Skip to main content

Morning Overview

Big U.S. data centers put out as much as 31 million tons of CO2 in a single year

The rush to build AI data centers has outrun almost every attempt to measure it. The building boom has moved so fast that it has been difficult even to collect basic information on what is being built, let alone plan how to supply it with clean electricity. A team of researchers has now pieced together the first facility-by-facility estimate of what the largest U.S. data centers consumed in a year, and what that electricity put into the air.

Read next on Morning Overview: Cadillac cut the price of its cheapest EV, which goes up to 317 miles on a charge

Their answer: 403 hyperscale data centers used roughly 68 to 99 terawatt-hours of electricity between May 2024 and April 2025, and the power plants that supplied it emitted an estimated 21 to 31 million metric tons of carbon dioxide.

403 hyperscale centers, up to 31 million tons of CO2

The study, published in Nature Sustainability by a team whose authors include Gianluca Guidi, Francesca Dominici and Tiziano Squartini, compiled facility-level information on U.S. hyperscale data centers and matched each to the power grid around it. In its abstract, the team reports that under its central scenario, the centers used about 82 terawatt-hours, about 1.8% of total U.S. electricity consumption, with attributable emissions of about 25.7 million metric tons. Across all of its load scenarios, emissions ranged from about 21 to 31 million tons.

As Ars Technica’s Scott K. Johnson explains, the researchers started with 675 facilities from a commercial database built on news reports, company statements and satellite imagery. They dropped any without a reported electrical capacity, or that they could not confirm was a large building with the expected cooling and electrical equipment, leaving 403 centers averaging about 40 megawatts each. The range in the results reflects uncertainty about how hard those centers run: some conventional data centers average about 48% of capacity, while AI training loads have run closer to 70%.

The upper figure for electricity use, 1.5 to 2.2% of the nation’s total, puts these centers roughly on par with the total consumption of Wisconsin or Indiana. The 21 to 31 million tons of CO2 amount to around 0.5% of total U.S. fossil fuel CO2 emissions.

A grid mix only slightly cleaner than average

The centers were not concentrated in especially clean grids. About 54% of the electricity attributed to them came from fossil fuel plants, dominated by natural gas, with 25% from renewables and 21% from nuclear. The paper puts their electricity-weighted carbon intensity at about 314 grams of CO2 per kilowatt-hour, compared with a U.S. average of about 348.

The estimate is also conservative in one important way. Another 272 facilities thought to be large data centers were left out because they could not be verified. And because the researchers could not separate AI computing from conventional data services, some of the centers included may not be AI facilities at all.

For scale, Ars Technica notes that a recent Lawrence Berkeley National Laboratory report estimated that all U.S. data centers, excluding cryptocurrency mining, used 76 terawatt-hours in 2018 and 192 terawatt-hours in 2024. That suggests the hyperscale centers in the new study account for something like one-third to one-half of all data center electricity use.

Behind-the-meter gas turbines are not in the count

One growing source of emissions falls outside the study entirely: power generated on site. The analysis counts only electricity drawn from the grid, not “behind-the-meter” generation such as the fleet of natural gas turbines that xAI’s Tennessee data center ran during the study period. Memphis residents and the Southern Environmental Law Center accused xAI of running more turbines than local officials knew about, without permits, at the site powering its Colossus supercomputer.

On-site generation is spreading quickly. The market intelligence firm Cleanview says it identified 59 data centers with a combined capacity of about 90 gigawatts that plan to build their own power, more than 25% of all planned U.S. data center capacity in its tracker, and that 92% of those projects have been announced since the start of 2025. “Power generation efficiency is out. Speed to power is all that developers care about,” the firm wrote. Cleanview found projects tied to Meta, Microsoft, Amazon and Oracle, and said AI labs including OpenAI and Anthropic have signed leases for more than 10 gigawatts of behind-the-meter power. In one case, a developer unable to secure conventional turbines placed a $1.25 billion order with Boom Supersonic, a company that has never sold a power generation product.

The equipment matters for emissions. Developers short of large, efficient combined-cycle turbines are turning to smaller and less efficient generators, which climate scientist Zeke Hausfather estimates emit 60 to 80% more greenhouse gases than grid power to produce the same electricity. Building on-site solar instead could do the reverse and cut a center’s footprint.

The new study gives the first clear baseline for tracking all of this. Its authors describe the work as a facility-resolved framework for measuring the carbon footprint of hyperscale data centers, and the next year of data will show how much the 21 to 31 million ton range grows as more of the AI buildout comes online.

This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.


More from Morning Overview

Morning Overview is reader-supported. Some links in our articles are affiliate links, and we may earn a commission at no extra cost to you. As an Amazon Associate I earn from qualifying purchases. Full disclosure.