Data center energy use could become cheaper and more efficient if AI facilities shift electricity consumption to off-peak hours, according to a new MIT study that highlights both economic and environmental trade-offs.
As artificial intelligence fuels a rapid expansion of data centres around the world, concerns are growing over how much electricity these facilities will consume—and whether power grids can keep up.
A new study by researchers at the Massachusetts Institute of Technology (MIT) suggests there may be a way to ease the pressure. Rather than consuming electricity around the clock at fixed rates, data centres could shift a significant portion of their energy use to off-peak hours, lowering electricity costs while making better use of existing grid capacity.
The findings, published in the journal iScience, indicate that if data centres adopt more flexible electricity consumption patterns, average power system costs could fall by as much as 5 per cent in Texas, 4 per cent in the Mid-Atlantic region and 2 per cent across western U.S. states.
Data Center Energy Use: Flexible Data Centers Could Reduce Energy Costs
The researchers modelled how expanding data centres would affect electricity grids in three regions that are expected to host about 82 per cent of U.S. data centres by 2030: Texas, the Mid-Atlantic and the Western Interconnect, which covers 11 western states.
Their simulations found that shifting at least one-fifth of a data centre’s electricity use away from peak-demand periods could reduce overall system costs. In some cases, as much as half of a facility’s energy demand would need to be moved to quieter periods of the day.
“The key with data centers is: How can we add them to the network without adding a lot to our peak usage?” said Christopher Knittel, economist at the MIT Sloan School of Management and co-author of the study, in a media statement.
“One way for data centers to do that — to add to average usage but not the peak usage — is if they provide some grid flexibility during those high-cost periods. And that’s what we’ve been interested in understanding.”
The researchers note that most data centres already have some operational flexibility because they typically run below full capacity. Instead of carrying out energy-intensive computing tasks during periods of peak electricity demand, many could shift those operations to midday, when solar power generation is often highest and overall demand is lower.
AI Growth Is Putting Pressure on Power Grids
The rapid expansion of AI has dramatically increased demand for computing infrastructure, raising questions about whether electricity grids can support hundreds of new data centres without driving up costs or emissions.
The study suggests that adding more data centres does not automatically translate into higher electricity prices. Because much of the cost of running a power grid comes from fixed infrastructure such as transmission lines, increasing electricity use can spread those costs across a larger customer base—provided peak demand does not rise at the same pace.
“It’s really just math,” Knittel said.
“There are two dimensions that data centers have to make decisions about. One is how much of their load in any one time period is flexible. And two, how many hours, plus or minus, can they move that computation?”
Flexible Data Centers May Have Different Climate Impacts
The environmental picture is more complex.
The researchers found that the projected growth in data centres by 2030 could significantly increase carbon dioxide emissions if electricity demand is met through fossil fuels. Compared with a scenario without new data centres, emissions could rise by 58 per cent in Texas, 20 per cent in the Mid-Atlantic region and 24 per cent in the western United States.
However, the impact varies depending on how regional electricity systems generate power.
In Texas, where wind energy accounts for a large share of electricity generation, shifting data-centre operations to times when renewable energy is abundant could reduce carbon emissions by as much as 40 per cent.
In contrast, the Mid-Atlantic region presents a different picture. There, flexible electricity use could unintentionally keep coal-fired power plants operating for longer periods.
“When data centers provide some flexibility in that latter scenario, the data centers actually move hours to when sun and wind energy production is slowing, and that allows a coal plant to stay on,” Knittel observed. “So it doesn’t necessarily attract more renewable investment. It attracts more coal investment.”
Policy Could Shape the Future of AI Infrastructure
The researchers argue that flexibility alone is unlikely to become common unless governments and grid operators create incentives for companies.
“That’s why we have policy,” Knittel said.
One option would be to allow data centres that agree to flexible electricity use to connect to the grid sooner.
“One big concern about these data centers now is how long it takes for them to connect to the grid,” Knittel said. “One way to provide flexibility now is what’s called ‘connect and manage,’ which is, connecting you faster to the grid if you agree to provide flexibility. Tech firms would take that deal. They would rather connect a year earlier, and throttle down computation a few hours a day, than to have to wait. We do this with power plants too.”
He added that industry-wide rules would help address competitive concerns.
“Tech companies say they won’t provide flexibility alone. But if everyone in the industry has to, it’s okay.”
Balancing AI Growth With Sustainable Energy
As governments and technology companies race to build the computing infrastructure needed for the AI era, the study suggests that when data centres consume electricity may prove to be as important as how much they consume.
The researchers conclude that smarter scheduling of electricity demand, combined with supportive public policy, could lower power system costs while reducing pressure on electricity grids. At the same time, the study highlights that the environmental benefits of flexible energy use will depend on how individual regions generate electricity, reinforcing the need for location-specific energy planning.