The rapid expansion of data centres across the United States, driven largely by the growing demand for artificial intelligence, has raised concerns about electricity consumption, grid reliability, and environmental impacts. A new study from the Massachusetts Institute of Technology (MIT) suggests that these effects will depend heavily on how and when data centres consume electricity. Rather than simply increasing pressure on the grid, facilities that shift a substantial share of their power use away from peak-demand periods could reduce overall electricity costs while helping utilities manage demand more efficiently. However, the environmental outcomes vary by region, with some areas benefiting from greater renewable energy use while others could see increased reliance on fossil fuels. “The key with data centers is: How can we add them to the network without adding a lot to our peak usage?” says Christopher Knittel, an economist at MIT Sloan School of Management and co-author of the study.

Published in iScience, the study, titled Flexible Data Centers Reduce Power System Costs But Can Increase Emissions, was conducted by Juan Ramon L. Senga, Shen Wang, and Knittel from MIT’s Center for Energy and Environmental Policy Research. The researchers used the Gen X model of the U.S. electricity grid to simulate a full year of energy demand under scenarios involving rapid data centre expansion. They focused on Texas, the Mid-Atlantic region, and the Western Interconnect, which together are expected to host about 82 percent of U.S. data centres by 2030. Their analysis compared traditional, inflexible electricity use with scenarios in which data centres shifted at least 20 percent—and in some cases as much as half—of their electricity demand to off-peak hours.

The modelling found that flexible electricity consumption could generate meaningful savings for the power system. Compared with inflexible demand, average electricity costs could fall by as much as 5 percent in Texas, 4 percent in the Mid-Atlantic region, and 2 percent across the western United States. Much of the savings comes from reducing demand during the most expensive periods of the day while making better use of existing infrastructure. Since a large share of grid costs is fixed, spreading those costs across greater electricity consumption can lower average prices if peak demand does not increase proportionally. Most data centres already operate below full capacity, allowing certain computing tasks to be shifted from morning and evening peaks to midday hours when electricity demand is lower and solar power generation is strongest.

The researchers also found that different types of AI workloads offer varying degrees of flexibility. Data centres devoted to training AI models typically run continuously and can more easily reschedule computing tasks without affecting performance. By contrast, facilities handling AI inference, such as responding to online search queries or other real-time user requests, must match consumer demand and therefore have less flexibility. Even so, Knittel notes that modest improvements in load shifting could have substantial financial benefits. Savings of just a few percentage points may appear small, but when applied to a power system worth hundreds of billions of dollars annually, they represent significant economic gains.

The environmental picture proved more complex. Without changes in operating patterns, projected data centre growth by 2030 would substantially increase carbon dioxide emissions, with the study estimating increases of 58 percent in Texas, 20 percent in the Mid-Atlantic region, and 24 percent in the western United States compared with scenarios without additional data centres. However, the impact of flexible electricity use depended on regional energy resources. In Texas, where wind power supplies more than half of grid electricity, flexible demand could encourage greater use of renewable energy and reduce emissions by as much as 40 percent. In the Mid-Atlantic region, however, shifting electricity use to periods when solar and wind generation declines could keep coal-fired power plants operating longer, leading to a modest increase in overall emissions despite additional renewable energy generation.

The findings suggest that the benefits of expanding data centres will depend not only on technological advances but also on public policy. Knittel argues that companies may be reluctant to voluntarily adopt flexible operating schedules unless competitors face the same requirements. One possible solution would be to offer faster grid connections to data centres that agree to reduce or shift electricity consumption during periods of peak demand, an approach known as “connect and manage.” Such policies could improve grid stability while encouraging investment in cleaner energy sources where they are most effective. As AI continues to fuel demand for new computing infrastructure, the researchers conclude that carefully designed policies will play a crucial role in determining whether future data centres become a burden on the grid or part of a more efficient and sustainable energy system.

More information: Juan Ramon L. Senga et al, Flexible data centers reduce power system costs but can increase emissions, iScience. DOI: 10.1016/j.isci.2026.116497

Journal information: iScience Provided by Massachusetts Institute of Technology