352
Productivity & Workflow352
Software Development247
Automation & Workflow239
Marketing & Growth204
AI Infrastructure & MLOps193
Photography & Imaging166
Writing & Content Creation192
Design & Creative171
Data & Analytics165
Customer Support155
Education & Learning142
Voice & Speech130
Sales & Outreach132
Operations & Admin119
A Harvard-led team estimates 403 large US data centers used 93.66 TWh of electricity and caused 52.69 million metric tons of CO2 in 2024–2025.
In short: A Harvard-led study estimates that 403 of the largest US data centers used huge amounts of electricity and were linked to 52.69 million metric tons of CO2 over one year.
Researchers estimated that 403 US “hyperscale” data centers, meaning very large facilities that run cloud and internet services, emitted 52.69 million metric tons of carbon dioxide from May 2024 through April 2025. They also estimated these sites used 93.66 terawatt-hours of electricity over that period. A terawatt-hour is a very large unit of power, like counting electricity use by the city instead of by the home.
The study says the emissions number is a conservative lower bound. The reason is that the 403 facilities in the sample are estimated to represent about 78% of all US hyperscale data center electricity use, not 100%.
A key challenge was missing public information. The team combined many data sources, checked facility locations using satellite images, estimated electricity use, and then matched each site to the power plants that likely supplied its electricity. For six facilities with missing capacity details, they used a statistical model, meaning an educated estimate based on patterns in similar sites.
The study counted emissions from the electricity the data centers consumed, even if the pollution happened at a power plant off-site. It estimated an average carbon intensity of 0.56 kilograms of CO2 per kilowatt-hour, and found that nearly 60% of the emissions came from fossil fuels.
These numbers cover hyperscale data centers only, not every data center in the US. They are also not a direct measurement of pollution caused by AI work alone. Next, watch whether operators and regulators push for more consistent disclosure, since better reporting would make future estimates less guesswork.
Source: NYTimes