---
title: "US hyperscale data centers emitted about as much CO2 as Denmark; Harvard study counts 25.7 million tons"
url: https://badiyaa.com/data-centers-carbon-emissions-denmark-harvard/
published: 2026-11-19T08:32:00+00:00
modified: 2026-11-19T08:32:00+00:00
section: Science
publisher: Badiyaa
---

# US hyperscale data centers emitted about as much CO2 as Denmark; Harvard study counts 25.7 million tons

A Harvard-led study ties 403 US hyperscale data centers to 25.7 million metric tons of CO2 a year, about 96% of Denmark's annual fossil emissions.

America’s 403 biggest data centers were tied to about 25.7 million metric tons of carbon dioxide in a single year. That is almost exactly what the whole of Denmark emits from fossil fuels and cement, a country of roughly 6 million people.

The figure comes from a Harvard-led study published November 16 in the journal Nature Sustainability, which estimated the power use and emissions of US hyperscale data centers facility by facility. [The Harvard T.H. Chan School of Public Health announced](https://hsph.harvard.edu/news/hyperscale-data-centers-energy-consumption-carbon-emissions-newly-estimated-by-researchers/) the results alongside the Harvard Data Science Initiative.

## Checking the Denmark comparison

The Denmark line is a comparison, not a quote from the study, so here is the math. The international EDGAR emissions database, as compiled in [this country-by-country emissions table](https://en.wikipedia.org/wiki/List_of_countries_by_carbon_dioxide_emissions), puts Denmark’s 2023 fossil CO2 output at 26.77 million metric tons. Divide the data centers’ 25.7 million by that and you get about 0.96. In other words, the data centers in the study matched roughly 96% of Denmark’s annual fossil carbon.

The comparison holds even though the two numbers measure slightly different things. The data center figure is attributed: it assigns each facility a share of the emissions from the power plants on its regional grid. Denmark’s figure counts what is burned inside its borders.

## How the researchers built the estimate

Few data center operators publish facility-level energy figures, so the team assembled its own list. Using satellite imagery, they identified 403 hyperscale sites across the United States, the giant campuses that run cloud services and train and serve AI models. For each one they estimated yearly electricity use from May 2024 to April 2025 under several workload scenarios. Then they used Environmental Protection Agency plant data to work out which fuels likely generated that power.

The headline results:

- Electricity: 68 to 99 terawatt-hours a year, depending on how hard the sites were running. The top of that range approaches a full year of power use in states such as Indiana, Michigan or Tennessee.

- Share of the grid: about 1.8% of all US electricity consumption.

- Fuel mix: roughly 54% fossil, mainly natural gas (38%) and coal (15%).

- Geography: Virginia, Oregon, Ohio and Iowa together accounted for more than half of the electricity studied.

The emissions estimate is two to three times higher than an earlier estimate for 2018, according to the school.

“Data centers are growing incredibly fast, driven largely by AI, and decisions about where to build them and how to power them are being made right now,” said lead author Gianluca Guidi, a postdoctoral fellow in biostatistics. “But there’s very little facility-level information about their environmental footprint.”

## Why location matters so much

A server draws the same power wherever it sits, but its carbon cost depends on the grid behind the plug. The study found that about a quarter of the electricity came from grid regions dirtier than the national average, and many facilities clustered in areas that lean heavily on coal. Two identical campuses could therefore carry very different footprints. Badiyaa’s earlier report on [how much electricity these same 403 data centers use](https://badiyaa.com/us-hyperscale-data-centers-electricity-study/) covered the power side of the study; the carbon side shows why siting decisions carry so much weight. Some operators are even [turning jet engine cores into on-site gas turbines](https://badiyaa.com/proenergy-jet-engine-turbines-data-centers/) to get power faster.

## Limits

The study could not cover every hyperscale site in the country, and its data could not separate general cloud computing from AI-specific work. Its numbers are modeled from grid averages, not measured at the meter. An earlier preprint of the paper, posted in June, gave a higher emissions range of 37 to 54 million metric tons. The researchers also say the totals have likely grown since April 2025. “What we’re seeing is already striking, but it’s only a snapshot of a rapidly expanding industry,” Guidi said. Next, corresponding author Francesca Dominici’s team plans to estimate the health effects tied to the power plants feeding these sites.

## Further reading

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

- [Paper DOI](https://doi.org/10.1038/s41893-026-01985-7)

*Image: Carl Lender, CC BY 2.0 via Wikimedia Commons*

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