Skip to content
Bramble

Technology

Harvard study: 403 largest US data centres used about 1.8% of national electricity

A Harvard-led study in Nature Sustainability estimates that 403 hyperscale US data centres used about 82 TWh in the year to April 2025, roughly 1.8% of US electricity. Outside researchers say the results are plausible but rest on assumptions that are hard to verify.

2 outlets · 1L · 1C · 0R First reported Account updated
Image: New Scientist
Image: Wired

1 / 2

The story, neutrally told

Mixed · 2A Harvard-led study published Monday in Nature Sustainability estimates that the largest US data centres together use about as much electricity as a major US state. Left · 1The researchers compiled 403 hyperscale data centres that were operational between May 2024 and April 2025, using industry research and satellite data, and modelled their energy use in several scenarios. Centre · 1New Scientist reports that the team estimates these facilities used about 82 terawatt-hours (TWh) of electricity, equivalent to about 1.8 per cent of total US electricity use, more than the annual consumption of states such as Missouri, Kentucky and New Jersey. The upper bound of 99 TWh a year would rival Indiana, Michigan and Washington.

Mixed · 2Wired reports that the majority of that electricity came from fossil fuels and that data centre emissions are three times higher than before the AI boom. New Scientist reports that about 54 per cent of attributed generation was fossil-fuelled, with annual carbon dioxide emissions of 25.7 million tonnes. Left · 1Data centres in Virginia, Oregon, Ohio and Iowa accounted for more than 50 per cent of the energy use in the study, according to Wired. Mixed · 2The study focuses on hyperscale facilities, and not all of those surveyed are used for AI. Its authors say such data is hard to gather because federal and state governments do not track data centre use separately and operators often treat it as proprietary.

Centre · 1Francesca Dominici of Harvard, who worked on the study for two years with colleagues, acknowledges that "there are definitely a lot of assumptions baked in" and wants operators to be more transparent. Mixed · 2Outside experts gave a mixed assessment. Alex de Vries-Gao of the Free University of Amsterdam said the findings match International Energy Agency estimates and are "not particularly shocking", but that the study relies on commercial data that is hard to verify. Yury Dvorkin of Johns Hopkins called the analysis "coarse". Jon Koomey, an energy researcher not involved in the study, said a peer-reviewed study is more rigorous than some back-of-the-envelope calculations. Left · 1Wired says the study may already be outdated: the average facility studied is 40 megawatts, while developers are now planning campuses of hundreds or thousands of megawatts, as lead author Gianluca Guidi noted. The study also excludes behind-the-meter power plants built onsite, which Koomey calls a "wild card"; as much as 189 gigawatts of such generation is in the pipeline, and BloombergNEF estimated that building it all could raise US emissions by 10 per cent.

Mixed · 2New Scientist adds context on public opposition, citing a poll in which 76 per cent of Americans would oppose a new data centre in their community, and on water and land use. Wired notes that the Energy Information Agency has completed three pilot surveys in 10 states and is designing a potential voluntary survey of data centre power use.

Every sentence links to the reporting it rests on. The pill in front of each says where its sources sit: Left, Centre or Right when one side supplies at least half of them, Mixed when they are evenly split. The number is how many outlets it cites.

Left1 outlet

Framing
Wired leads with a striking comparison (400 data centres use more electricity than Michigan) and stresses that the study understates the problem because it omits off-grid gas plants and larger new campuses.
Emphasis
Lack of federal tracking, the omission of behind-the-meter power, rapid growth, the BloombergNEF emissions estimate and the EIA's planned survey.
Leaves out or plays down
Does not give the 82 TWh or 1.8% figures, the 99 TWh upper bound, or the outside critics' doubts about verifiability.
Charged language
“wild card”
For example
“New Harvard research shows that data-center power demand is outstripping states—and that doesn’t even include facilities with off-grid gas plants.” — Wired

Centre1 outlet

Framing
New Scientist leads with the share of national electricity (about 2 per cent) and the state comparison, then weighs the study's limits through outside experts and the lead researcher.
Emphasis
Precise figures (82 TWh, 1.8%, 25.7 Mt CO2), methodological caveats, public opposition and local impacts on water, land and prices.
Leaves out or plays down
Does not mention the regional concentration in Virginia, Oregon, Ohio and Iowa, the EIA survey effort, or the 40 MW average facility size.
Charged language
“fast and furious”“most painful type of facility”
For example
“The largest data centres in the US are responsible for about 2 per cent of the country's entire annual electricity use” — New Scientist
“You can't really validate the results” — New Scientist

Right0 outlets

No right outlet in our sources has covered this story yet.

Ask about this story

Loading…