The rapid expansion of artificial intelligence is forcing utilities to confront a question they have not faced in more than a decade: how to meet a surge in electricity demand.
Across the country, companies are racing to build data centers that train and run AI systems.
Other WRAL Top Stories
Just this week, Microsoft announced plans to move forward with a data center in Person County. Officials said on Tuesday that the data center is expected to grow the tax base and create jobs across the county.
Utilities, in turn, are planning new power plants to supply them. In many regions, that likely means more natural gas.
A new study from Duke University suggests the scale of new construction could depend less on how much electricity data centers use than on when they use it.
The report found that if large data centers shifted a portion of their computing to times when the grid is less stressed, the United States could avoid up to $150 billion in power plant, fuel and transmission costs over the next decade. Much of the deferred investment would otherwise go toward natural gas facilities.
“At any given time there’s plenty of unused power on the grid,” said Jackson Ewing, director of energy and climate policy at Duke's Nicholas Institute. “The core challenge is those peak moments, which only happen a few times a year.”
Utilities build systems to withstand those brief periods of extreme demand, often during heat waves or winter storms. Even if power plants operate at full output for only a handful of hours annually, customers still pay for the infrastructure year-round through regulated rates and fuel charges.
The concept at the center of the Duke study is known as load flexibility. In practical terms, it means allowing data centers to move some computing tasks to different times of day or even different geographic locations when electricity is more abundant.
“Flexibility means that data centers are able to shift their energy needs across time,” Ewing said, “in ways that allow them to not use as much power when other demands on the grid are particularly high.”
The modeling builds on earlier Duke research that identified unused capacity in existing systems. This year’s analysis looks ahead 10 years, asking how flexibility could alter the investment decisions utilities are making now.
The timing matters. Electricity demand had been relatively flat for years. Now it is rising again, driven by AI, advanced manufacturing and electrification. Utilities typically plan generation and transmission investments years in advance. If they overestimate future demand, ratepayers can be left paying for assets that are underused.
“There’s a risk here of overbuilding in anticipation for load that doesn’t ultimately materialize,” Ewing said.
Electricity prices were already climbing before the AI boom. Aging infrastructure, fuel costs and extreme weather have all put upward pressure on bills.
“These data centers are coming like a train on top of those other pressures,” said Allison Clements, a former commissioner at the Federal Energy Regulatory Commission. “They certainly add to the need to find solutions quickly.”
Whether data centers raise or lower costs depends largely on regulatory design. If companies pay their share of infrastructure upgrades and coordinate with utilities, they can spread fixed costs across a larger customer base. If those arrangements are misaligned, Clements said, “there can be upward pressure on rates.”
In North Carolina, Duke Energy says its newest large-load contracts, particularly for data centers, require customers to reduce or shift electricity use during periods of peak demand if requested. Data centers currently account for less than 1% of the company’s peak demand in the Carolinas, though Duke expects them to make up about 10% of total electric sales by 2030.
The utility says large-load customers can lower overall costs by spreading infrastructure expenses across a broader base. In a recently signed 1-gigawatt data center agreement, Duke estimates up to $1 billion in customer savings over 15 years.
At the same time, Duke’s latest resource plan shows that even with 1,000 megawatts of curtailable load, the company would still need new combined-cycle natural gas plants to meet year-round demand.
"If it is built smartly and in an integrated way, we could end up with a more balanced and affordable, resilient energy system," Ewing said.