Harvard researchers collected metrics on more than 400 data centers to determine their energy use and carbon emissions.
During the midterm elections, Americans made one thing very clear: Data centers must be reined in.
A vast majority of communities across the country voted to ban or restrict the construction of data centers, as concerns about their impact on the environment and public health grow.
A new study by Harvard researchers shows that these fears are not unfounded. The findings published today in Nature Sustainability show that the carbon emissions and energy use of hyperscale data centers amount to as much as those of entire states and more than those of the country of Denmark.
I spoke with Gianluca Guidi, a postdoctoral fellow in the Department of Biostatistics and lead author of the paper, to discuss how the team collected data on energy use despite a lack of industry transparency, the rapid evolution of AI infrastructure and future research goals.
Hyperscale data centers are the largest and most powerful data center facilities, featuring thousands of servers to power cloud computing, model training and the global use of AI. Guidi said he was inspired to look into data center emissions and the broader impacts of generative AI after reading a paper on bitcoin mining’s electricity use, which found that just 34 mining facilities, combined, consumed as much electricity as the entire city of Los Angeles.
The study narrowed in on 403 of these facilities to provide the most up-to-date and transparent measures of their electricity use and carbon emissions between May 2024 and April 2025.
Source link







