Even modest measures to curb data centers’ energy use during peak hours could substantially reduce the amount of new generation capacity needed to meet growing U.S. electricity demands over the next decade, a new report on the potential economic benefits of data center load flexibility said.
The report was released on Feb. 17 by Duke University’s Nicholas Institute for Energy, Environment & Sustainability.
“In an era of rapidly expanding electricity demand, all options need to be on the table to avoid unnecessary capital investments,” said Martin Ross, senior research economist at the Nicholas Institute. “Flexible operations at data centers will have to play a central role in lowering costs to all customers.”
The report builds on previous research into load flexibility, including a widely cited Duke study released last year that estimated nearly 100 GW of large new loads could be integrated into the existing grid with minimal impacts.
The new analysis incorporates the economics of electricity capacity planning, dispatch and reserve requirements to look at the effects of data center flexibility over the next 5 to 10 years, a timeframe that fits with utilities’ investment horizons. Several modeling scenarios cover a broad range of assumptions about both data center demand growth and the levels and types of flexibility.
Key takeaways from the report include:
The ability to shift data center workloads over hours of the year and across regions offers the potential to avoid billions of dollars in capital investments. Estimated savings for the electricity system in terms of avoided capital, operating and fuel costs for most of the scenarios modeled range from $40 billion to $150 billion over the next decade.
Data center flexibility could shift energy investments from natural gas toward renewables. If flexibility is limited, new natural gas combined cycle (NGCC) power plants are likely to be the most significant source of new utility-provided electricity for data centers over the next five years.
New NGCC capacity investments are cut by 10%-50% in the flexibility scenarios modeled. The lower end of the range assumes demand flexibility that only avoids 1% of peak hours. Combining peak-hour flexibility with broader temporal data center adjustments can extend the benefits by further reducing the need for new gas units. Moving data centers around the country to take advantage of low-cost generation—particularly from renewables—has even greater potential for reductions than temporal flexibility.
Across the modeled scenarios, electricity prices fall for both data centers and retail customers. Data centers could see average electricity prices between $2.5/MWh and $8.0/MWh lower as the result of flexibility. Meanwhile, the modeling projects retail electricity prices to fall by roughly 0.5% to 2.8% since data center flexibility reduces stress on the grid during the highest-cost peak hours and lowers capital expenditures.
Long-term carbon emissions are generally lower in the modeled scenarios. Emissions may rise temporarily in the near term if flexibility affects the dispatch of existing fossil fuel units. However, long-term emissions decrease in the analysis because data center flexibility leads to fewer new gas plants—even without clean electricity incentives or other climate policies.
“Demand response by data centers is critical in lowering stress on the grid today and in the future,” said Jackson Ewing, director of energy and climate policy at the Nicholas Institute. “Potential benefits include cost savings, emissions reductions and avoiding stranded generating assets in the future.”
Ross and Ewing note that future analysis should move beyond the stylized scenarios in the report to more fully reflect the evolving characteristics of data center loads and power system constraints. Future studies should also reassess the extent to which flexibility from other customer classes—such as electric vehicles or industrial or commercial electrification—could deliver comparable system benefits, and under what conditions higher levels of participation might realistically be achieved.
