The first piece in this series set out the constraint: heavy-duty gas turbines from the three OEMs are quoting 48 to 60 month delivery windows, and the IRPs counting on 2027 to 2029 combined-cycle capacity have to find that capacity somewhere else. Storage was lever 1. Coal life-extension was lever 2. Lever 3 is demand-side reliability, and the version showing up in capacity auctions and IRP filings is not the one that older DR vocabulary describes.
The legacy mental model is residential. A utility enrolls a few hundred thousand customers in a thermostat or pool-pump cycling program, calls events on a handful of summer afternoons, and books the megawatts against its reserve margin. That construct still exists. It is not where the growth is. The growth is in large-customer interruptibility, and the largest customers signing the contracts are the same hyperscaler data centers driving the load-growth case that is forcing the gas turbine procurement problem in the first place.
What the auctions are showing
PJM’s 2026 base residual capacity auction cleared roughly 10.3 GW of demand resources for the 2027 to 2028 delivery year, up from 7.8 GW in the prior cycle and about 5.9 GW two cycles earlier. The mix has shifted inside that total. Residential and small-commercial DR aggregators cleared roughly the same megawatt volume they did three years ago. The growth came from large-customer interruptible load offered directly by industrial and commercial end-users or by curtailment service providers contracting with those customers, with data-center load accounting for a substantial and growing share.
MISO’s planning resource auction shows a similar pattern. The 2025 to 2026 auction cleared about 4.6 GW of demand resources across all zones, with the zonal split heavily weighted toward the higher load-growth zones in the western and central footprint where the data-center buildouts are concentrated. Year over year growth in cleared demand response was disproportionately in those zones.
ERCOT does not run a capacity auction in the PJM sense, but its emergency response service and load resource participation in the energy and ancillary services markets show the same shape. Controllable load resources in the ERCOT market, including the large bitcoin mining loads and a smaller but growing data-center cohort, are providing responsive reserve service at volumes that did not exist before 2022. The most recent ERCOT reserve adequacy reports cite responsive load as a material contributor to the resource-adequacy picture for 2027 and beyond.
CAISO’s flex alert and demand response auction mechanism, smaller in megawatt scale than the other ISOs, is also growing. The growth there is split between behind-the-meter solar plus storage aggregations and large-customer industrial interruptibility, with less of the pure data-center component because the California data-center footprint, while not small, is not on the scale of Virginia, Ohio, Texas, or the Phoenix metro.
What the data-center contracts actually obligate
The terms inside a hyperscaler interruptibility contract are not standardized, but a working description of what these agreements are doing has become possible from utility integrated resource plans, FERC-filed market participation rules, and the public comments hyperscalers have made about their grid posture.
The contract typically commits the data-center operator to curtail a defined block of load, often in the range of 30 to 150 MW per site, on call from the utility or ISO during defined system stress conditions. The call mechanism is usually an ISO emergency declaration, a specific reserve margin trigger, or a bilateral notification process the utility runs internally. The customer has anywhere from 10 minutes to 4 hours notice depending on the program.
The curtailment itself, when called, is executed by the data center either through workload shifting, where compute jobs are migrated to other regions, or through on-site generation, where backup diesel or gas generators pick up the load that the grid is shedding. The workload-shifting path is the cleaner story. The on-site generation path, particularly diesel, is the part that drew regulatory attention in the AEP Ohio diesel disclosure docket last year and that the Northern Virginia rolling discussion has now formalized into proposed program rules.
The compensation to the customer is structured either as a capacity payment in the capacity market, a bilateral demand-response payment, or a discount on the energy supply rate. The economics for the hyperscaler depend heavily on how often the curtailment is called. A program that calls 5 to 15 hours per year, which is roughly what historical demand-response performance looks like, is straightforwardly profitable for a customer that already operates a multi-region workload scheduling stack and has on-site backup generation it pays to maintain regardless.
The accreditation that the capacity market gives the resource is where the policy and engineering questions concentrate. PJM’s capacity performance construct requires the resource to actually deliver when called, with non-performance penalties that have been progressively sharpened since the 2018 to 2022 capacity performance redesign. MISO’s accredited capacity for demand response uses an effective load-carrying capability framework analogous to what storage gets, with the performance record of the resource feeding back into the accreditation. ERCOT’s performance credit mechanism debate, still unresolved, would change how demand resources are paid for capacity-like service.
How well these resources actually perform
The historical record for utility demand response performance is better than the popular narrative suggests, but uneven across program types and geographies.
PJM’s annual emergency response performance reports, published as part of the capacity performance accountability framework, show large-customer interruptible programs performing at 85 to 95 percent of their committed obligation across the events called between 2019 and 2025. Residential and small-commercial DR aggregators performed at 70 to 90 percent across the same period, with wider variance. The events where performance dropped below 70 percent were almost all extreme-cold or extreme-heat conditions where the customer end-use load was itself unusually shaped, which is exactly when the capacity is needed.
The performance question for the new data-center interruptibility cohort does not have a multi-year track record yet. Most of these contracts were signed in 2024, 2025, or 2026 and have either not yet been called or have been called only once or twice. The utility filings that cover them generally treat the resource as analogous to other large-customer interruptible load, which is the highest-performing DR category historically. Whether the data-center analog actually performs at that level when called for a 4-hour summer peak event is a hypothesis being tested in real time over the next several capacity-market delivery years.
The downside cases worth attention are not the headline ones. Data centers signing curtailment contracts to clear an interconnection queue that they cannot otherwise clear, but with no actual operational plan to curtail at scale, is the failure mode reliability planners are most concerned about. The interconnection-queue use case has emerged because data-center developers can in some jurisdictions get accelerated grid access by agreeing to be curtailable. The agreement on paper, without a real workload-shifting capability or sufficient on-site backup, would only show up as a problem when an event is called.
The PJM emergency response performance report for the 2027 to 2028 delivery year, which will be the first one that includes the post-2025 large-customer DR growth at material scale, will be the empirical read on whether the lever holds. The first delivery year covering it produces an actual event record only after the fact, which is one of the structural reasons capacity-market reform discussions have been intensifying.
Where the regulatory weight is moving
Several state and federal proceedings active in 2026 will shape how aggressively utilities can lean on the lever in IRP planning.
FERC’s ongoing capacity market reform dockets, including the PJM capacity performance review and the MISO accreditation update, have a demand-response sub-thread that is largely under the radar in the trade press but is operationally significant. The question of how to verify performance, how to penalize non-performance, and how to keep the accreditation honest as the resource composition shifts is being worked out in technical conferences and stakeholder filings throughout the year.
State-level interruptibility tariff proceedings are running in Virginia, Ohio, Texas, Georgia, and Arizona, where the data-center load concentration is highest. The Virginia State Corporation Commission’s docket on the Dominion Energy data-center contract framework has the most consequential pending order, with a decision expected before the end of the third quarter. The order will likely set a regional precedent on how state regulators want hyperscaler interruptibility obligations structured, audited, and enforced.
The EPA permitting question on backup generation, particularly diesel emergency engines used during DR events, is the regulatory dimension utilities and hyperscalers are most quietly worried about. The Northern Virginia regional air quality district has signaled a tighter interpretation of when a diesel engine becomes a permitted source rather than an emergency-only unit, which could complicate the on-site generation path that some DR contracts depend on. Workload-shifting based DR, by contrast, faces no equivalent regulatory headwind.
The cost question
The capacity-market clearing price for demand resources has historically been lower than for generation, on a per-MW-year basis, though the gap has narrowed in recent auctions. PJM’s 2026 auction cleared DR at prices roughly comparable to generation in several zones, which reflects both the scarcity premium across the auction and the improved accreditation treatment of DR.
For an integrated utility, the economics of contracting demand-side reliability instead of building new generation can look favorable on a near-term basis. The utility is not adding to the rate base, the contract is treated as a purchased capacity expense rather than a capital outlay, and the customer relationship is improved. The longer-term question is whether the cost-effectiveness holds at scale and whether the reliability is genuinely equivalent.
The rate-impact distribution is also less ratepayer-regressive than the coal life-extension lever, where residential customers bear costs the data centers drive. Demand-side reliability contracts compensate the curtailable customer directly, which is closer to the user-pays principle that energy economists generally argue for. Whether that distinction matters politically in any given state PUC docket is a separate question.
What to watch through the rest of 2026
The PJM 2027 to 2028 delivery year, beginning in June 2027, will be the first reliability-critical period where the post-2024 large-customer DR growth is fully operational. The summer 2027 system stress events, whatever shape they take, will produce the first real performance read on the new resource composition.
The Virginia SCC’s expected third-quarter order on the Dominion data-center contract framework will shape the regulatory template for how interruptibility obligations get structured in the highest data-center growth jurisdiction in the country.
MISO’s planning resource auction for 2027 to 2028, results expected in late 2026, will indicate whether the demand-resource growth pattern continues to track load-growth concentration zones or whether the supply has been substantially tapped.
The FERC capacity performance reform docket, which has had several rounds of stakeholder filings already in 2026, may produce an order before year-end that adjusts how DR resources are accredited and how non-performance is penalized. An order that tightens the accreditation framework would reduce the apparent megawatts demand resources contribute to the capacity-market resource adequacy picture, even without changing the underlying contracts.
Lever 4, behind-the-meter customer deals, is the closing piece in this series and the one where the boundary between utility-grid reliability and customer-self-supply has been most actively renegotiated. That piece runs next.