The rapid expansion of fossil-fuel-powered data centres, particularly those supporting artificial intelligence (AI) applications, risks reversing 15 years of greenhouse gas emission reductions in the United States power sector, according to an internal analysis by the Sierra Club.
The environmental organization projects that the US power sector’s carbon dioxide emissions could reach 2.1 billion metric tonnes by 2030 due to planned AI-related data centre projects. This represents an increase of approximately 618 million tonnes, or 42 percent, compared to 2025 levels. Jeremy Fisher, the Sierra Club’s principal adviser for climate and energy, noted that earlier estimates made around 18 months ago were significantly lower, forecasting only a quarter to half of the current projected emissions growth.
A significant portion of the expected emissions growth is attributed to off-grid power sources. Many data centre developers are turning to natural gas-fired plants that operate independently from the main electrical grid to meet the substantial and immediate energy demands. These off-grid facilities alone could generate roughly 460 million tonnes of carbon dioxide annually by 2030, equivalent to the yearly emissions of Mexico, according to Fisher.
BloombergNEF analyst Helen Kou echoed concerns, stating that substantial increases in gas-powered electricity generation for these new facilities would likely raise emissions from current levels. However, considerable uncertainty remains regarding the total scale of the buildout, the technological choices made, and the amount of renewable energy capacity added in the near future. The Sierra Club’s projections are notably higher than some other assessments, illustrating the wide range of possible outcomes.
In contrast, the Union of Concerned Scientists estimated in January that fossil fuel-based electricity for data centres could increase power-sector emissions in the US by 19 to 29 percent, which would amount to as much as 342 million tonnes annually by 2035. This lower figure highlights the variability in forecasts depending on different assumptions about facility build rates, fuel sources, and energy efficiency measures.
Brendan Pierpont, director of electricity at the Energy Innovation policy think tank, emphasized that if the full suite of projected data centres were built on schedule—many equipped with inefficient natural gas turbines running around the clock—power sector emissions could rise significantly. He cautioned that such an increase could erase years of progress in reducing emissions, noting, “There’s certainly a risk that that’s enough emissions to push us back many years of our progress on reducing power-sector emissions.” However, Pierpont also stressed that this outcome is not inevitable.
The potential surge in emissions comes as the US power sector has been steadily reducing its carbon footprint through a combination of cleaner technologies and policy efforts over the past two decades. The developments in AI infrastructure, if powered primarily by fossil fuels, could challenge these climate gains, raising critical questions about balancing technological growth and environmental sustainability.