AI Data Centers Delay Coal Plant Retirements, Extend Fossil Fuel Use

The explosive growth of AI chatbots is creating an unexpected obstacle to America's clean energy transition: a wave of data centers that require so much electricity they're keeping aging fossil fuel plants online and driving construction of new gas infrastructure.

At least 15 fossil fuel power plants scheduled for retirement have had their closures postponed to meet surging data center demand, according to a new report from the Frontier Group. Those facilities alone emit more climate pollution than the entire state of Massachusetts produces annually.

Why it matters

The AI boom threatens to derail U.S. plans to phase out coal power by 2040. As tech companies race to deploy large language models like ChatGPT, Claude, and Gemini, the infrastructure required to run them is locking in fossil fuel dependence for years or decades longer than planned — directly conflicting with climate goals.

Training models consumes massive energy

Generative AI tools require extraordinary amounts of electricity both to train and operate. Training a single large language model like ChatGPT-3 can consume up to 10 gigawatt-hours of power — roughly equivalent to the annual electricity use of over 1,000 U.S. households, according to AI researcher Sajjad Moazen at the University of Washington.

This demand is concentrated in data center hubs like Loudoun County, Virginia, home to over 200 facilities. Sarina Virmani, a local high school student who published research on data center environmental impacts, noted that many people don't realize AI "lives in these massive buildings" with tangible physical footprints.

Zombie power plants and new gas infrastructure

Quentin Good, coauthor of the Frontier Group report, describes the phenomenon as "zombie power plants" — facilities that should be retired but are being kept operational to serve data centers. In southern Virginia, one coal plant originally scheduled to close in 2025 now has its operations extended indefinitely. In 2023, that single facility emitted pollution equivalent to more than 65,000 gasoline-powered vehicles.

Beyond延ed retirements, entirely new fossil fuel infrastructure is being constructed. Wired reported in April 2026 that new gas plants linked to just 11 U.S. data center campuses could generate more climate pollution than the entire country of Morocco emitted in 2024.

Some data centers also rely on diesel generators as backup power during peak demand periods, adding another pollution source.

Water use varies by location

Data centers require substantial water volumes to cool servers, though the environmental impact depends heavily on location. In Virginia, Good found water use isn't currently a crisis, though drought conditions could change that calculation.

In water-stressed regions like parts of Colorado, data center water demand poses more serious concerns, particularly during hot, dry months when river flows are lowest and competing water needs peak. Discharged water also carries elevated sodium levels and heat that may affect aquatic ecosystems — an area Good identified as needing more research.

What can be done

Despite a Trump administration executive order in late 2025 attempting to limit state regulation of data centers, NPR has reported bipartisan support in Congress and state governments for AI industry oversight.

At the local level, residents have successfully shut down projects, passed moratoriums, and blocked electricity rate increases through organized opposition at public hearings. Individuals can also reduce unnecessary use of generative AI tools to lower demand for energy-intensive systems.

These details were first reported by Yale Climate Connections, based on reporting by Samantha Harrington.