Load Follows Generation

Load Follows Generation: How AI Data Centers Are Quietly Repealing a Century of Utility Planning Logic

Beneath the consensus that AI is straining the grid, FERC's December 2025 PJM colocation order, EIA's 2026 retirement-delay data, and a 66% surge in gas-plant build costs together signal a structural inversion: where new electric load now chooses to sit next to existing or new generation, slowing coal retirements and locking in fossil capacity inside the energy transition.

The consensus 2026 energy-transition narrative is familiar: AI data centers are blowing through grid capacity, interconnection queues are the bottleneck, the answer is more transmission, renewables and storage. Beneath that headline a quieter inversion is under way. For a century US grid planning treated load as fixed and generation as variable; the new pattern reverses that. Data centers now site next to existing nuclear and gas plants, build their own gas peakers, and use behind-the-meter generation to bypass queues entirely. The board question is no longer "when will the grid catch up?" but "what generation gets locked in — and what gets cancelled — when load stops waiting for the grid?"

Signal Identification

A structural inversion in US electricity-system planning. The signal is regulatory and operational acceptance that load now sites itself next to generation, restructuring coal-retirement timing, gas-peaker economics, transmission planning, and the de-facto fuel mix of the AI-era energy transition — a regulatory and infrastructure pivot.

Time horizon: 2-5 years (PJM colocation rules from Feb 2026; gas-turbine supply tight to early 2030s; structural lock-in 2027-2030) Plausibility band: High Geographic / Jurisdictional Scope: Primary: PJM, ERCOT, MISO, SPP. Spillover: UK, Ireland, Germany, the Nordics, Japan. Sectors exposed: Hyperscale data centers; nuclear and gas IPPs; gas-turbine OEMs; utilities and RTOs; renewables developers; net-zero reporting.

What's Changing

On 18 December 2025 (FERC, 18/12/2025) ordered PJM to develop rules for colocating data centers and large loads at power plants, ruling the existing tariff and behind-the-meter generation rules "unjust and unreasonable." PJM compliance filings are due 20 January and 16 February 2026 establishing three new transmission services. (Utility Dive, 19/12/2025) reports Capstone analysts calling the decision a "major victory" for Constellation, PSEG and Vistra; FERC Chairman Laura Swett framed it as enabling AI development without ratepayer cost-shifting.

The fossil consequence is visible in retirement data. (EIA, 23/02/2026) reports owners planned to retire 12.3 GW of US capacity in 2025 but retired only 4.6 GW — the least since 2008 — after DOE emergency orders extended coal plants. Six coal retirements totalling 3.5 GW were postponed from 2025 to 2026, and "any renewed or new emergency orders could affect retirements planned for this year." (RMI, 17/03/2026) places this against a 2.2 TW backlog of generation and storage projects in interconnection queues, with average request-to-COD time risen from under 2 years in 2008 to nearly 5 in 2024.

Gas peakers are the de-facto new build. (TechCrunch, 27/04/2026), citing BloombergNEF, reports CCGT build cost up 66% in two years to $2,157/kW, construction 23% longer, gas-turbine prices up 195% over 2019, waitlists into the early 2030s. (Utility Dive, 13/04/2026) reports PJM proposing a 14.9 GW two-phase plan including bilateral matchmaking between data-center buyers and generation sellers — the first PJM central capacity procurement, evidence the grid operator is now a load-to-generation broker.

The retirement reversal: planned vs actual coal capacity exits

Sources: EIA 23/02/2026; TechCrunch (BloombergNEF / Wood Mackenzie) 27/04/2026.

Disruption Pathway

Stage one (mid-2026 to end-2026): PJM and other RTOs publish colocation tariffs; aging coal plants receive RMR or DOE emergency-order extensions through 2027-2029; hyperscalers accelerate behind-the-meter gas builds at nuclear-adjacent and greenfield sites. Stage two (2027-2028): the gas-turbine OEM bottleneck (Mitsubishi, GE Vernova, Siemens Energy) caps new-build pace and concentrates rent among IPPs with existing assets; central capacity procurement spreads from PJM to MISO, ERCOT and SPP; sovereign and corporate PPAs lock in 15-25 year gas and nuclear offtake. Stage three (2028-2030): the grid becomes a federated system — bulk public service plus an archipelago of colocated private grids — with renewables relegated to grid-edge build and corporate PPAs.

Stresses concentrate at three points. Ratepayer cost-shifting is the political flashpoint; (ITIF, 07/04/2026) notes PJM capacity auction prices rose from ~$60/kWh in 2024 to >$300/kWh in 2025; large-load tariffs with 15-year minimums and 85% load guarantees are emerging to ringfence costs. Gas-turbine supply is the hard physical bottleneck. Net-zero credibility is the third stress: each delayed retirement and new gas plant lengthens the 2030-2035 decarbonisation glide path. Operationally, hyperscaler procurement displaces utility-led IRP planning; institutionally, FERC, state PUCs and DOE evolve toward bespoke large-load-tariff frameworks that codify private-grid status alongside bulk-system service.

Why This Matters

For utility boards and IPP CFOs, generation-adjacent siting reorders the value chain: dispatchable capacity (nuclear, CCGT, peakers) and existing interconnection rights become the structural asset, hyperscaler PPAs the new annuity. Renewables developers face queue-bypass economics that tilt against marginal solar-plus-storage unless they can colocate too. For institutional ESG investors, the implicit decarbonisation timeline is slipping 3-5 years; portfolios should reweight transition-risk exposure. For policymakers, the question is no longer whether to permit colocation but how ratepayer protections, reliability and net-zero credibility survive it.

Decision-action posture for this signal: Prepare — the inversion is codified in FERC tariff law and visible in retirement data; gas-turbine supply and the ratepayer fight will set 2027-2030 trajectory, but positioning windows remain open for IPPs, hyperscalers and renewables developers willing to commit this cycle.

Counter-Argument

The strongest objection is that the load-follows-generation framing overstates a transitional adjustment. (ITIF, 07/04/2026) argues hyperscaler capex runs ahead of revenues; only one-third of announced 240 GW of new US data center capacity is actually being built; deals fell over 40% Q3-Q4 2025; the Stargate flagship has stalled; demand-management, large-load tariffs and hyperscaler "bring-your-own-power" will absorb the shock without permanent fossil lock-in. On this reading, behind-the-meter colocation is transitional, not structural.

This underweights the regulatory and physical lock-in already in place. FERC's December 2025 order is codified tariff law in the largest US RTO, gas-turbine supply is sold out into the early 2030s, and DOE has shown willingness to use emergency Federal Power Act authority to keep coal plants online. Even if hyperscaler demand softens by half, the 5-7 year planning horizon for replacement low-carbon dispatchable capacity is gone; the fossil capacity now being built will run for 30-40 years. The signal is precisely that the architecture is solidifying inside the window where the demand surprise might still reverse.

Implications

The signal catalyses durable change in US electric-system architecture. The FERC PJM order is canonical: a federal regulator certifying that large loads can choose generation and bypass the bulk system, with ratepayer guardrails layered on top. Nuclear and gas IPPs and turbine OEMs capture asymmetric rent; renewables developers face higher hurdle rates outside colocation; utilities are pushed into a brokering role; net-zero architectures lose 3-5 years of glide path inside the same 2050 headline. The grid stabilises as a two-tier system — bulk public service plus colocated private generation — not the integrated all-clean planning model of the prior decade.

Early Indicators to Monitor

PJM tariff revisions for Firm and Non-Firm Contract Demand services accepted by FERC by Q2 2026.
Two or more 2026 coal retirements receive DOE emergency-order extensions by Q4 2026.
MISO, ERCOT or SPP files its own large-load colocation tariff with FERC by end-2026.
A hyperscaler announces a >2 GW behind-the-meter gas-plus-data-center campus outside Texas before mid-2027.
A US state PUC publishes a large-load tariff with 85% minimum-load guarantee and 15-year term by H1 2027.

Disconfirming Signals

Hyperscaler capex pulls back >30% in 2026, freeing gas-turbine slots.
FERC reverses key elements of the PJM colocation order on appeal before end-2026.
A US state PUC blocks behind-the-meter colocation as ratepayer cost-shift, adopted by two or more peer states by 2027.
2026 coal retirements come in at or above the 11.0 GW planned.
SMR or long-duration storage outpaces gas peakers at colocated sites before 2028.

Strategic Questions

Should renewables developers reposition to colocation-eligible sites now, or wait for state tariffs to clarify cost-shift rules?
For nuclear and gas IPPs, what share of capacity should go to 15-25 year hyperscaler PPAs versus merchant capacity revenues?
At what gas-turbine waitlist length does SMR or long-duration storage become economic at scale?
How should net-zero investors reweight transition-risk exposure as fossil lock-in extends 3-5 years?

Keywords

Load follows generation; behind-the-meter generation; FERC PJM colocation order; data center power; gas turbine bottleneck; coal retirement delay; large load tariff; interconnection queue; capacity auction; reliability-must-run; AI energy demand; hyperscaler PPA; CCGT cost surge; net-zero glide path

Bibliography

Source tiers: Tier 1, governments, regulators and intergovernmental bodies. Tier 2, think-tanks, academic institutes, major consultancies and quality data providers. Tier 3, quality journalism and specialist trade press. Tier 4, vendor, company and practitioner sources, used only as directional corroboration.

Tier 1 FERC Directs Nation's Largest Grid Operator to Create New Rules to Embrace Innovation and Protect Consumers. Federal Energy Regulatory Commission (FERC) (18/12/2025).
Tier 1 Retirement delays of U.S. electric generating capacity may continue in 2026. U.S. Energy Information Administration (23/02/2026).
Tier 2 The Interconnection Queue Continues to Be a Barrier to American Economic Competitiveness. Rocky Mountain Institute (17/03/2026).
Tier 2 Four Reasons New AI Data Centers Won't Overwhelm the Electricity Grid. Information Technology and Innovation Foundation (07/04/2026).
Tier 3 FERC orders PJM to craft large load colocation rules. Utility Dive (19/12/2025).
Tier 3 PJM proposes adding 14.9 GW with bilateral contracts, central procurement. Utility Dive (13/04/2026).
Tier 3 Data center demand drives 66% surge in natural gas power plant costs. TechCrunch (27/04/2026).