The grid wakes up: AI demand, FERC 1920, and the PJM, MISO interconnection cliff: Deluair Consultancy

US electricity demand is growing again. The Energy Information Administration's Short Term Energy Outlook (STEO) for 2025 projects retail sales growth of roughly 2 percent year on year, against a 2014 to 2023 average closer to 0.4 percent. The PJM 2025 to 2026 capacity auction cleared at 14.7 billion dollars, up from 2.2 billion the year prior, and MISO Zone 4 cleared at 666.50 dollars per megawatt-day for 2025 to 2026 against 30 dollars per megawatt-day the prior cycle. FERC Order 1920, finalized May 13, 2024, requires every transmission planner to run a 20 year horizon study, and Order 2023, July 2023, replaced first come first served interconnection with cluster studies. The pipeline is large. The throughput is not. This brief works through what AI demand actually requires, where FERC reform helps and where it does not, and what utilities, hyperscalers, independent power producers, and state regulators should do across 2026 to 2028.

Demand is back, and the curve has bent #

From 2014 to 2023, US retail electricity sales grew at a compound rate of roughly 0.4 percent per year, the slowest decade in modern record. EIA's Short Term Energy Outlook released in 2025 marks that line broken: retail sales are projected at roughly 2 percent year on year growth, with the commercial sector leading. EIA's Annual Energy Outlook 2025 reference case carries that momentum forward through the late 2020s.

The driver is concentrated. BloombergNEF's late 2024 forecast adds 38 gigawatts of US data center load between 2024 and 2030. Goldman Sachs Research, in an April 2024 note widely cited by FERC and ISO planners, projected data center electricity demand could rise more than 160 percent through 2030, lifting the data center share of total US power consumption from roughly 3 percent to 8 percent. The Department of Energy's December 2024 Lawrence Berkeley National Lab report projects 176 terawatt-hours of data center consumption in 2023, rising to 325 to 580 terawatt-hours by 2028.

Geography is even more concentrated. Northern Virginia, the Dallas Fort Worth and West Texas corridor, central Ohio, Phoenix, central Washington, and the Atlanta region account for the bulk of announced load. Dominion Energy's 2024 IRP filed with the Virginia SCC shows summer peak load rising from 22.5 gigawatts in 2024 toward more than 40 gigawatts by 2039, with data center load alone above 25 gigawatts. That is the entire 2018 system peak on top of the existing system.

PJM and MISO: capacity auctions price the squeeze #

PJM Interconnection's base residual auction for the 2025 to 2026 delivery year, cleared in July 2024, set a new record. Total capacity payments came to 14.7 billion dollars, against 2.2 billion in the previous auction, a 568 percent jump. Clearing prices hit 269.92 dollars per megawatt-day across most of the footprint and 444.26 dollars in the Dominion zone, both at or near the cap. Maryland, Pennsylvania, New Jersey, Ohio, and Virginia regulators warned residential customers of bill increases of 20 to 30 percent on the capacity component once the higher clearing prices flowed through retail tariffs in mid 2025.

MISO told a similar story with sharper cliffs. The 2024 to 2025 planning resource auction cleared Zone 4 (Indiana) at 30 dollars per megawatt-day for summer. The 2025 to 2026 auction, run April 2025, cleared the same Zone 4 at 666.50 dollars per megawatt-day, the seasonal cap, with similar caps in Zones 1, 3, 5, 6, and 7. The footprint wide capacity shortfall, which MISO had been flagging since 2022, finally hit price.

The signal from both auctions is the same. Reserve margins are eroding faster than new resources can clear interconnection. Coal retirements are running ahead of replacement, the gas peaker pipeline is constrained by turbine lead times above 4 years, and the renewable plus storage queue is moving but slowly. Capacity prices are not the cause. They are the thermometer.

Market	Auction year	Clearing price (USD per MW-day)	Total capacity cost (USD billion)
PJM RTO wide	2024 to 2025	28.92	2.2
PJM RTO wide	2025 to 2026	269.92	14.7
PJM Dominion zone	2025 to 2026	444.26	included in 14.7
MISO Zone 4 summer	2024 to 2025	30.00	below 1
MISO Zone 4 summer	2025 to 2026	666.50	approximately 4.4

Sources: PJM Interconnection 2025 to 2026 BRA results (July 2024), MISO 2025 to 2026 PRA results (April 2025).

FERC Order 1920 and Order 2023: the reform stack #

FERC issued Order 2023 in July 2023 to fix the interconnection queue. The reform replaced first come first served serial studies with first ready, first served cluster studies, imposed firm study deadlines with financial penalties on transmission providers who miss them, and added withdrawal penalties to clear speculative requests. The order directly addressed the queue stagnation Lawrence Berkeley National Lab had documented, with PJM holding more than 200 gigawatts of generation in queue, MISO 173 gigawatts, and ERCOT, which is not FERC jurisdictional but adopted parallel reforms, around 60 gigawatts of conventional plus more in storage and renewables.

FERC Order 1920, issued May 13, 2024, is the deeper reform. It requires regional transmission organizations and non RTO planning regions to run long term, scenario based transmission planning over a 20 year horizon, evaluate at least seven specified benefits when assessing projects, and use a default cost allocation method that recognizes those benefits. Order 1920-A, issued November 2024, gave states a stronger formal voice in cost allocation in response to filings from the Organization of MISO States and others. The Department of Energy's 2023 National Transmission Needs Study, the analytical foundation FERC cited, found the US needs 47,300 to 65,400 miles of new high voltage transmission by 2035 to meet expected load and generation, with regional and interregional transfer capacity requirements rising sharply.

Both orders matter. Neither delivers electrons quickly. Order 2023 cluster study cycles for PJM, MISO, SPP, and CAISO are working through their first end to end runs. Order 1920 long term plans must be filed by mid 2027, and the first projects identified under those plans will not energize before 2031 to 2034. The reform stack fixes the system that builds the next decade. It does not solve 2026 to 2030.

What hyperscalers are actually doing while the queue clears #

Hyperscalers have moved past waiting for the grid. The pattern across 2024 and 2025 is colocation with existing nuclear, behind the meter generation, and direct PPAs that bundle new build renewables with firming capacity. Talen Energy's March 2024 sale of its Cumulus campus to Amazon Web Services, sited next to the 2,494 megawatt Susquehanna nuclear plant in Pennsylvania, contracted up to 960 megawatts of nuclear capacity to AWS. The structure faced FERC challenge from AEP and Exelon, and a November 2024 FERC order rejected the amended interconnection service agreement, sending the parties back to design.

Constellation Energy's September 2024 announcement to restart Three Mile Island Unit 1, rebranded the Crane Clean Energy Center, contracted 100 percent of the 835 megawatt unit to Microsoft for 20 years, with target restart in 2028 pending NRC review. Vistra's Comanche Peak nuclear plant in Texas is in active discussions with hyperscaler offtakers per Vistra's Q4 2024 earnings call. Meta issued a request for proposals in December 2024 seeking 1 to 4 gigawatts of new nuclear generation in service by the early 2030s.

Behind the meter gas is filling the immediate gap. GE Vernova reported aeroderivative gas turbine bookings, primarily LM6000 and LM2500 units sized 25 to 100 megawatts, running well ahead of the prior baseline, with most going to behind the meter data center applications. The attraction is speed: behind the meter installations can be permitted as on site generation, often outside the full ISO interconnection queue, and online in 18 to 30 months versus 4 to 7 years for grid connected combined cycle. The cost is regulatory, environmental, and political, and several states (Virginia, Texas, Ohio) have opened proceedings on whether and how behind the meter load should pay transmission and capacity charges.

Generation mix, the IRA, and the policy reset #

Despite the queue, generation is being added. EIA Form 860 data shows 2024 utility scale capacity additions of roughly 53 gigawatts: 30 gigawatts of solar photovoltaic, 11 gigawatts of battery storage, 5.4 gigawatts of natural gas, and 2.4 gigawatts of wind, with the balance distributed across nuclear uprates and other categories. Distributed solar added another roughly 14 gigawatts. The 2024 build was the largest single year addition in modern US record. The Inflation Reduction Act technology neutral production and investment tax credits, sections 45Y and 48E, took effect for projects starting construction after January 1, 2025, and the One Big Beautiful Bill Act signed July 4, 2025 modified but preserved the credits for storage, geothermal, nuclear, and a phase down for wind and solar.

The Trump administration's National Security Presidential Memorandum 2 (NSPM-2) of February 4, 2025 ordered a halt on new offshore wind permits and review of existing approvals, removing roughly 30 gigawatts of planned offshore capacity from the 2025 to 2035 outlook. CHIPS Act manufacturing incentives were reaffirmed. Several IRA energy provisions are under active review by Treasury, with section 45Y and 48E remaining in force as of April 2026, but with modified guidance on foreign entity of concern provisions and domestic content. Net effect on the 2026 to 2030 build: solar and storage continue, onshore wind slows modestly, gas peakers accelerate, nuclear restart and small modular reactor pre development advance, offshore wind effectively pauses.

This combination tightens the binding constraint. The fastest to deploy new capacity, solar and battery storage, depends on cluster study throughput and transmission upgrades that Order 1920 plans for but does not deliver inside the window. The slowest to deploy capacity, large gas combined cycle and new nuclear, depends on turbine and pressure vessel supply chains with multi year backlogs. Behind the meter gas and nuclear restart are bridges, not the build.

Resource	2024 US capacity additions (GW)	Status under 2025 policy
Utility scale solar	30.0	45Y, 48E credits sustained, FEOC tightened
Battery storage	11.0	45Y, 48E credits sustained
Natural gas	5.4	Turbine lead times above 4 years
Onshore wind	2.4	Credits sustained, phase down post 2027
Offshore wind	0.0 net new	NSPM-2 February 2025 halted permits
Nuclear (restart, uprate)	below 0.5	Three Mile Island restart targeted 2028

Sources: EIA Form 860 and Electric Power Monthly 2024, executive orders and NSPM-2 (Feb 4, 2025).

What to do, by stakeholder, 2026 to 2028 #

For vertically integrated utilities and load serving entities. Treat the integrated resource plan as a real time document. Dominion, AEP, Duke, Entergy, and Xcel are running IRP refresh cycles inside 18 month windows where 36 month was normal. Build flexibility into procurement: layered capacity contracts, demand response options, and explicit colocation tariffs. Use Order 1920 long term planning to lock interregional transfer rights now.

For hyperscalers and large data center operators. Stop assuming greenfield grid connect inside 4 years. Build colocation with existing nuclear and combined cycle as the base case, behind the meter aeroderivative gas as the bridge, and grid connect as the long term. Sign multi decade PPAs with firming, not slice of system renewable contracts that the 2024 capacity auctions revealed as unhedged. Engage state commissions early on cost allocation: silence is not a winning position.

For independent power producers and developers. The 2025 capacity prices are real revenue. Move ready to build projects through cluster studies aggressively, post withdrawal collateral early, and stage projects to Order 2023 milestones. Storage paired with solar continues to be the highest IRR profile in PJM, MISO, and CAISO under current capacity and ancillary prices. SMR pre development should target 2032 to 2035 commercial operation with hyperscaler offtake, not merchant.

For state public utility commissions and FERC. Cost allocation is the binding political question. Order 1920-A gave states a louder voice, and that voice has to choose between keeping residential bills flat and funding the transmission build that load growth requires. The honest answer is some combination: large customer contribution rules that put incremental data center load on the hook for incremental transmission, paired with broader rate base recovery for shared backbone investments.

For policymakers in Washington. The reform stack is correct. Permitting reform on transmission siting, including FERC backstop authority under section 216 of the Federal Power Act, needs to be exercised, not held in reserve. The binding constraint by 2027 will be skilled labor, transformers, and high voltage cable, not financing or policy.

Sources #

Cite this brief

@misc{hossen2026uspowergrid2026,
  author = {Hossen, Md Deluair},
  title  = {The grid wakes up: AI demand, FERC 1920, and the PJM, MISO interconnection cliff},
  year   = {2026},
  url    = {https://deluair.com/consultancy/insights/us-power-grid-2026},
  note   = {Deluair Consultancy briefs}
}

Hossen, M. D. (2026). The grid wakes up: AI demand, FERC 1920, and the PJM, MISO interconnection cliff. Deluair Consultancy briefs. https://deluair.com/consultancy/insights/us-power-grid-2026

Hossen, Md Deluair. "The grid wakes up: AI demand, FERC 1920, and the PJM, MISO interconnection cliff." Deluair Consultancy briefs, 2026-04-26. https://deluair.com/consultancy/insights/us-power-grid-2026.

On the watchlist

Upcoming dates that bear on this brief.

See the full firm watchlist for the rest of the calendar.

December 15, 2026 Energy

EIA STEO winter heating outlook

Henry Hub winter premium with 14 plus Bcf/d feedgas demand, JKM/TTF arbitrage, and DUC inventory below 4,200.

Related insights

Adjacent reading.

Energy and transition economics

The grid wakes up: AI demand, FERC 1920, and the PJM, MISO interconnection cliff