Hyperscale data centers in the United States consumed an estimated 82 terawatt-hours (TWh) of electricity and were associated with 25.7 million metric tonnes of carbon dioxide emissions over a 12-month period, according to new research led by Harvard T.H. Chan School of Public Health and the Harvard Data Science Initiative. Published in Nature Sustainability, the study examined 403 hyperscale facilities supporting cloud computing, artificial intelligence, and other large-scale digital services and found that about 54% of their electricity came from fossil fuels.
The findings provide one of the first facility-level assessments of electricity consumption and associated emissions from US hyperscale data centers during the rapid expansion of AI infrastructure. The analysis covers May 2024 through April 2025 and estimates that the facilities accounted for approximately 1.8% of total US electricity consumption under its central scenario. Researchers warn that the sector's environmental footprint may already be considerably larger as new facilities continue to be developed.
US Hyperscale Data Centers Consume Up to 99 TWh Annually
The researchers used satellite imagery and facility-level information to identify 403 hyperscale data centers across the United States. They estimated electricity demand under different operating conditions, finding that annual consumption ranged from 68 TWh to 99 TWh, with a central estimate of approximately 82 TWh. At the upper end, the electricity demand approaches the annual consumption of individual US states such as Indiana, Michigan, and Tennessee.
Associated carbon dioxide emissions were estimated at between 21 million and 31 million metric tonnes across the operating scenarios, with a central estimate of 25.7 million tonnes. Harvard researchers noted that this central estimate was two to three times higher than an earlier 2018 estimate, reflecting the growth of large-scale computing infrastructure.
Read more: LŌ-GOI, A An Food Target 2M Carbon Credits From Low-Methane Rice
Fossil Fuels Supply 54% of Data Center Electricity
The study found that approximately 54% of the electricity attributed to the analyzed facilities came from fossil fuel generation. Natural gas accounted for around 38% of the electricity mix, while coal contributed approximately 15%. Nuclear power supplied around 21%, with renewable sources accounting for approximately 25%.
These findings highlight the importance of regional electricity generation in determining the carbon footprint of data centers. Facilities operating in areas with substantial coal and natural gas generation may be associated with higher emissions than those connected to grids with higher shares of renewable and nuclear electricity. The study also found that approximately one-quarter of hyperscale electricity demand was located in grid regions with above-average carbon intensity.
Four States Account for Over Half of Electricity Demand
Virginia, Oregon, Ohio and Iowa accounted for more than half of the electricity consumption attributed to the facilities examined. Virginia had the largest concentration, with 142 hyperscale data centers representing approximately 21 TWh of annual electricity demand. The geographic concentration reflects the development of major data center clusters around established digital infrastructure and electricity networks.
The researchers identified substantial differences in electricity emissions intensity across regions. Some facilities were located in areas heavily dependent on coal-fired generation, while others operated in regions with cleaner electricity supplies. The study's overall electricity-weighted carbon intensity was approximately 314 grams of CO₂ per kilowatt-hour, compared with a national generation average of approximately 348 grams, showing that regional conditions significantly influence the results.
Explore OneStop ESG Marketplace: Building energy & decarbonization
Data Gaps Complicate Climate and Public Health Assessments
The study combined satellite imagery, facility capacity estimates, and US Environmental Protection Agency electricity generation and emissions data to assess energy consumption and associated carbon emissions. Because comprehensive facility-level electricity consumption data were unavailable, researchers modeled operating conditions rather than directly measuring electricity use at every site. The analysis also could not distinguish electricity used specifically for AI workloads from other computing activities.
The researchers highlighted potential public health consequences associated with fossil fuel electricity generation. Power plants supplying data center demand can also release fine particulate matter, nitrogen oxides, and sulfur dioxide, which are associated with respiratory and cardiovascular health risks. Francesca Dominici, Director of the Harvard Data Science Initiative, said, "Decisions about where data centers go and how they’re powered are critical environmental and public health decisions, and they should be made with good data."
The research provides a framework for evaluating how data center location, electricity sourcing and operational demand influence environmental impacts. However, the estimates cover only the 403 facilities included in the study and should not be interpreted as the entire US data center industry's electricity consumption or emissions. As investment in AI infrastructure continues, improved facility-level transparency and access to lower-carbon electricity will be important for understanding and managing the sector's growing environmental footprint.
Source: Harvard T.H. Chan School of Public Health
Subscribe to our newsletter for more insights, case studies, and ESG intelligence.
Keep abreast of the top ESG Events on OneStop ESG Events.
OneStop ESG Educate: Your go-to source for top ESG courses and training programs tailored to your needs.
Stay informed with the latest insights on OneStop ESG News.
Discover meaningful career opportunities on OneStop ESG Jobs.
Ankit Palan is a Canada based writer who has been writing about sustainability for the past four years. He focuses on making topics like climate change, ESG, and responsible business easier to understand and more relatable. His work looks at how sustainability plays out in the real world, across businesses, finance, and everyday decisions, without overcomplicating it.
.png%3Falt%3Dmedia%26token%3D660a5d9a-2b92-43b8-a237-e5ed16daace2&w=3840&q=75)

.png%3Falt%3Dmedia%26token%3Dc9f5cf07-15b8-411b-987e-6a8951d79c5e&w=1920&q=75)
.png%3Falt%3Dmedia%26token%3Da24c295f-0418-4077-8eef-f31b5019491b&w=1920&q=75)
Comments
Have a thought on this? Share it with other readers.