Field report W-25 / K6 · sheet /macro
Boardroom MacroFiled from the territory
Harvard Study: Hyperscale Data Centers Now Outdraw Entire States
Hyperscale data centers used up to 99 TWh in a year — rivaling Indiana or Michigan — and emitted 25 million metric tons of CO2, a Harvard study finds.
- Grid
- W-25 / K6
- Surveyed
- November 16, 2026
- Reading
- 4 min read
- Correspondent
- Olivia Hart

Field notes
- 403 hyperscale data centers used 68–99 TWh of electricity from 2024 to 2025, rivaling states like Indiana and Michigan (~100 TWh each).
- The fleet emitted over 25 million metric tons of CO2 in that period, about three times its 2018 emissions.
- Fossil fuels supplied 54% of the power mix: 38% natural gas and 15% coal.
- The study, published in Nature Sustainability, is the first peer-reviewed hyperscaler emissions accounting since 2018.
- Researchers compiled the data via satellite imagery, permit scraping and EPA data, and published findings at neverjustabuilding.org.
America's hyperscale data centers consumed between 68 and 99 terawatt-hours of electricity in a single year — as much as or more than entire states such as Indiana, Michigan and Tennessee, each of which uses roughly 100 TWh annually. The figure comes from a two-year Harvard-led study published in Nature Sustainability, the first peer-reviewed accounting of hyperscaler emissions since 2018, before the AI buildout accelerated.
The stakes for executives are straightforward. The same fleet emitted more than 25 million metric tons of carbon dioxide from 2024 to 2025, roughly triple its 2018 emissions. Most of that power came from fossil fuels. For anyone siting, financing or powering AI infrastructure, the study establishes a public baseline that regulators, utilities and communities can now cite — and the researchers built it without any cooperation from the operators themselves.
Francesca Dominici, the study's author and director of the Harvard Data Science Initiative, said there has been little peer-reviewed science on data center emissions despite mounting public scrutiny. To her knowledge, the only prior peer-reviewed work on hyperscaler greenhouse gas emissions and electricity consumption dates from 2018.
How did researchers measure a black box?
Dominici and her co-researchers identified 403 hyperscale data centers using satellite imagery, then scraped web filings, applied remote sensing and drew on Environmental Protection Agency data to calculate emissions based on the mix of power plants surrounding each facility. The method itself, she argues, reveals the problem.
The workaround "tells you about the fact that there is a lot of secrecy about these facilities," Dominici said. "Why can't we just have a database which would tell us where are the facilities, how big they are, what's the total capacity?"
What does the electricity figure compare to?
One terawatt-hour powers roughly 90,000 homes. Against that yardstick:
- The hyperscale fleet's annual draw matches or exceeds South Carolina (about 84 TWh) and dwarfs Oregon (60 TWh) and Maine (11 TWh).
- Texas, California and Florida remain far larger consumers, at 475, 251 and 248 TWh respectively.
The scale surprised even the lead researcher. "It's something so much bigger that I was actually not even expecting myself," Dominici said.
The numbers are also, in her words, likely an underestimate. The paper covers 2024 to 2025, so the data is already a year old. "So much more construction and development of hyperscaler data centers has been happening just in the last year," she said.
Why is the power mix so carbon-heavy?
Fossil fuels supply about 54% of the electricity feeding these facilities: natural gas accounts for 38% and coal for 15%. About a quarter of the fleet's electricity comes from grid regions dirtier than average, where coal-fired plants dominate. The specific grids data centers plug into "really matter" to their overall impact, Dominici said.
The research also shows, she added, that "there has been no attention on renewable sources." Wind and solar remain the cheapest sources to build, but neither delivers constant high power without battery storage. Fossil fuels still offer the fastest route to quick, stable capacity — the core tension every data center energy strategy now confronts.
What happens next?
Alongside the paper, the team launched neverjustabuilding.org, a website illustrating the findings. The researchers say they want to reach everyone touched by data center development: communities near construction sites, policymakers, utilities and the developers themselves. The site also lets visitors sign up for updates on forthcoming research.
This study is the first in a series. Follow-up work will examine not only emissions and energy use but noise, water consumption and the health impacts of data center pollution.
For practitioners, the framework is the point. "This gives us now a framework . . . to start not only tracking," Dominici said, "but also to really start identifying what we can do and where we should build and how to build, so [data centers] will have less of an impact."
The message for operators is that the measurement gap has closed. Independent researchers have demonstrated they can size the fleet, locate its power sources and price its emissions without disclosure — which makes voluntary transparency a cheaper option than waiting for the next satellite pass.
via fastcompany.com/rss (Source)
More from Olivia Hart
Show full bio
News editor covering regulation, policy and their impact on corporate strategy and chief-executive leadership at CEO Frontier.
456 articles
Adjacent sheets
- Data Center Gas Plants Could Add 10% to US Power Emissions
- Data Centers Need Community Stakes and Regulatory Accountability
- Midterm Voters Rejected Data Centers. A Clean Energy Standard Is the Answer, Climate Advisers Says
- Nvidia Urges Treating Data Centers as Community Assets
- Restricting Data Centers Is a Strategic Mistake: The Case for Building Megawatts