Flexibility on the grid doesn’t work like flexibility in yoga. When stretching into complex poses, what matters is practice and persistence, breath and balance. When it comes to data center flexibility, though, it’s all about configuration and contracts.
Energy Secretary Wright’s proposal for direct interconnection of large loads to the interstate transmission system gives a big nod to data center flexibility, recommending expedited access for facilities that agree to be dispatchable or curtailable during times of grid stress. But what does that actually look like?
The concept of data center flexibility has been around for a while, but really seemed to burst upon the scene in July 2024, when the Secretary of Energy’s Advisory Board recommended looking into ways to optimize data centers’ consumption and “explor[ing] opportunities for temporal and spatial flexibility in AI training and inference.”
Then, in February, Duke University researchers Tyler Norris, Tim Profeta, and others released a seminal white paper on the issue, defining data center flexibility as “the ability of end-use customers to temporarily reduce their electricity consumption from the grid during periods of system stress by using on-site generators, shifting workload to other facilities, or reducing operations.”
The paper suggested the “headroom” available for data centers to reduce their energy consumption during system peaks — those critical (and expensive) hours when the grid calls upon every spare electron — could “unlock” up to 100 gigawatts of spare capacity to integrate these large loads.
But early examples of data center “flexibility” look impossibly simplistic in today’s environment. A 2016 cloud data center that wanted to run on clean energy but needed 24/7 reliability would simply procure a wind PPA and build a natural gas backup generator. Today, co-location funded by the hyperscaler is increasingly common and may become the new standard. As recently as the time of the Duke white paper, co-location represented “only a portion of load interconnections,” and wasn’t considered “a long-term system-wide solution.” It is now.
If a data center’s power needs are entirely accommodated by its own on-site generation, that should make the facility effectively “grid neutral” — something four governors in PJM recently endorsed. So when we talk about facilities with load flexibility potential, we’re really talking about those that still need to draw from the grid, even if they’re “hybrid” (sited alongside a power plant).
But the load profile matters here. Data center load is different from other industrial loads that can simply slow or halt energy-intensive operations if the financial incentive is great enough. AI workload is notoriously “jumpy,” — which casts doubt on the Duke researchers’ assumption that the “new load is constant,” and therefore easily curtailed.
The future of curtailment, then, is unlikely to look like switching over to your co-located backup power source (because in that case why not just run on your own juice all the time?) or ramping down your operations (because you’re not really in control of them).
But it might look like building a large battery as part of your data center configuration, which has already been touted as a way to shorten the interconnection process in certain jurisdictions. Next year, the 96MW Aurora facility, a landmark demonstration flexible data center from partners Nvidia, Emerald AI, EPRI, and others, will come online, showing what a “polite” data center could look like.
But to really become “grid-positive,” one other option is to contract for capacity through a “bring-your-own-VPP” construct, like those proposed by Voltus and WattCarbon. In that scenario, the demand response function is performed by distributed resources elsewhere on the distribution grid — perhaps through smart thermostat curtailment, activation of home batteries, or vehicle-to-grid deployment.
This flexibility option may provide hyperscalers with a quick-to-market contractual option, provided FERC recognizes that not all flexibility needs to be within the fence-line of the data center facility. It also has the benefit of scaling sustainable infrastructure that puts equity and resilience in the hands of ratepayers.
And that, dear reader, could be the hyperscalers’ moment of zen.
Elizabeth K. Whitney is a managing principal at Meguire Whitney, a government relations firm representing not-for-profit utilities. She has twenty years of federal policy experience, focusing on energy policy, and writes the This Week in Hyperscaling newsletter. Her government relations practice centers on energy markets, environmental regulation, climate change, and nuclear power. The opinions represented in this contributed article are solely those of the author, and do not reflect the views of Latitude Media or any of its staff.
