Every time you ask ChatGPT a question, generate an image with Gemini, or have Claude help write an email, a data center somewhere is doing the work. Most of us never think about what it takes to keep those servers running, but new research suggests America's AI infrastructure already has an enormous environmental cost.
A new Harvard-led study published in Nature Sustainability finds that 403 of the largest U.S. data centers consumed about 1.8% of the country's total electricity between May 2024 and April 2025.
That's an estimated 82 terawatt-hours of electricity in a single year, enough to power nearly 7 million American homes for a year, according to an Associated Press report.
And electricity consumption is only part of the problem. Researchers estimate the facilities produced 25.7 million metric tons of carbon dioxide emissions, roughly equivalent to Denmark's annual emissions.
Companies are racing to build more infrastructure to support increasingly powerful AI models, but questions remain over what the technology's rapid growth means for the environment and the people living near the facilities powering it.
Using satellite imagery, facility information and federal electricity-generation data, the team estimated how much power each location consumed and the resulting carbon emissions.
Approximately 54% of the electricity attributed to the data centers came from fossil-fuel generation, including 38% from natural gas and 15% from coal.
The estimated emissions were also two to three times higher than an earlier assessment of hyperscale data-center emissions in 2018. The study's measurements ended in April 2025, which means they don't even account for the additional data centers built or expanded over the past year and a half.
Source link







