Skip to content
Economy Media Stay Informed Anytime, Anywhere
Tech

Harvard: 403 US Data Centers Use 1.8% of Nation’s Electricity

Harvard researchers estimate 403 large US data centers used about 82 TWh in a year, roughly 1.8% of national consumption, with more than half from fossil fuels.

CAMPUS - Military Data Center
Photo: Welcome to Switzerland backstage! via Flickr, CC BY 2.0

A group of 403 large U.S. data centers consumed roughly 82 terawatt-hours of electricity in a single year, about 1.8% of national electricity use, according to a new Harvard-led study. Gizmodo reported that the research was published Nov. 16 in the journal Nature Sustainability. Wired framed the findings as showing that just 400 or so facilities use more electricity than the state of Michigan.

What the study found

The researchers, led by Gianluca Guidi, a postdoctoral researcher at the Harvard T.H. Chan School of Public Health, examined “hyperscale” data centers, the very large facilities run by cloud and technology companies. The study covered the period from May 2024 to April 2025.

In the central scenario, the facilities used about 82 terawatt-hours (TWh) a year. Because actual power loads are not directly observed, the team gave a range of 68 to 99 TWh depending on assumptions, according to the authors’ preprint, which was posted on the arXiv research server.

About 54% of the electricity attributed to these data centers came from fossil fuels: roughly 38% from natural gas and 15% from coal. The rest came largely from nuclear power (about 21%) and renewable sources (about 25%). The preprint put the average carbon intensity of that power at about 545 grams of CO2 per kilowatt-hour, around 48% above the 2023 national grid average of roughly 370 grams.

Four states, Virginia, Oregon, Ohio and Iowa, accounted for more than half of the studied facilities’ consumption. Virginia alone had 142 of the data centers in the sample and about 21 TWh of use in the central scenario, the preprint said.

Emissions estimates differ between reports

Gizmodo reported the study’s emissions figure as about 28 million tons of CO2 and said that was two to three times higher than earlier estimates. The preprint version of the paper gave a range of 37 to 54 million metric tons, with a central estimate near 45 million. The numbers may differ because the published paper and the earlier preprint use different versions or scenarios; Economy Media could not confirm which figure appears in the final journal article.

Why it matters

Guidi told Gizmodo the 82 TWh figure is “approaching the annual electricity consumption of an entire mid-sized American state.” He also said that rising demand met by fossil fuels makes an increase in emissions “inevitable.” Fossil-fuel plants also release fine particulate matter, nitrogen oxides and sulfur dioxide, which are linked to heart and lung disease, and Guidi said people living near and downwind of those plants may bear health costs.

The scale of expansion adds to the concern. Guidi said more than 1,000 hyperscale facilities are already operating in the United States, with about 2,000 more planned.

What is not known

The study has several stated limits. It did not cover every U.S. hyperscale data center: the preprint says only 403 of 675 identified facilities were retained. The data cannot separate artificial intelligence workloads from general-purpose computing, so the specific role of AI is not measured. The authors also used average emissions factors for power plants, which do not capture how new data centers change grid operations, and they relied on estimated rather than metered facility loads. The detailed facility-level dataset is held under a data use agreement, so only aggregated data has been made public.

Wired noted that the research does not count facilities that rely on off-grid gas plants. Guidi said his next study will compare air pollution from grid-connected data centers with that from facilities that generate their own power on site.

No data center operators or utilities have been reported as responding to the findings in the coverage reviewed.