Policy
U.S. Data Centers Emitted 28 Million Tons of CO2 Before AI Boom
Harvard study reveals the 400 largest facilities matched Denmark's annual emissions, prompting new voluntary climate pledges at COP31.
Key takeaways
- The 400 largest U.S. data centers emitted 28.3 million tons of CO2 annually before AI's full expansion—matching Denmark's total national emissions.
- Harvard researchers found emissions were two to three times higher than previous estimates, with plans for 2,000 additional data centers raising future concerns.
- A new voluntary pledge at COP31 calls for clean power, efficient water use, and greater transparency from data center operators.
- Tech company electricity consumption jumped 240% between 2022 and 2025, preventing major firms from meeting climate commitments.
- Capturing waste heat from data centers could supply approximately 10% of European homes by 2030, according to the International Energy Agency.
The 400 largest data centers in the United States generated approximately 28.3 million tons of carbon dioxide between May 2024 and April 2025—equivalent to Denmark's entire national emissions—according to new research from Harvard University published in Nature Sustainability. The findings arrive as the artificial intelligence sector prepares for massive expansion, raising concerns about technology's climate impact.
The facilities consumed 82 terawatt-hours of electricity during the study period, enough to power nearly 7 million American homes for a year. More than half of that power came from fossil fuels, researchers found. The emissions estimate runs two to three times higher than previous studies had indicated, said senior author Francesca Dominici, a Harvard biostatistics professor.
Measurement challenges and modeling
Harvard researchers analyzed 403 of the largest U.S. data centers using available data combined with modeling techniques. The modeling approach became necessary because technology companies typically do not disclose detailed information about their facilities' power consumption and emissions.
Several independent experts in climate science, energy systems, and AI told the Los Angeles Times that the study's methodology and conclusions appeared sound. Cornell University energy systems scientist Fengqi You, who was not involved in the research, emphasized that reliable measurement and disclosure are essential for managing environmental impacts effectively.
Dominici noted that plans for 2,000 additional data centers make the emissions trajectory particularly concerning. "We are going backwards toward the goal of controlling CO2 emissions," she said.
New voluntary pledge at climate talks
At the COP31 climate negotiations in Antalya, Turkey, officials launched the Antalya Artificial Intelligence Pledge this week. The voluntary agreement, proposed by host nation Turkey and the International Telecommunication Union, calls for data centers to use clean power sources, implement efficient energy and water systems, and increase transparency about their environmental costs.
U.N. climate chief Simon Stiell emphasized the urgency during the pledge's Monday launch: "Claims about future emissions savings cannot excuse rising pollution now, making climate disasters even worse."
COP31 President Murat Kurum stated that world leaders must openly address AI's growing energy consumption and establish governing terms.
The efficiency paradox
Technology companies' electricity consumption increased 240% between 2022 and 2025, according to Climate Action Against Disinformation. This surge has prevented major tech firms from meeting their stated climate goals despite investments in renewable energy.
Former Vice President Al Gore offered a more optimistic perspective in an Associated Press interview, citing research by economist Nicholas Stern predicting that AI could enable a 6% to 9% reduction in greenhouse gas emissions by 2030 through improved efficiency in materials and energy use.
However, Climate Action Against Disinformation spokesman Philip Newell expressed skepticism: "They're selling us hope while hiding the harms from us."
Heat recovery as mitigation
One practical approach to reducing data centers' environmental impact involves capturing their waste heat for use in nearby buildings. The International Energy Agency estimates that data centers could supply enough heat for approximately 10% of European homes by 2030 if fully integrated into district heating systems.
Danish engineering company Danfoss has identified waste heat as "the world's largest untapped energy source," noting that data centers provide particularly consistent heat output as their power consumption grows. In Texas and North Dakota, industrial development firm BaRupOn is planning facilities designed to recapture heat for electricity generation and community heating.
Why it matters
The study arrives at a critical juncture as governments and businesses make trillion-dollar infrastructure decisions about AI deployment. With thousands of new data centers planned, the technology sector's emissions trajectory could significantly impact national and global climate targets. The research provides the first comprehensive baseline for measuring progress—or regression—as the AI industry scales, making transparency and accountability possible where previously only corporate estimates existed.
These details were first reported by the Los Angeles Times, with research by Seth Borenstein and Jennifer McDermott of the Associated Press.