Putting the Numbers in Context
The figure of 25.7 million metric tons of carbon dioxide is staggering on its own, but its true scale becomes clearer with comparison. This level of emissions, generated between May 2024 and April 2025, is equivalent to the entire annual carbon pollution
of Denmark, a country of nearly six million people. The study, published in the journal Nature Sustainability, represents one of the first major peer-reviewed attempts to quantify the emissions of these 'hyperscale' data centres since the boom in artificial intelligence began. The electricity consumed by these facilities over the study period was estimated to be between 68 and 99 terawatt-hours, accounting for nearly 2% of total US electricity use. That's more power than New York City and Los Angeles combined.
The AI Boom and an Unquenchable Thirst for Power
The primary driver behind this explosive growth in energy consumption is the artificial intelligence revolution. Training and running complex AI models requires immense computational power, which in turn demands vast amounts of electricity. The International Energy Agency (IEA) has projected that global electricity demand from data centres could double between 2022 and 2026, largely fueled by AI adoption. This new Harvard research shows a footprint two to three times higher than a previous estimate from 2018, underscoring the rapid acceleration. Researchers noted that their figures are likely a conservative estimate, as the pace of data centre construction has only increased since the study period concluded in April 2025.
Not All Energy Is Created Equal
A crucial finding of the study is the source of this power. Over half—54%—of the electricity consumed by these massive facilities came from burning fossil fuels, primarily natural gas and coal. This reliance on carbon-intensive energy sources is the direct cause of the high emissions figures. The report highlighted significant regional differences; for instance, a data centre in Wyoming, where coal is a major power source, generated more than twice the carbon for the same amount of electricity as one in Oregon, which benefits from hydropower. Virginia, Iowa, and Ohio were identified as hotspots, with data centres in those three states alone accounting for about half of the total estimated emissions.
A Challenge of Transparency
The researchers behind the study had to engage in significant 'detective work' to compile their data, highlighting a major issue in the industry: a lack of transparency. Many tech companies do not publicly disclose the electricity consumption of individual data centres, making it difficult for independent bodies to accurately assess their environmental impact. To conduct the study, the Harvard team used a commercial database and satellite imagery to identify and confirm the existence of 403 hyperscale facilities before estimating their power use and calculating emissions based on the local energy grid's mix. This opacity has become a point of contention, with growing calls for regulation requiring companies to report their energy and water consumption, similar to new rules in the EU.
The Industry's Response and the Path Forward
In response to mounting pressure and their own climate goals, major tech companies are making significant investments to mitigate their environmental impact. Google, Microsoft, and Amazon Web Services have all set ambitious targets to power their operations with 100% renewable energy. These efforts often involve long-term power purchase agreements (PPAs) for wind and solar energy, on-site generation, and investments in energy efficiency and advanced cooling technologies. For example, Google reported that despite a 27% jump in electricity use in 2024, its data centre emissions fell by 12% due to clean energy procurement and efficiency gains. However, the sheer growth of AI demand is outpacing the deployment of clean energy in many regions, creating a carbon debt that complicates net-zero ambitions.
















