U.S. AI data centers emit carbon levels rivaling entire countries
A new study reveals that hyperscale data centers produced 25.7 million metric tons of CO2 last year, driven largely by the rapid expansion of AI.

The short version
- U.S. hyperscale data centers emitted 25.7 million metric tons of CO2 between May 2024 and April 2025.
- The emissions are two to three times higher than previous estimates from 2018.
- About 54% of the electricity powering these facilities comes from fossil fuels, primarily natural gas and coal.
- The power demand for these centers reached about 1.8% of the total electricity consumption in the United States.
- Virginia, Oregon, Ohio, and Iowa accounted for the majority of the electricity demand studied.
Why it matters
As AI becomes integrated into daily life, the physical infrastructure required to run it is creating a significant environmental footprint. This data suggests that the growth of the AI industry is creating immediate, measurable challenges for national carbon reduction goals.
The scale of the carbon footprint
A new study published in Nature Sustainability reveals the massive environmental cost of the modern internet. Researchers found that more than 400 hyperscale data centers in the United States produced approximately 25.7 million metric tons of carbon dioxide during the period from May 2024 to April 2025. This figure is significant enough to rival the annual carbon emissions of some entire countries.
To put this scale into perspective, the report notes that these emissions are roughly equal to the annual output of Denmark or Bolivia. This represents a sharp increase from previous years, as the study found these figures to be two to three times higher than estimates from 2018. The growth is largely attributed to the rapid expansion of cloud computing and the training of large language models.
The study highlights a growing tension between technological advancement and climate goals. As companies race to deploy more AI tools, the physical infrastructure required to host them is consuming resources at an unprecedented rate. The data shows that the carbon footprint of these facilities is no longer a niche concern but a significant portion of the national energy profile.
How the data was gathered
The research team, organized by the Harvard T.H. Chan School of Public Health and the Harvard Data Science Initiative, used a multi-step process to track these emissions. They first utilized satellite imagery to identify and build a comprehensive list of hyperscale data centers across the country. These are the largest class of data centers, which provide the backbone for modern AI and cloud services.
Once the facilities were identified, the researchers used data from the Environmental Protection Agency to assign carbon emissions to each site. They did this by determining which specific power plants fed the regional electricity grids where the data centers were located. This allowed them to calculate the carbon intensity of the power being drawn by each facility based on its geographic location.
The study found that the power demand for these centers ranged from 68 to 99 terawatt-hours. This amount of electricity is substantial, equaling about 1.8% of the total electricity consumed by the entire United States. The researchers noted that the upper end of this estimate is nearly as much as the annual power consumption of states like Indiana, Michigan, and Tennessee.
Reliance on fossil fuels
Despite the push for green energy in the tech sector, the study shows a heavy continued reliance on traditional power sources. Approximately 54% of the electricity used by these hyperscale facilities comes from fossil fuels. Specifically, natural gas accounted for 38% of that power, while coal provided another 15%.
The research also pointed out that a quarter of the electricity used by these centers comes from grid regions that are considered dirtier than the national average. This suggests that data centers are often placed in areas where cheap, carbon-intensive power is readily available. The study identified Virginia, Oregon, Ohio, and Iowa as the regions representing the majority of the electricity demand covered.
The findings underscore the difficulty of decarbonizing the AI boom quickly. While many tech giants have pledged to reach net-zero goals, the current reality of the grid shows that a majority of the energy supporting AI tools is still carbon-heavy. This creates a significant hurdle for companies trying to balance rapid scaling with environmental responsibility.
Limitations and future outlook
The researchers acknowledged certain limitations in the scope of the study. Because of data availability, they were unable to examine every single hyperscale data center in the country. Furthermore, the current data did not allow the team to distinguish between general-purpose computing and operations specifically dedicated to AI at each individual facility.
Despite these limits, the authors warn that the situation is likely worsening. Lead author Gianluca Guidi stated that the data center boom has not slowed down since the study period ended in April 2025. He noted that the electricity use and carbon footprint reported in the study could already be substantially larger by the year 2026.
The next phase of research will likely focus on the human element of this infrastructure growth. Corresponding author Francesca Dominici, director of the Harvard Data Science Initiative, stated that measuring health effects is a next step for the team. She emphasized that because these facilities are growing so fast, the decisions regarding where they are built and how they are powered are critical public health decisions that require high-quality data.
What to watch
- How data center operators will respond to new carbon reporting requirements.
- Whether new federal policies will incentivize moving data centers to 'greener' grid regions.
- The development of new cooling technologies to reduce the overall power demand of AI chips.
Try it
Look up the electricity mix of your own state to see what percentage of your local grid is powered by fossil fuels versus renewables.
Sources and links
- Reportingqz.com: the article we sourced
First reported by qz.com. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.