The integrated resource plan is the single most consequential document a regulated utility files. It is the long-range demand forecast, the resource stack that meets it, and the cost test the state public utility commission uses to approve or reject capital investment over the next twenty years. Every coal retirement, every gas turbine order, every solar PPA, every nuclear license renewal flows from an IRP assumption set. For most of the last decade the demand forecast in those documents looked roughly the same year over year. Load was approximately flat, with efficiency gains roughly offsetting electrification, and the resource discussion was almost entirely about how to replace coal at the lowest delivered cost.
That has now broken. The 2025 IRP filings, and the 2026 updates already in front of regulators in Virginia, North Carolina, South Carolina, Georgia, Ohio, Indiana, and Texas, contain load forecasts that are not on the same curve as the 2022 and 2023 filings from the same utilities. Peak demand and energy growth have both stepped up, and the size of the revision is not subtle.
The size of the load revision
The clearest data point is Dominion Energy Virginia. The 2023 IRP filed with the Virginia State Corporation Commission projected summer peak load growth in PJM Dominion zone of roughly 1.5 to 1.7 percent per year. The 2024 update raised that to roughly 5 percent per year. The 2025 IRP, filed in late 2025 and updated in early 2026, projects compound annual growth of roughly 5.5 to 7 percent for the next decade depending on the scenario. The 2025 plan adds approximately 13 gigawatts of incremental peak demand by 2035 relative to the 2023 plan, on a starting base of roughly 21 gigawatts.
Duke Energy Carolinas and Duke Energy Progress filed a joint Carolinas Resource Plan update in 2025 that raised the combined 2030 peak from approximately 32 gigawatts in the 2023 plan to approximately 39 gigawatts in the 2025 update, with most of the revision concentrated in 2027 through 2031. Georgia Power’s 2025 IRP triennial update raised the company’s projected winter peak by roughly 8 gigawatts through 2031, again concentrated in the back half of the period.
American Electric Power’s Indiana Michigan Power and AEP Ohio IRP updates raised data-center-driven load in central Ohio and northern Indiana by roughly 6 to 9 gigawatts depending on scenario. Entergy filed revised load forecasts in Louisiana and Mississippi reflecting hyperscaler campuses that did not exist in the 2022 IRP. Even the more conservative Midwestern utilities are showing 2 to 3 percent forecast revisions where the prior cycle had near-zero.
These numbers are not industry averages, they are headline filings, and they are publicly available through each state’s docket portal. The size of the revision is, in aggregate, larger than anything US utilities have planned through since the post-war electrification expansion.
Where the load is coming from
The load revision is not generic economic growth. Residential demand growth in most service territories is still flat to slightly negative on a per-customer basis. Commercial space conditioning is up modestly. Manufacturing is up modestly where reshoring is real, primarily in semiconductor, battery, and EV plant clusters. Transport electrification is contributing, but slower than the 2022 forecasts assumed because EV sales growth slowed in 2024 and 2025.
The dominant contributor is data center load. The pattern is consistent across filings: hyperscaler and colocation campuses requesting interconnection at the 100 to 500 megawatt scale per site, with announced or contracted facilities clustered in a small number of territories. Northern Virginia continues to lead, but the new growth is more geographically distributed than the prior cycle. Central Ohio, the Atlanta suburbs, the Phoenix area, the Dallas-Fort Worth corridor, Louisiana, and the Carolinas are all carrying multi-gigawatt new data center pipelines in their 2025 IRPs.
The training side of the AI compute buildout has been the proximate driver of the campus footprint. Inference deployment, which carries different load shape characteristics and may end up more geographically diffuse, has not yet shown up in the IRP numbers in a material way. That is a forward risk on the demand forecast, in both directions.
How the resource stack is responding
If utilities were simply replacing retiring coal one for one with the cheapest available marginal capacity in 2026, that would be solar plus storage in most of the country, gas combined cycle in the rest, and a small contribution from new nuclear and demand response. With load growing at this rate, the planning calculation changes.
The 2025 IRPs are doing four things, roughly in order of contribution.
First, they are deferring announced coal retirements. The Dominion 2025 plan defers the retirement of the Chesterfield coal units and revises the Clover and Mount Storm timelines. Duke’s Carolinas update defers Marshall and Belews Creek coal retirements by two to five years. AEP’s Ohio plan signals review of the Cardinal and Gen J M Gavin schedules. The pattern is consistent: announced retirements through 2028 are largely intact, but units scheduled for 2029 through 2032 are being pushed three to five years later.
Second, they are adding gas. New combined cycle units in the 700 to 1100 megawatt range are appearing in five-year capital plans that did not include them in the 2023 cycle. Gas peakers in the 50 to 250 megawatt range are filling the prompt reliability gap. The constraint on this path is the gas turbine orderbook. GE Vernova and Siemens Energy heavy-duty turbines are on multi-year backlog through 2029, and aeroderivative units from GE and Mitsubishi Power are tight through 2027.
Third, they are expanding nuclear at the edges. License renewals, uprates, and the early life extension of plants previously scheduled for closure are all in the 2025 plans. New small modular reactor commitments are showing up as planning assumptions in the back half of the IRP windows, primarily 2032 and later, with the caveat that no SMR has yet completed first-of-a-kind commercial deployment in the United States.
Fourth, they are adding solar and storage, but at roughly the pace of the prior IRP cycles, not faster. The 2025 plans are not solving the load gap with renewables. The IRA tax credit transferability rules and the Section 45X manufacturing credit have stabilized the project economics, but interconnection queues, transmission capacity, and grain-oriented electrical steel for transformers are all binding the rate at which new solar and storage can be added regardless of the policy signal.
The constraints on what the IRPs are actually planning
A resource plan is only worth the steel that gets built. The 2025 plans contain physical project lists that face three external constraints that the utility cannot directly relieve.
The interconnection queue. PJM, MISO, ERCOT, CAISO, and the Southeast pseudo-pools all currently sit on multi-hundred-gigawatt queues. FERC Order 2023 has begun to clear out speculative projects, but the cluster study process for the projects that survive remains slow. Battery storage interconnection wait times are converging on transmission wait times, which means that the resource the IRP wants to add quickly is the resource the queue cannot deliver quickly.
Transformer supply. Large power transformers in the 138 to 345 kV class are quoting 24 to 36 month lead times, and the binding input is grain-oriented electrical steel and skilled assembly labor at a small number of OEM plants. The same transformer order is needed for a gas plant, a solar farm, a battery site, and a data center substation. The IRP can request the resource. The transformer order has to be placed two years before mechanical completion.
The gas turbine orderbook. Heavy-duty F-class and H-class units from GE Vernova and Siemens Energy are on backlog through the end of the decade. Mitsubishi Power’s aeroderivative LMS100 units are tight through 2027. Several utilities have placed reservation deposits for delivery slots in 2028 through 2030 that do not yet have full equipment specifications. That is unusual, and it indicates that the demand has run ahead of the OEM production response.
The three constraints stack. A new gas combined cycle plant needs the turbine, the transformer, and the interconnection. A coal deferral needs a relicensed environmental compliance position. A new solar farm needs the transformer and the queue position. Every line item in the 2025 IRPs is competing for the same finite OEM and queue capacity.
What the regulators are signaling
State public utility commissions are not approving the 2025 IRPs as filed. The Virginia SCC, the North Carolina Utilities Commission, the Georgia Public Service Commission, and the Public Utilities Commission of Ohio have all signaled discomfort with the magnitude of the rate impact embedded in the 2025 plans. The point of friction is not the load forecast itself, which the commissions broadly accept based on the executed interconnection requests from hyperscaler customers. The point of friction is the cost allocation. Who pays for the gas turbine that a data center campus needs, the data center customer through a special tariff or the residential ratepayer through general rate base?
Virginia has moved on a data center tariff structure with separate cost allocation. North Carolina is in early proceedings on the same question. Ohio is considering a separate generation cost band for large new loads. The legal and regulatory architecture for who carries the marginal cost of data-center-driven generation is still being built, and the IRP that lands on a state commission’s docket in 2026 looks different from the same IRP filed in 2025 because of these proceedings.
What this means for the energy transition narrative
The headline reading of the 2025 IRPs is that coal is sticking around longer than the climate scenarios assumed, gas is being built faster than the climate scenarios assumed, and renewables are growing at roughly the rate the climate scenarios assumed. The net emissions trajectory of the US power sector through 2030 is therefore higher than it looked in the 2023 IRPs.
The countervailing reading is that the load was going to happen anyway. If a hyperscaler campus does not interconnect to the regulated utility, it interconnects to a behind-the-meter gas plant that is invisible to the IRP, or it locates in a different country. The IRP at least keeps the new load inside a planning framework where solar, storage, nuclear, and gas are competing on a transparent cost basis. That is not the same outcome as net-zero on schedule, but it is meaningfully different from the unregulated alternative.
Either way, the 2025 IRPs are now the document to read for anyone trying to understand what physical infrastructure will actually be built between now and 2032. The financial models, the policy debates, and the news coverage are catching up to a set of facts that the utilities already filed in their resource plans. Read the IRPs.