On October 9, 2026, the Federal Energy Regulatory Commission issued the second of two orders that together define how PJM Interconnection will — and will not — handle the largest load growth in its history. The first, covered earlier in this publication, accepted PJM’s one-time Reliability Backstop Procurement and then suspended it for five months. The second, in Docket No. ER26-3515-000, concerns a different mechanism: the Interim Resource Adequacy Service, or IRAS, PJM’s plan to let new large loads connect without matching new generation if they accepted being curtailed first in emergencies.
FERC rejected IRAS, the Bring Your Own New Capacity (BYONC) rules that would have let loads escape curtailment by supplying their own new generation, the requirement that load-serving entities designate new capacity for new large loads, and the proposal to exclude uncovered new large loads from the capacity market’s demand curve starting with the 2029/30 delivery year. What survived: PJM’s definition of a Large Load — end-use customer load with a cumulative peak of 50 MW or more at a single electrical site — and a Large Load Registry to track those loads, both effective October 12, 2026. The vote was notational and unanimous among the five commissioners, with Commissioner David Rosner writing a concurrence that may matter more for what comes next than the order itself. FERC acknowledged what it called the imminent capacity shortfall in PJM. It rejected the tool, not the diagnosis.
The proposal, filed August 13, was PJM’s answer to a specific arithmetic problem. PJM has said that beginning with the 2027/28 delivery year, it will be short of the capacity needed to meet its regional reliability requirement, driven particularly by data-center load growth and the pace of data-center interconnection. Its filing projected peak load growth of 32 GW from 2024 to 2030, roughly 30 GW of it from data centers. The July capacity auction for 2028/29, covered in our earlier reporting, procured 138,318 MW of unforced capacity at the $325/MW-day cap and still fell 6,831 MW short of the reliability requirement — the second consecutive auction to miss.
IRAS was designed as a bridge across that gap. New large loads — those arriving on or after June 1, 2027, under the plan — whose suppliers had not lined up new capacity for them would fall under a new curtailment service: in an emergency, PJM could direct electric distributors to reduce load from those Eligible New Large Loads, ahead of other customers. The escape hatch was BYONC: a load that brought qualifying new capacity, either through the backstop procurement or through its own arrangement, could reduce or eliminate its IRAS exposure. A third element would have excluded uncovered new large loads from the capacity auction’s demand curve starting in 2029/30, so existing customers would not bear higher capacity costs driven by loads that brought no supply of their own. The logic was cost causation. IRAS offered large loads a choice rather than a bill — bring supply, or accept the risk of being switched off first.
Why FERC said no
FERC’s central objection was jurisdictional, and it turned on the gap between what PJM described in its cover letter and what its tariff actually said. PJM’s transmittal letter said it would tell electric distributors only how many megawatts to shed, leaving the choice of customers to the states. The tariff as filed, FERC found, did something else: it appeared to place obligations directly on the Eligible New Large Loads themselves and to require electric distributors to curtail that specific class of retail customers. That distinction is the whole ballgame under the Federal Power Act. FERC governs wholesale markets; states govern retail service and retail rates. The loads IRAS would have curtailed are, in FERC’s words, for the most part retail customers taking retail service. A federal tariff that dictates which retail customers get cut, and that sets up a compensation framework FERC read as paying those loads with money collected from other retail customers, appears to establish a retail rate, a matter for state commissions.
Jurisdiction was the headline objection, but FERC gave two more reasons any refiling must survive. First, discrimination: PJM had not offered sufficiently detailed evidence, FERC found, that treating load-serving entities differently based on their retail customers’ characteristics — size, vintage, whether they brought new capacity — is not unduly discriminatory. PJM’s capacity market has historically bought capacity for all network load alike; a regime that sorts loads by size and arrival date and curtails one class first needs a record showing why those characteristics matter to resource adequacy, and FERC found that record wanting. It rejected the New Large Load definition without prejudice. Second, accreditation: the BYONC rules would have locked in a resource’s capacity accreditation — its credited contribution to reliability — for the life of its arrangement, a lock-in FERC found insufficiently justified. Accreditation determines what resources earn, and freezing it for BYONC resources while the rest of the market’s accreditation evolves would create a two-tier system with no demonstrated basis.
FERC also flagged a reliability concern. Under Reliability Standard EOP-011-4, firm load may be involuntarily curtailed only after the reliability coordinator initiates an Energy Emergency Alert Level 3 (EEA3). PJM’s proposal would have curtailed certain firm load involuntarily earlier, at EEA1, which FERC said appears to ignore the standard’s plain language. Commissioner Rosner’s concurrence sharpened the point into a broader critique of sequencing: PJM’s filing, he wrote, flips the accepted paradigm by proposing to force IRAS load off the system in the earliest stages of emergency conditions, before PJM deploys demand response — the customers who are paid to volunteer for exactly this purpose.
The most consequential passage in the concurrence concerns the demand curve. IRAS would have excluded uncovered new large loads from the capacity market’s demand curve — the sloped curve, described earlier in this series, that sets the capacity price based on how much supply is offered against the target. Rosner warned that excluding those loads would produce a market signaling that PJM has more than enough capacity while the region sheds thousands of megawatts of demand on an involuntary basis. Remove a large and growing class of load from the forecast, and the curve reports that less capacity is needed than the physical system actually requires. Prices soften, investment signals weaken, and meanwhile the excluded loads still draw power in normal conditions and still get cut when conditions tighten. Rosner’s bottom line was a principle and an assignment: in an emergency, data centers not serving critical functions should be shut off before ordinary customers — homes, hospitals, and schools should never be curtailed before data centers — but states, not PJM, should decide the curtailment order, and such measures should come only when necessary to keep the system reliable, after voluntary demand response has been exhausted. The concurrence sketches the shape of whatever comes next: state-set priorities, voluntary flexibility first, mandatory curtailment last.
What survived: the definition and the registry
The two pieces FERC accepted deserve more attention than consolation prizes. The Large Load definition counts all affiliated load facilities within a one-mile radius as one site — an anti-gaming provision FERC said will prevent an owner from escaping the definition by siting several smaller data centers, each below 50 MW, close together. FERC also declined to limit the definition to data centers.
The registry requires load-serving entities to report each large load they serve to PJM — location, in-service dates, peak load and ramp schedule, telemetry, backup generation. Detailed data stays confidential under PJM’s existing rules; the public gets aggregated figures on PJM’s website. FERC found the registry valuable separate and apart from IRAS, because PJM needs visibility into all large loads, existing ones included, to forecast load growth and plan its transmission system. That visibility is the foundation for everything that follows: you cannot allocate costs to loads you cannot see, or set curtailment priorities for customers you cannot identify. The registry does not add a megawatt or cut a bill. It makes the next decade of argument proceed from a shared set of facts rather than competing guesses about what is connected.
Who pays now
With the cheapest route in closed, the cost of AI’s grid access now travels through two remaining doors. The first is the bilateral door: private deals in which large loads contract directly for new supply. That door was never dependent on IRAS. On October 6, Constellation Energy called its 890 MW nuclear uprate deal with Google a direct response to PJM’s bring-your-own-power proposal. The proposal it answered has now fallen, but the deal outlives it. PJM’s matchmaking process connecting large loads with developers also continues.
The second door is the state door, and FERC’s order pushed the hardest questions through it. States must now decide curtailment priorities, and they are already deciding cost allocation through retail rate cases. The clearest template is Ohio: in July 2025, the Public Utilities Commission of Ohio approved a rule for AEP Ohio’s largest data-center customers requiring loads above 25 MW to pay for at least 85 percent of the power they reserve each month, whether they use it or not, for up to 12 years, with exit fees for walking away. The effect on the queue was immediate — AEP Ohio, which had faced interconnection requests totaling more than 30 GW, saw speculative requests fall away once requesting power came with a price. Jefferies analysts named Ohio and Pennsylvania as the key states to watch for state-led revival of the IRAS concept.
The default path is the regular capacity market, where costs land on everyone. The July 2028/29 auction cleared at the $325/MW-day cap, and PJM has said that without the FERC-approved collar, the auction would have cleared at nearly $555/MW-day across most of the footprint and $777/MW-day in the Commonwealth Edison zone, pushing total costs to roughly $29.7 billion instead of $16.4 billion. The cap protects consumers from that bill. But a market clearing at its cap while missing its reliability requirement two auctions running is rationing by something other than price. Jefferies said the IRAS rejection could mean higher capacity prices for longer in PJM, favorable to independent power producers such as Constellation, NRG Energy, Talen Energy, and Vistra.
Three clarifications matter. First, the rejection does not stop data centers from connecting; new large loads still interconnect, they simply no longer face a federal tariff assigning them mandatory curtailment priority if they bring no new capacity. Second, it does not add a megawatt: the 6,831 MW shortfall from the July auction is untouched, and the separate backstop procurement designed to fill it remains suspended. Third, it does not hand the whole question to the states: FERC set no state deadline, is still deciding backstop cost allocation itself, and invited PJM to refile. The timing underscored the stakes: the registry’s October 12 effective date coincided with PJM’s first-ever Capacity Advisory, issued October 9 for October 12 and 13, with PJM forecasting peaks of about 102,000 MW and 104,000 MW against roughly 70,000 MW of scheduled generation outages.
What happens next
The calendar now has four dates. October 22 brings a special PJM Members Committee meeting to consult with members — but that meeting concerns the backstop procurement refile, not IRAS. October 29 is PJM’s stated target for filing a revised backstop procurement proposal in Docket No. ER26-3380, answering FERC’s September 29 order. That date belongs to the backstop case, not IRAS. The third date is the 30-day compliance filing in this docket, ER26-3515 — a deadline that falls on a Sunday, so effectively about November 9. The fourth is December, when PJM runs the Base Residual Auction for the 2029/30 delivery year. Because FERC rejected the demand-curve exclusion, PJM must now model all forecast demand for that auction.
PJM has also told FERC it intends to make Section 205 filings in the separate large-load show-cause proceeding by early to mid-November, including new transmission services for flexible large loads. A voluntary version of the IRAS idea could surface there: a non-firm service a load chooses in exchange for faster or cheaper connection. That opt-in bargain is the most plausible shape of a revival.
Two things will determine whether this order is remembered as a detour or a dead end. The first is the states. FERC told them they have a crucial role in designating which retail customers must be curtailed when the system is stressed. Watch Ohio and Pennsylvania first, and watch whether other PJM states adopt take-or-pay rules or curtailment priority orders of their own. The state that writes the first clear curtailment-priority rule will set the template for thirteen others.
The second is the December auction. With all forecast demand back in the demand curve, the 2029/30 auction will show whether two consecutive shortfalls were an anomaly or the new normal. The bridge PJM tried to build has lost one of its two spans: the backstop is suspended, the curtailment service rejected. The registry remains. The loads keep coming. And the question FERC declined to answer — who pays for AI’s grid access — now sits with the states, the bilateral market, and a capacity auction clearing at its cap.