A new study of U.S. data centers found that hyperscale data centers use nearly as much electricity in a year as the entire state of Michigan — and that the total carbon dioxide they release is on par with Denmark’s emissions in a year.
Predictably, that’s way up from the sector’s footprint pre-artificial intelligence boom. Researchers at Harvard University’s T.H. Chan School of Public Health and the Harvard Data Science Initiative said their new study identified two to three times the greenhouse gas emissions from data center energy use as compared with a previous peer-reviewed estimate from 2018.
The new study, published today in the journal Nature Sustainability, estimated that hyperscale data centers emitted 25.7 million metric tons of carbon dioxide while using 82 terawatt-hours of electricity in the year from May 2024 through April 2025: right as the AI boom was ramping up. That’s about 1.8% of total U.S. electricity generation.
The researchers said that a little more than half of the energy used by data centers, or 54%, came from fossil fuels. Most of that energy use was concentrated in just four states — Iowa, Ohio, Oregon, and Virginia. Location matters because the fuel mix for electricity generation is different on different grids. The study found that about a quarter of data center demand comes in grid regions that are dirtier than the national average. In Ohio, for example, roughly 80% of electricity is generated by fossil fuels, according to the Energy Information Administration, and about 20% still comes from burning coal.

“Our goal is to give the public, policymakers, investors, and developers a much more concrete, up-to-date picture than was available before,” said the study’s senior author Francesca Dominici, who directs the Harvard Data Science Initiative. “Decisions about where data centers go and how they’re powered are critical environmental and public health decisions, and they should be made with good data.”
Dominici said further research will focus on the direct health impacts associated with air pollution from data center energy use. However, the researchers already see evidence that when AI’s growth relies on fossil fuels, it will make people sick.
“The companies building AI have plenty of money to do better,” said Dr. Vanessa Kerry, director of global and climate health policy in the Chan School’s Department of Environmental Health. “They must invest in new clean power and publish the energy use and emissions of every data center; we should not have to choose between the promise of AI and the health of the people it is meant to serve.”
A mapping tool
The study is the first peer-reviewed work to measure data center emissions and energy use at the facility level since the rise of AI, according to the researchers.
But it’s not just an academic exercise; Dominici and her colleagues said they want to make the information useful for stakeholders in host communities. They developed an explorer tool, Never Just a Building, to “turn that science into something people can act on: one set of independent facts for any proposed data center, before anything is built.”
The idea is that anyone from a developer to a local resident could use the platform to compare four candidate sites for a hypothetical 100 MW data center, with each location scored on a 0 (best) to 100 (worst) scale for its environmental and public health impact. The tool traces a proposed facility to the power plants that supply it and maps the fine-particle pollution downwind, census tract by census tract.
In the example, a new data campus in Loudoun County, Virginia — home to “Data Center Alley” — rates an undesirably high 76 because of the regional power supplier’s continued dependence on fossil fuels. The explorer tool also allows a user to envision different scenarios for data center development. Switch the hypothetical new facility in Loudoun County to 24/7 clean power, for instance, and the score improves from 76 to 54 as the associated pollution plume nearly disappears.
“The explorer takes no side on whether a project should go ahead,” the researchers wrote. “It gives developers, big tech companies, communities, local officials and regulators the same facts.”


