Why we publish the research
We ask candidates for plain, unedited answers. We hold our own sources to the same standard.
Two of our ten line items, rural data center expansion and power grid reliability, turn on how much energy data centers really use. Each summary below says what the study found, and where it stops, so you can read the source and decide for yourself.
Peer-reviewed study, 2026, open access
From servers to services: modeling data centers as heat-active urban energy prosumers
Authors: Sai Sudharshan Ravi, Theresa Sophie Löffler, Eduardo Antonio Pina, Shivom Sharma, Dorsan Lepour, Cedric Terrier, François Maréchal, École Polytechnique Fédérale de Lausanne
Published in: Applied Energy, volume 402 (2026), article 127049
What it studies. How data centers could stop being only large electricity users and instead feed useful heat and flexible demand back into a city’s energy system. The researchers modeled five scenarios for the EPFL university campus in Lausanne, Switzerland, and compared two ways of recovering waste heat.
What it found. In that case study, data center waste heat could supply up to 40 percent of the campus heat demand, and up to 60 percent of the heat the data centers reject could be put to use. Shifting computing work to match solar output raised the share of solar power used on site by up to 28 percent in some cases.
Background it cites. Data centers use about 2 to 3 percent of global electricity, a share that has been projected to rise to between 10 and 51 percent by 2030.
What it does not cover. This study is about reusing waste heat in a Swiss campus district. It does not measure how much data centers warm the land or air around them, so it is not a source for heat island temperature figures.
Science Policy Forum, 2020
Recalibrating global data center energy-use estimates
Authors: Eric Masanet, Arman Shehabi, Nuoa Lei, Sarah Smith, Jonathan Koomey
Published in: Science, volume 367, issue 6481 (February 28, 2020), pages 984 to 986
What it studies. How much electricity the world’s data centers actually used from 2010 to 2018, built up from data on servers, storage, network equipment, and cooling rather than projected from growth in internet traffic.
What it found. Global data center electricity use rose from about 194 terawatt-hours in 2010 to about 205 terawatt-hours in 2018, around 1 percent of the world’s electricity. That is a 6 percent increase, while the computing work done in data centers grew by 550 percent. Most of the difference came from more efficient servers and storage and a shift to large cloud and hyperscale facilities.
How much each measure grew from 2010 to 2018, as a multiple of its 2010 level.
Show as a table
| Measure, 2010 to 2018 | Growth |
|---|---|
| Data storage capacity (estimated) | 25 times |
| Internet traffic at data centers | More than 10 times |
| Compute instances | 6.5 times (up 550 percent) |
| Electricity use | 1.06 times (up 6 percent) |
Source: Masanet, Shehabi, Lei, Smith, and Koomey, Science 367 (2020), pages 984 to 986.
Read it in context. This analysis uses data through 2018, before the recent wave of AI data center construction. The authors warned that energy demand could grow sharply once today’s efficiency gains are used up, and they called for public reporting of data center energy use for transparency and accountability.
Working paper, 2026
Measuring the impact of data centers in the United States economy
Author: Nicholas Z. Muller, Carnegie Mellon University
Published as: National Bureau of Economic Research Working Paper 35100, April 2026
What it studies. It puts a dollar figure on the air pollution and greenhouse gases from the electricity data centers use, and splits that figure by state.
What it found. Data centers use roughly 250 terawatt-hours of electricity a year, about 5 to 6 percent of all United States generation, and cause something near $25 billion a year in damages, within a range of $10 billion to $33 billion. Texas and Virginia together carry about 30 percent of the national total.
Why it matters here. This is the study that puts Texas specifically in the frame. It is also the clearest answer to the argument that a data center’s costs land somewhere else.
Read it in context. A working paper has not been through peer review yet. Damage figures also depend on which social cost of carbon the author uses, and he shows a range rather than one number for that reason.
Expert discussion, 2026
Data centers, urban heat, and AI growth
Published by: Yale School of the Environment, July 10, 2026, with Karen Seto, Daniel Esty, Kenneth Gillingham, and Yuan Yao
What it covers. Nearly all the electricity a data center draws turns into heat that has to be moved somewhere. This is a discussion of where that heat goes and what it does to the neighborhood around it.
What it reports. Work on data centers in Phoenix found waste heat raising land surface temperatures nearby by as much as 16 degrees Fahrenheit, and air temperatures in neighborhoods close by by as much as 4 degrees, measurable up to half a kilometer away, which is about five city blocks.
Why it matters here. It fills the gap we flagged in the Applied Energy study above, which is about reusing waste heat rather than measuring how much a data center warms the area around it.
Read it in context. The temperature findings come from Phoenix, a desert metro. Texas summers are hot but not identical, so treat the figures as a warning to check locally, not as a Texas measurement.
Peer-reviewed, 2025
The carbon and water footprints of data centers, and what this could mean for artificial intelligence
Published in: Patterns, Cell Press
What it studies. The carbon and the water behind the computing itself, and what happens to both as AI workloads grow.
What it found. The authors estimate the carbon footprint of AI systems at between 32.6 and 79.7 million tons of carbon dioxide in 2025, which they compare to the footprint of New York City, and a water footprint in the range of the world’s annual consumption of bottled water.
Read it in context. These are global estimates with a wide range, not a measurement of any one facility, and the authors present them as a range for that reason.
Peer-reviewed, open access, 2026
Health implications of the rapid rise of data centers in Virginia
Authors: Neha Gour, Luis Ortiz, Edward Maibach
Published in: Frontiers in Climate, February 6, 2026
What it studies. What the concentration of data centers in northern Virginia does to the health of the people living around them.
What it found. It traces several separate paths: air pollution from the power plants and from diesel backup generators, freshwater use, noise from equipment that runs every hour of every day, land cleared for construction, and energy costs that fall hardest on low income households. Virginia data centers draw roughly 5,050 megawatts, a figure the authors expect to double by 2040.
Why it matters here. Virginia is the closest thing America has to a preview. Texas is now on the same path, which is what the Data Center Moratorium and Local Impact Study Act asks communities to study before, rather than after.
Read it in context. The authors call this an exploratory assessment. It maps the ways harm could travel and does not claim to measure health outcomes in Virginia residents.
Policy report, 2025
What happens when data centers come to town?
Authors: Terry Nguyen and Ben Green, Science, Technology, and Public Policy Program, Gerald R. Ford School of Public Policy, University of Michigan, July 2025
What it studies. What actually lands on a town after a data center arrives, and which policies change that.
What it found. New demand pushes utilities into costly upgrades that get passed to residents as higher rates. Tax breaks written to attract data centers reduce local revenue, including school district revenue, without producing the promised number of lasting jobs. Cooling technology currently forces a choice between saving energy and saving water, so a facility tends to improve one at the cost of the other.
Why it matters here. It is a checklist of the things a local impact study would have to look at, written before Texas had a bill asking for one.
Read it in context. The report draws mainly on Michigan, Virginia, and Nebraska, not Texas. One of its summary lines also mixes up units of power and energy, so we have not repeated any figure from it here.
How we share research
- Open access papers are posted in full when their license allows it, with the authors and license credited.
- Copyrighted papers are summarized here and linked at the publisher, never reposted.
- Every summary says what a study does not cover, so a finding never gets stretched past what the authors showed.
Know of a study we should add? Send it to us.