US hyperscale data centers used an estimated 82 terawatt-hours of electricity between May 2024 and April 2025, according to a study published in Nature Sustainability. The researchers estimated that the facilities’ electricity use produced 25.7 million metric tons of CO2 emissions. The figures point to a central factor in the climate cost of computing: the power source behind the machines.
A large electricity demand, estimated from the outside
The research team identified 403 hyperscale data centers in the US using satellite imagery. These large facilities, which can span millions of square feet, are being built in part to support generative AI. The study estimated their electricity consumption from each site’s electrical capacity.
To estimate emissions, the team used Environmental Protection Agency data to assess what each facility would have produced if connected to its local grid during the study period. The resulting estimate of 82 terawatt-hours amounted to about 1.8 percent of all US electricity consumption.
That total is greater than the annual electricity use of about 148 nations considered individually, according to John Sterman of MIT. The comparison offers a sense of scale: the electricity demand of this subset of US data centers rivals that of many countries, even though it is concentrated in a few hundred facilities.
The estimate is not a full accounting of every data center in the country. Hyperscale facilities are only a fraction of the thousands of US data centers, and the researchers had to infer consumption because operators do not disclose the information needed for a direct tally.
The grid changes the emissions picture
Electricity use alone does not determine a facility’s carbon footprint. The study estimated that about 54 percent of the power consumed by these hyperscale centers likely came from gas and coal. That mix helps explain the estimated 25.7 million metric tons of CO2 emissions, a figure comparable to Denmark’s annual carbon footprint.
The local electricity supply also matters. The study found that generating electricity in heavily coal-reliant Wyoming led to roughly seven times more carbon emissions per kilowatt-hour than in hydropower-rich Oregon. Two facilities using the same amount of electricity can therefore have very different estimated emissions depending on where they draw power.
That difference complicates claims that AI or data centers have one fixed environmental impact. The computing equipment may be similar, but the electricity system supporting it can change the outcome. Where facilities are built and how they are powered are choices with climate consequences.
What the estimates leave out
The study focused on electricity-related carbon emissions, so it does not capture every environmental effect linked to meeting data center demand. It does not account for local air pollution from new gas infrastructure built to supply these facilities, or from gas turbines some AI companies have installed to generate power at their sites without connecting to the grid.
Nor does every hyperscale facility serve AI workloads alone. These centers support other computing needs too, while the broader data center population includes many smaller sites beyond the 403 in the analysis. The study provides a substantial estimate for a defined group, rather than a complete picture of all US data center activity.
There is also uncertainty in the estimates because operators do not provide researchers with direct energy-use figures. Gianluca Guidi, the study’s lead author, and Fengqi You of Cornell University both highlighted the difficulty of assessing impacts without that information. Guidi and colleagues are using their research to build a website that tracks the environmental impact of individual data centers.
Choices made now could last
Sterman argues that data center operators should be required to bring new clean energy online. The study’s findings support the importance of power procurement and siting: cleaner electricity can reduce emissions for a given level of computing, while reliance on fossil fuels raises them.
Fengqi You also cautioned that new power plants and related infrastructure built to meet rising data center demand may operate for decades. That creates the possibility of long-term carbon lock-in: decisions made to serve facilities today can shape emissions well into the future.
Greater transparency could help communities and policymakers weigh those decisions. A bipartisan bill making its way through Congress could require more national disclosure of data centers’ electricity use. Gina McCarthy, former White House national climate advisor and EPA administrator, said reliable, independent information is important for understanding what these facilities mean for air, health and the climate.