Hyperscale data centers use 2% of US electricity

Study attempts to tally hyperscalers’ energy use and emissions.

One flashpoint for the tech industry’s all-in push on Large Language Models has been the sudden rush to turn investment money into hyperscale data centers. The rush hasn’t allowed anyone to plan how to supply the required electricity cleanly or with minimal impact on ratepayers. Instead, it has been difficult to even collect information on what’s being built.

A new study has tallied up energy use and estimated emissions for hyperscalers operating between May 2024 and April 2025 in the United States. They weren’t able to differentiate between AI compute and more traditional data services, but it provides a clear point of reference that could be tracked moving forward.

Finding the centers

The team started with a list of 675 facilities from a commercial database based on news reports, company statements, and satellite imagery. They dropped any that didn’t have a reported electrical capacity or that they couldn’t verify was a suitably large building with the expected cooling and electrical connection equipment. That reduced the list to 403 data centers with an average electrical capacity of about 40 megawatts.

In a way, that makes this a very conservative estimate, since 272 facilities believed to be large data centers were left out. But if you’re interested in AI data centers, specifically, it’s also true that some of the facilities included may not fall in that category.

That fact also introduces some uncertainty in the power usage of these data centers. Relying on other research, they use a range—some conventional data centers run at an average of about 48 percent of capacity, while AI training loads have been running higher, at around 70 percent.

Putting that together, the total power usage for the year by these 403 data centers was estimated at 68 to 99 terawatt-hours. That amounts to 1.5 to 2.2 percent of total US electricity use, on par with the total used by Wisconsin or Indiana.

A recent Lawrence Berkeley National Laboratory report estimated the total electricity use of data centers of any kind in the US (minus crypto mining) at 76 terawatt-hours in 2018. In 2024, that number was up to 192 terawatt-hours. That means the hyperscale data centers identified by this study account for something like one-third to one-half of all data center energy use.

Electrons come with carbon

The researchers also wanted to estimate the greenhouse gas emissions associated with generating all that electricity. It’s fairly straight-forward to get a reasonable number here—calculate the mix of generation in each data center’s region of the grid, and apply that CO2 per watt-hour rate to a data center’s electricity use.

This yields 21 to 31 million metric tons of CO2 for the year. That’s around 0.5 percent of total US fossil fuel CO2 emissions.

These data centers aren’t located in areas of especially clean energy—their energy mix was only slightly cleaner than the US average. About 54 percent of the electricity was generated by fossil fuel power plants (dominantly natural gas), with 25 percent from renewables, and 21 percent from nuclear.

This doesn’t include any “behind-the-meter” generation on-site, though. The xAI data center in Tennessee notably ran a fleet of natural gas turbines during this study’s time period, for example. Cleanview, which describes itself as a market intelligence company, has tracked a boom in on-site generation since early 2025, including 116 current or planned data centers. While this may not affect this study’s numbers much, it is becoming increasingly significant over time.

This matters because the types of generators being used are much less efficient than utility power plants, emitting 60 to 80 percent more greenhouse gas to produce the same amount of electricity. (Conversely, if a data center built solar generation instead, it could reduce the carbon footprint of its energy.)

Although this study can’t pinpoint the energy use of AI data centers alone, repeating this analysis in subsequent years could reveal the growth related to the AI buildout—at least as long as that buildout lasts.

Nature Sustainability, 2026. DOI: 10.1038/s41893-026-01985-7 (About DOIs).

Original source Hyperscale data centers use 2% of US electricity

Back to home