---
title: "US hyperscale data centers used 68 to 99 TWh in a year, about 1.8% of US electricity, Harvard-led study finds"
url: https://badiyaa.com/us-hyperscale-data-centers-electricity-study/
published: 2026-11-17T12:33:00+00:00
modified: 2026-11-17T12:33:00+00:00
section: Tech
publisher: Badiyaa
---

# US hyperscale data centers used 68 to 99 TWh in a year, about 1.8% of US electricity, Harvard-led study finds

US hyperscale data centers used 68 to 99 TWh of electricity from May 2024 to April 2025, about 1.8% of national use, a Harvard-led study estimates.

The 403 largest data centers in the United States used somewhere between 68 and 99 terawatt-hours of electricity in a single year, according to a new facility-by-facility study. That is roughly what an entire mid-sized state such as Indiana, Michigan or Tennessee uses for everything in the same period.

Under the researchers’ middle estimate, about 82 terawatt-hours, these so-called hyperscale data centers accounted for around 1.8 percent of all electricity consumed in the US between May 2024 and April 2025.

## Who did the counting, and how

The study, posted online as a preprint, was led by researchers in the Department of Biostatistics at the Harvard T.H. Chan School of Public Health, working with colleagues from Italy, UCLA, the mapping company Esri and the data center research firm Baxtel.

Hyperscale facilities are the giant campuses run by cloud and AI companies, typically drawing tens of megawatts each. The team built a list of them, checked each site against satellite imagery, and estimated how much power every one drew under four different load assumptions. It then traced which power plants serve each facility’s region using EPA plant-level data.

The authors say plainly why these buildings matter now: their growth has been “mainly driven by the adoption of artificial intelligence.”

## Where they are and what powers them

Virginia dominates. It hosts 142 of the 403 hyperscale centers in the dataset, followed by Oregon with 56 and Ohio with 38. Put another way, about 35 percent of the facilities sit in a single state.

The electricity behind them is still mostly fossil. The study attributes about 53.9 percent of their power to fossil-fuel plants, with natural gas at 38.2 percent and coal at 15.2 percent. Nuclear supplies 20.9 percent and renewables 25.3 percent.

That mix gives these facilities a carbon intensity of about 545 grams of CO2 per kilowatt-hour, against a US grid average of 370 grams over the same period. Divide one by the other and each unit of hyperscale electricity carried roughly 1.47 times the emissions of the average US unit, which is where the study’s figure of about 48 percent higher comes from. Total attributable emissions came to an estimated 37 million to 54 million metric tons of CO2, with Virginia alone responsible for about 11.5 million tons under the central scenario.

## Bigger than before

The comparison with earlier work shows the speed of the change. An earlier study put hyperscale electricity use at 22.85 terawatt-hours in 2018. The new central estimate is about 3.6 times that figure, an increase the authors link to large cloud and AI facilities replacing smaller, less efficient data centers.

For scale, Texas, the biggest electricity user among US states, consumed about 475 terawatt-hours. The central hyperscale estimate equals about 17 percent of that.

## Limits of the estimate

The researchers flag several caveats. Their data cannot separate AI work from general-purpose computing inside a given facility, so hardware and cooling details stay partly hidden. Coverage is incomplete, which would push the true totals higher, while the load assumptions could push them in either direction. Cryptocurrency mining operations were left out entirely because they have been studied separately. The paper is a preprint, so it has not yet completed peer review.

Data center power has become a local issue as well as a national one; see how [a Utah AI data centre fight reached a Saturday Night Live sketch](https://badiyaa.com/snl-kevin-oleary-utah-ai-data-centre/), and how a [Texas campus microreactor cleared its final DOE safety review](https://badiyaa.com/last-energy-pwr-5-texas-am-rellis-doe-safety-analysis/).

The team has also built a public web tool that maps hyperscale electricity demand and emissions by state and by grid region, so the numbers can be updated as new plant data arrives.

## Further reading

- [arXiv: Assessing the Carbon Emissions and Energy Consumption of U.S. Hyperscale Data Centers](https://arxiv.org/html/2606.05420)

*Image: Theodore Christopher, CC0 via Wikimedia Commons*

*This article was produced with AI assistance and reviewed by our Editorial Desk. [How we work](https://badiyaa.com/editorial-policy/).*
