The Sobering Statistics
Researchers from Harvard University have provided one of the most detailed accountings yet of the environmental cost of our digital lives. By analysing 403 of the largest 'hyperscale' data centres in the United States between May 2024 and April 2025,
they uncovered some staggering figures. Together, these facilities produced an estimated 25.7 million metric tons of carbon dioxide. To put that into perspective, it's roughly equivalent to the annual emissions of the entire nation of Denmark. The study, published in the journal Nature Sustainability, found that this handful of buildings consumed around 82 terawatt-hours of electricity, which is nearly 2% of the total U.S. consumption and enough to power almost 7 million American homes for a year.
Following the Digital and Energy Trail
A major challenge in understanding the impact of data centres has been a lack of transparency from tech companies, who often treat their energy usage as a proprietary secret. The Harvard researchers worked around this by creating a novel methodology. They started by identifying the facilities using commercial real estate data and then verified their size and existence with satellite imagery. From there, they estimated each facility's power consumption based on its capacity and a range of operating scenarios. To calculate the carbon footprint, they mapped each data centre to its regional electricity grid and used U.S. Environmental Protection Agency data to determine the local energy mix. This allowed them to see not just how much power was used, but where it came from.
The Dominance of Fossil Fuels
Despite the tech industry's frequent pledges for a greener future, the study revealed a heavy reliance on fossil fuels. More than half (about 54%) of the electricity powering these 403 data centres came from sources like natural gas and coal. This finding highlights a core tension: even as companies make individual investments in renewable energy, many of their facilities are plugged into grids that are still predominantly 'dirty'. This is especially true as the demand for artificial intelligence explodes, pushing utilities to bring more fossil fuel power plants online to meet the unprecedented energy needs. The study found that some of the largest data centre clusters are in regions with carbon-intensive energy grids, meaning their environmental impact is disproportionately large.
Not All States Are Created Equal
Geography plays a crucial role in a data centre's carbon footprint. The research showed a massive concentration of energy use in just a few states. Virginia, Oregon, Ohio, and Iowa alone accounted for more than half of the total electricity consumption across all 403 sites studied. Virginia, often called 'Data Centre Alley', hosts 142 of the facilities and consumes about a quarter of the total power. However, a higher power draw doesn't always mean higher emissions. A facility in a state with a greener energy mix, like Oregon, might have a smaller carbon footprint than a less power-hungry facility in a state that relies heavily on coal. The study found that data centres in Wyoming, for instance, produce about seven times more carbon dioxide for the same amount of electricity as those in Oregon.
An Underestimate in the Age of AI
Perhaps the most alarming takeaway is that these already massive figures are likely a significant underestimate. The study’s data collection ended in April 2025, which, as lead author Gianluca Guidi noted, is 'ancient history' in the context of the AI boom. Since then, hundreds more data centres have come online, and an additional 2,000 are reportedly planned or under construction. The explosion in generative AI requires immense computational power, driving a surge in energy demand that is outpacing both sustainability efforts and researchers' ability to track it. This new data provides a crucial, if conservative, baseline for a problem that is growing exponentially.
















