electricity in a year, with natural gas and coal supplying more than half.
In a recent development, every backed-up photo, every chatbot answer and every streamed show lives somewhere called the cloud, which sounds like nowhere at all. It is a set of windowless warehouses.
Industry observers note that at the high end of the estimate, 403 of the largest U.S. data centers together used nearly as much electricity in a year as Indiana uses statewide.
More than half of that electricity came from burning gas and coal.
In a recent development, A team led by the Harvard T.H. Chan School of Public Health and the Harvard Data Science Initiative mapped 403 of these buildings, known as hyperscale data centers.
The researchers then estimated what each one drew from the grid between May 2024 and April 2025.
The totals are estimates built from each building’s electrical capacity, not meter readings.
“Data centers are growing incredibly fast, driven largely by AI, and decisions about where to build them and how to power them are being made right now,” said lead author Gianluca Guidi, a postdoctoral fellow in the Department of Biostatistics.
“But there’s very little facility-level information about their environmental footprint.”.
“Our study helps fill that gap,” Guidi added.
Hyperscale data centers are the largest class of data center, the kind that runs cloud services and trains substantial AI models. The buildings in the sample had an average electrical capacity of about 40 megawatts.
Industry observers note that the team started with 675 candidate sites from a commercial database and kept the 403 it could confirm. Each address was matched to a rooftop and its footprint checked against OpenStreetMap.
Then the researchers looked at satellite and aerial images for the signs of a data center: significant uniform roofs, cooling equipment and a substation next door.
Virginia had 142 of the 403, more than any other state. Oregon had 56 and Ohio 38.
No data center runs at its full rated capacity every hour, and the companies don’t publish their real loads. So the team multiplied each facility’s capacity by the hours in a year and by a load factor.
The researchers ran four versions of that factor, from a light load to one meant to bracket heavy AI training.
Sources indicate that across the four versions, the 403 facilities used an estimated 68 to 99 terawatt-hours of electricity over the year. The central estimate is about 82 terawatt-hours, or roughly 1.8% of all the electricity the country used.
Recent findings suggest that indiana, Michigan and Tennessee each use about 100 terawatt-hours a year. So at the high end these warehouses matched a whole state, and even the central estimate covers most of one state’s use.
Industry observers note that the central figure is about 3.6 times a 2021 estimate of what hyperscale centers used in 2018. Projections of AI’s appetite for electricity look ahead; this study counts the buildings that exist now.
To work out where the electricity came from, and the power emissions behind it, the team assigned each data center to its regional grid operator.
The team then allocated the data centers’ electricity demand across power plants in each region based on how much electricity each plant generated.
The plant-by-plant generation and emissions figures came from the Environmental Protection Agency.
On that basis, 53.9% of the electricity came from fossil fuels: 38.2% from natural gas and 15.2% from coal.
Recent findings suggest that nuclear plants supplied 20.9%, and hydro, wind, solar, biomass and geothermal together supplied 25.3%. Nuclear and renewables together fell short of half.
The result was about 25.7 million metric tons (28.3 million U.S. tons) of carbon dioxide under the central version, and 21 to 31 million metric tons across all four.
A 2021 study estimated that U.S. hyperscale data centers were responsible for greenhouse gas emissions equivalent to about 10.5 million metric tons of carbon dioxide in 2018.
Recent findings suggest that virginia alone accounted for about 21 terawatt-hours, roughly a quarter of the total. Oregon followed with about 11, Ohio with about 9 and Iowa with about 7.
Virginia, Oregon, Ohio and Iowa together made up more than half of the electricity use, so the industry is heavily concentrated.
According to recent reports, oregon’s data centers drew more electricity than Ohio’s. Yet Oregon’s were tied to about 1.3 million metric tons of carbon dioxide and Ohio’s to 2.9 million, because Oregon’s grid runs far cleaner.
The fuel mix behind each state’s grid decides the carbon.
Virginia’s data centers were tied to about 7 million metric tons, the most of any state.
In a recent development, PJM, the grid operator serving Virginia, oversaw the region with the highest data center electricity demand. Its territory spans parts of 13 states and the District of Columbia.
There, about 60% of the attributed electricity came from fossil fuels and 33% from nuclear plants.
According to recent reports, the method gives each data center the average fuel mix of its regional power grid. It doesn’t trace which plant physically fed which building.
It also doesn’t count the renewable power contracts some operators sign, unless that power shows up in the regional mix. The authors wrote that the numbers describe the grid, not any manufacturer’s performance.
Measured that way, the data centers’ electricity was slightly cleaner than the national average: about 314 grams of carbon dioxide per kilowatt-hour against 348 for the country.
About 23% of the demand, though, sat in regions dirtier than that average.
The data also can’t separate AI work from ordinary cloud computing inside a building. And the 272 candidate sites the team couldn’t verify were left out, so the totals likely run low for the industry as a whole.
Guidi noted the picture has already moved on.
“The data center boom hasn’t slowed down since our study period ended in April 2025. Fresh facilities have been built, existing campuses have expanded, and demand for AI computing continues to grow,” he noted.
The researchers estimated electricity use and carbon dioxide emissions.
The team looks to examine the health effects next, according to study co-author Francesca Dominici, director of the Harvard Data Science Initiative.
“The fossil fuel power plants in the regional electricity systems associated with this demand also release other pollutants, such as fine particulate matter, nitrogen oxides, and sulfur dioxide, that are directly linked to heart and lung disease and premature death,” Dominici said.
The people who breathe those pollutants live near the power plants, which may be far from any data center.
Earth.com has covered the industry’s other local footprints, from heat zones around server farms to the one in four data centers in water-stressed regions.
The full study was published in the journal Nature Sustainability.
Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.
