
Electricity prices have risen sharply across the U.S. in recent years. (Image by Zenobillis on Shutterstock)
Coal Could Make a Comeback Thanks to AI and Crypto Power Demands, Study Warns
In A Nutshell
- AI and cryptocurrency data centers could push U.S. power-sector CO2 emissions up to 28% higher by 2030 than a future with no new digital infrastructure growth.
- Rather than turning to wind and solar, the grid is expected to lean on existing coal and natural gas plants to meet the surge in demand.
- Some regions, particularly Northern Virginia, could see wholesale energy costs rise by as much as 57%, with ordinary consumers potentially feeling the impact over time.
- Restoring federal clean energy tax credits would significantly reduce the fossil fuel share of new data center electricity, cutting natural gas’s contribution from 70% down to 41%.
Coal plants expected to fade further from the grid could get a second life. In Northern Virginia, one of the world’s biggest concentrations of data centers, a new modeling study suggests surging electricity demand from artificial intelligence and cryptocurrency mining could push aging coal facilities into heavier use, along with the pollution that comes with them. Researchers warn this is a preview of what could happen across the country if the electricity demands of the digital economy keep growing unchecked.
Scientists at North Carolina State University and Carnegie Mellon University published a study in Environmental Research Letters using a detailed computer model of the U.S. power grid to project how expanding data centers and cryptocurrency mining could reshape the country’s electricity system by 2030. Power-sector carbon dioxide emissions in 2030 could be up to 28% higher than in a future with no new data center growth, largely because the grid would lean on natural gas and coal plants to keep up. In some regions, modeled wholesale energy costs could rise by as much as 57%.
Data centers, the massive warehouse-like buildings that power everything from cloud storage to ChatGPT to Bitcoin, already consumed 176 billion kilowatt-hours of electricity in 2023, about 4.4% of all U.S. electricity use. That figure could more than triple by 2028, depending on how fast demand grows.
AI and Crypto Are Straining the Power Grid
Rather than assuming a single future, the team tested 13 different scenarios, varying how fast data centers grow, where they’re located, how much natural gas costs, and whether federal clean energy incentives are in place. Under a middle-of-the-road growth scenario, total data center and cryptocurrency electricity demand could reach roughly 790 billion kilowatt-hours by 2030, more than four times the 2023 level.
When pushed, the grid doesn’t reach for the cleanest options first. It reaches for what’s available and cheap. In the middle-growth scenario, coal contributes between 12% and 14% of all the extra electricity generated to meet data center demand, while natural gas supplies between 64% and 76%. Wind and solar account for a much smaller share, and most of that extra fossil fuel power comes not from newly built plants but from running existing coal and gas facilities much harder than they’d otherwise be used.
Old Coal Plants, New Life in Northern Virginia
Perhaps the starkest regional finding involves Northern Virginia. When data center demand surges in the model, coal plants in neighboring Ohio and West Virginia ramp up and ship more power east. Virginia may set ambitious clean energy goals, but it can’t fully control whether its electricity is powered by coal burning in a neighboring state. Researchers call this “carbon leakage,” where emissions are effectively exported from the demand center to wherever the cheaper, dirtier power happens to be.
Texas tells a somewhat different story. The Lone Star State accommodates data center growth primarily through expanded natural gas generation, partly because of a different grid structure and resource mix. Nationally, the picture is sobering either way: under the mid-growth scenario, the additional CO2 emissions tied to data center demand in 2030 would total 330 million metric tons, larger than the annual emissions of many entire countries.
What AI Data Center Demand Could Mean for Electricity Costs
Retail electricity rates in major grid regions from Illinois to New Jersey already rose between 23% and 40% from 2020 to 2024, according to figures cited in the study. The researchers warn their results point toward continued upward pressure where data centers are clustering. The study measures wholesale power costs, a generation-cost metric that doesn’t capture every component of a monthly household bill but can feed into retail prices over time.
At the national level, energy costs rise by roughly 6% to 29% across scenarios. Central and Northern Virginia face increases of up to 57%, because the last bit of power needed to meet peak demand there is often supplied by expensive, aging coal or gas plants that wouldn’t otherwise run much.
Federal policy turns out to matter a great deal. The 2022 Inflation Reduction Act’s clean energy tax credits, repealed for new facilities starting after 2027 per the paper, kept natural gas’s share of additional data-center electricity to 41% when reinstated, down from 70% without them. Wind and solar fill a much larger portion of the gap when incentives are on the books.
Natural gas prices also deliver a counterintuitive result. Cheap gas leaves coal plants sitting idle, and when data center demand surges, those parked plants become the go-to source of extra power. Low gas prices are paradoxically associated with higher incremental emissions from data centers. High gas prices push more of the new load toward renewables, moderating emissions even as overall energy costs climb.
Data centers and cryptocurrency operations aren’t going away, and the AI boom shows no signs of slowing. Simply letting that growth happen without deliberate planning carries real consequences: upward pressure on electricity costs for American consumers, a potential reversal of years of progress on power-sector emissions, and regions like Northern Virginia watching a coal resurgence that few residents asked for. The study’s authors frame their results as a call for “proactive planning and targeted policy,” and the numbers make that case hard to ignore.
Disclaimer: This article is based on a peer-reviewed modeling study and reflects projected scenarios through 2030, not confirmed outcomes. The findings depend on assumptions about data center growth rates, energy policy, fuel prices, and regional grid conditions, all of which remain subject to change.
Paper Notes
Limitations
This study focuses on the U.S. power grid through 2030, a deliberately near-term window chosen to limit uncertainty, which means longer-term technology and policy shifts are not captured. The model assumes new data center demand follows historical geographic patterns tied to existing data transmission infrastructure, and future siting decisions could diverge significantly, potentially changing both cost and emissions outcomes. Emissions are attributed to where electricity is generated rather than where digital services are consumed, a production-based accounting choice the authors acknowledge could be viewed differently under a consumption-based framework. Sustained high utilization of coal plants, while modeled, may also be constrained in practice by plant availability and environmental regulations not fully reflected in the scenarios.
Funding and Disclosures
Financial support for the Open Energy Outlook initiative was provided by the Alfred P. Sloan Foundation. No other funding sources or conflicts of interest were identified. The authors note that, in an irony not lost on them, they used ChatGPT (OpenAI, GPT-5-mini) to assist in editing portions of the manuscript; all interpretations, analyses, and conclusions are solely those of the authors.
Publication Details
Authors: Jeremiah X. Johnson (North Carolina State University), Cameron Wade (Sutubra Research), Michael Blackhurst (University of Pittsburgh), Joseph F. DeCarolis (Carnegie Mellon University), Anderson R. de Queiroz (North Carolina State University), I. Daniel Posen (University of Toronto), Paulina Jaramillo (Carnegie Mellon University) | Journal: Environmental Research Letters (IOP Publishing) | Paper Title: Power system costs and emissions from data center and cryptocurrency mining expansion in the United States | Year: 2026 | DOI: 10.1088/1748-9326/ae6c3d | Data Availability: https://doi.org/10.5281/zenodo.17349998







