Almost exactly a year ago, researchers at Duke University published a now-widely cited study on the potential of flexible data centers, explaining that the U.S. grid may have more than 100 gigawatts of headroom for new load — without the need for major capacity expansion.
It contributed to a buzz of investment, industry speculation, and regulatory scrutiny about large load demand response as a primary grid reliability tool, as well as skepticism about whether flexibility might actually result in increased overall emissions from the very tech companies that have committed to aggressive sustainability goals. Its lead author, then-PhD student Tyler Norris, has since stepped into an energy policy role at Google.
That original paper, however, was limited in scope. Today, new research published by Norris’ former colleagues at the Nicholas Institute for Energy, Environment, and Sustainability, builds on last year’s finding, forecasting how specifically flexible data centers could impact other grid stakeholders, and the grid itself, in the next decade.
In the long term, the study found, flexibility tends to favor renewables development over gas, and results in up to $150 billion in cumulative savings for customers. Without flexibility, new natural gas combined cycle units are likely to be the most significant source of new utility-provided electricity for data centers over the next five years, it found.
Authors Martin Ross and Jackson Ewing assessed several distinct scenarios, defined by temporal and spatial adjustments to data center flexibility. Temporal flexibility — i.e. shifting when a data center performs a particular task — is simpler to model than regional flexibility, in which complex external factors such as siting, access to networking resources, or labor availability come into play, Ross explained.
Even the lowest levels of flexibility — a scenario assuming data centers fully disconnect from the grid during the costliest 1% or 2% of peak hours a year — could have a significant impact on the grid, the authors found. That scenario would reduce the amount of new natural gas construction by up to 15%, or roughly 12 GW.
In a scenario with slightly more flexibility, assuming 20% of a data center’s total demand is flexible across the year while 80% is fixed to standard commercial load patterns, new natural gas construction would be cut by nearly 20% and data centers would see average electricity price decreases of up to $3 per megawatt-hour. The study also assesses a more aggressive scenario, in which 50% of total demand is flexible across the year. In that case, nearly half of all new gas capacity required in the non-flexibility scenario would be avoided.
Reducing demand during peak hours increases the relative value of renewables, Ross explained: “Relatively small changes [in demand] are enough to noticeably change what the capacity mix and investment needs might look like over the next five to 10 years.”
That’s not to say that concerns over flexibility efforts leading to increased emissions in the near term, thanks to an increased reliance on existing baseload fossil plants, aren’t valid. “It’s entirely possible…that flexibility can potentially shift things around in a way between gas and coal and renewables that would have a short-term increase in emissions,” Ross said. “In the longer term, I think flexibility is fairly unequivocally going to be good for lowering emissions, because you do get this shift in the generation mix away from gas and into renewables.”
Speed to power policy
The theoretical potential of flexibility is clear in both Duke papers, both in 2026 and 2035. But significant real world hurdles remain regarding reliability, resource adequacy, and cost allocation.
Policymakers, grid operators, and utilities around the country are increasingly considering a speed to power bargain, Ewing explained: If hyperscale customers want to connect quickly in constrained regions, they must then accept some form of operational flexibility in return. This “quid pro quo” is already showing up in early-stage proposals and utility tariffs, he added. And it’s playing out both at the Federal Energy Regulatory Commission, where rulemaking is underway to standardize interconnection for large loads (and potentially prioritize those offering to be flexible) and at the market level in PJM, where the board recently approved a framework that lets data centers either bring their own new generation, or accept early curtailment in exchange for faster interconnection.
This second Duke study gives those conversations a more concrete analytical backbone, Ewing said. Rather than treating flexibility as a vague aspiration, the study puts numbers on different degrees of flexibility, and what each could mean for system-wide investments, costs, and generation over the next decade.
Still, the complicated realities of making flexibility work on the ground remain. The study doesn’t attempt to prescribe any specific tariff designs or commercial structures, and the modeling doesn’t account for some of the trickiest questions about flexing AI workloads, the strict availability requirements from cloud customers, and enforcement mechanisms for disconnecting from the grid.
“We recognize that flexibility is not some panacea,” Ewing said. But given how many policy efforts are unfolding around flexibility at the moment, he added, “it’s worth looking into what some of these implications might be on cost and generation mix in particular.”
Transition-AI 2026, Latitude Media’s flagship event at the center of the AI-energy infrastructure buildout, will convene developers, utilities, regulators, hyperscalers, and capital providers to move towards coordinated, sustainable solutions for AI-driven demand. Register now to secure the lowest rates and join us in San Francisco April 13-14.
