Plate Nº 35 · recorded November 16, 2026
Earth & ClimateReported finding
U.S. hyperscale data centers rival states in electricity use
Harvard researchers estimated that 403 U.S. hyperscale data centers used 68–99 terawatt-hours from May 2024 to April 2025, linking their power demand to 25.7 million metric tons of CO₂.
By Marcus Bennett5 min read967 words
In brief
- Researchers identified 403 U.S. hyperscale data centers and estimated their electricity use from May 2024 through April 2025.
- Combined annual demand was estimated at 68–99 terawatt-hours, or about 1.8% of U.S. electricity consumption.
- The demand was associated with about 25.7 million metric tons of CO₂, two to three times the 2018 estimate.
- Fossil fuels supplied about 54% of the electricity: 38% from natural gas and 15% from coal.
- Four states—Virginia, Oregon, Ohio and Iowa—accounted for more than half of the electricity use studied.

By Marcus Bennett
The 403 U.S. hyperscale data centers studied used an estimated 68 to 99 terawatt-hours of electricity from May 2024 through April 2025—about 1.8% of total U.S. consumption.
Their demand was associated with approximately 25.7 million metric tons of carbon dioxide, two to three times an earlier estimate for 2018.
A hyperscale data center is one of the largest and most powerful facilities in the industry. Such sites support cloud computing, train large language models and deploy artificial intelligence tools globally.
The Harvard T.H. Chan School of Public Health and the Harvard Data Science Initiative led the study, which provides one of the first facility-level estimates of data-center electricity use and carbon dioxide emissions since the rise of AI.
The paper, "Assessing the carbon emissions and energy consumption of US hyperscale data centres," is scheduled for publication on Nov. 16, 2026, in Nature Sustainability. Its DOI is 10.1038/s41893-026-01985-7.
What did the researchers measure?
Researchers combined facility identification, operating scenarios and regional power-plant data to estimate the footprint of 403 hyperscale sites.
- They used satellite imagery to identify hyperscale data centers across the United States.
- They estimated each facility's annual electricity consumption between May 2024 and April 2025 under several operating scenarios.
- They used Environmental Protection Agency data to attribute emissions according to the mix of power plants feeding each facility's regional grid, or local electricity system.
Few earlier studies had estimated both facility-by-facility electricity use and the fuels behind that electricity. Lead author Gianluca Guidi, a postdoctoral fellow in the Department of Biostatistics, said: "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. But there's very little facility-level information about their environmental footprint."
He added: "Our study helps fill that gap."
Corresponding author Francesca Dominici, the Clarence James Gamble professor of biostatistics, population and data science and director of the Harvard Data Science Initiative, said the study gives the public, policymakers, investors and developers a more concrete, up-to-date picture.
She said decisions about where data centers go and how they operate should rest on sound environmental and public-health data.
Takeaway: The study pairs each facility's power demand with the fuel mix of the grid serving it, giving site planners a more useful baseline.
How large is the combined electricity and carbon footprint?
At the upper estimate, the 403 facilities' combined 99 terawatt-hours approach the annual electricity use of states such as Indiana, Michigan and Tennessee.
A terawatt-hour is a unit of electrical energy; the full range represents one year of demand across all 403 studied sites.
- Annual electricity demand: 68 to 99 terawatt-hours, depending on the operating scenario.
- Share of U.S. consumption: About 1.8% during the study period.
- Associated carbon dioxide emissions: Approximately 25.7 million metric tons, two to three times the earlier estimate for 2018.
The range reflects different assumptions about how intensively the facilities operate. It therefore presents a modeled footprint rather than a single metered total.
Takeaway: Under the higher operating scenario, the studied data centers draw about as much electricity as several individual states.
How does regional power change the footprint?
The fuel mix behind the electricity varies substantially across the country.
- Fossil fuels: About 54% of the electricity powering the studied facilities.
- Natural gas: 38% of the electricity supply.
- Coal: 15% of the electricity supply.
- Higher-emitting grids: About one-quarter of the electricity came from regions the researchers described as dirtier than average.
Many facilities were concentrated in areas where coal-fired power plants generated much of the electricity. Other regions relied more heavily on natural gas or had different mixes of generation, producing markedly different associated emissions for similar levels of data-center demand.
Four states—Virginia, Oregon, Ohio and Iowa—together accounted for more than half of the electricity use studied. That concentration makes regional generation policies especially relevant to the footprint of this particular set of facilities.
Takeaway: The same amount of electricity can produce very different associated emissions depending on which regional grid supplies it.
What remains unknown about health and today's total?
The study's coverage was incomplete because researchers lacked enough data on every hyperscale data center in the United States. The records also could not separate power used for general-purpose computing from power used for AI-specialized work.
The researchers did not directly measure health effects, and they plan to pursue that work next. Dominici said: "Quantifying health impacts is an important next step."
The study identifies two main pathways:
- Climate pathway: Carbon dioxide drives climate change, which affects health through heat waves, wildfire smoke, flooding and other routes.
- Pollution pathway: Fossil-fuel power plants also release fine particulate matter, or tiny airborne particles, along with nitrogen oxides and sulfur dioxide, combustion gases linked to heart and lung disease and premature death.
Communities near and downwind of power plants may bear health costs even when they live far from the data center drawing electricity from those plants.
Guidi said the data-center boom had not slowed since the study period ended in April 2025. New facilities have opened, existing campuses have expanded and demand for AI computing continues to grow. He said the reported electricity use and carbon footprint could already be substantially larger in 2026.
The researchers therefore expect current demand and emissions to be higher than during the study window, while the incomplete facility list and inability to isolate AI workloads limit how precisely the total can be allocated.
Takeaway: The 2024–25 estimate is a substantial snapshot, not a complete tally of the data centers operating in 2026.
via phys.org/rss-feed/ (Source)
More from Marcus Bennett
Show full bio
News editor covering regulation, policy and their impact on general science news at SciBeat.
449 articles
Nearby plates
- The Data Centre Problem Bigger Than Water or Power: the Money
- Government-Owned Power Plants Emit Most, Respond Least to Climate Policy
- DOE's AI Push Is Squeezing Funding for Research Grants
- DOE Budget Proposal Boosts Supercomputers, Cuts Research Funding
- New Study Ties Major Fossil Fuel Emitters to Extreme Weather Risk