On Thursday, PJM’s independent market monitor delivered a report with two verdicts that point in opposite directions. The energy market was competitive in the first nine months of 2026, according to Monitoring Analytics’ 2026 Quarterly State of the Market Report for PJM: January through September, released November 12. The capacity market was not — in any of the last four auctions, for delivery years 2025/2026 through 2028/2029 — “primarily as a result of the inclusion of forecast demand for data centers.” The monitor puts the cost of that inclusion at $29.4 billion, calls the increase irreversible, and warns the total “will increase significantly when PJM runs its next capacity auction in December 2026.”
The split verdict matters because it separates two arguments: prices — wholesale power cost $111.61 per MWh in the first nine months of 2026, up 40.9 percent from $79.19 a year earlier — and design, the capacity market pricing a forecast of data-center load rather than the load itself. The report turns a political argument about data centers and bills into an audited number, and puts the December auction at the center of the cost debate, on the same day PJM convenes a workshop to discuss redesigning the market.
The nine-month bill, component by component
Start with the totals. Energy is the largest component of wholesale cost at 63.5 percent, capacity 17.4 percent, transmission 17.2 percent. The total cost of energy per MWh rose $22.42, or 46.3 percent, from $48.47 to $70.89. The total cost of capacity per MWh rose $8.70, or 81.1 percent, from $10.72 to $19.42. Transmission barely moved: up $0.84, or 4.6 percent, to $19.16 per MWh.
The energy price itself tells a similar story. The real-time load-weighted average locational marginal price rose $24.47 per MWh, or 48.4 percent, from $50.51 to $74.98. The largest piece is not fuel: the transmission constraint penalty factor accounted for $8.81 of the $24.47, or 36.0 percent, against $5.73 (23.4 percent) for fuel and consumables. The energy market is expensive this year in large part because the wires are full, not because the fuel is dear.
That distinction is the report’s quiet rebuke to anyone who would read a 48 percent LMP increase as evidence of a broken energy market. Prices were set, on average, by units operating at or close to their short-run marginal costs, the monitor says — “evidence of generally competitive behavior and competitive market outcomes,” though high markups for some marginal units did affect prices. Load grew, too: the real-time hourly average load rose 3.4 percent, from 93,683 MWh to 96,845 MWh. The energy market absorbed more load and more congestion, and still produced competitive results. The capacity market did the opposite.
The $29.4 billion, and the dispute over what it means
The monitor reached the same conclusion about the same four auctions in its first-half report; the nine-month report restates it, and the figure has not moved. Including existing and forecast data-center demand in the capacity market “has resulted in an irreversible increase of $29.4 billion in capacity market costs for all customers to date.” That is the monitor’s estimate, and it is disputed by PJM: the grid operator has said the monitor’s total “does not equate to the total cost to load.”
The dispute is about what the number represents, not the arithmetic behind it: the monitor’s figure is a counterfactual — what capacity revenues would have been without data-center load in the demand curve — while PJM’s objection is that a change in auction revenues is not the same as a change in what customers ultimately pay, given hedging, rate cases, and allocation rules that delay and redistribute wholesale costs before they reach a retail bill. The monitor’s own framing is more precise than the headline: across the four auctions, data-center load accounts for $29.4 billion of $63.6 billion in total capacity market charges — 46.2 percent. In the most recent auction alone, for the 2028/2029 delivery year, the monitor estimates $6.3 billion of the $16.4 billion in charges — 38.2 percent — was a result of data-center load.
In the first nine months of 2026, the total cost of wholesale power would have been $101.66 per MWh if not for data-center load and its impact on capacity costs for the 2025/2026 and 2026/2027 delivery years, against an actual $111.61. The difference is $9.95 per MWh, or 9.8 percent — the data-center adder for the nine-month period, covering the capacity market only and excluding any data-center effect on energy or transmission costs. That adder is smaller than the first-half figure of $11.11 per MWh — the arithmetic of a fixed charge spread over more megawatt-hours.
The direction of the total, however, is the other way. Each auction buys capacity for a single delivery year, and once an auction clears, the commitments are struck and the payments flow whether or not the load that justified them ever connects. That is what “irreversible” means here: the $29.4 billion is the sum of incremental revenues across four one-year delivery years, not an annual run rate, and no revision to the forecast can claw any of it back.
Why the forecast is the problem, not the load
The monitor’s case is not that data centers should not use power. It is that a capacity market procures for the future on the basis of a forecast, and it does not true up when the forecast is wrong. Data-center load included in the capacity auctions rose from 7,927 MW in the 2025/2026 auction to 11,993 MW, then 17,071 MW, then 21,474 MW in the 2028/2029 auction — most of it forecast rather than connected. When that forecast load enters the demand curve three years ahead, the auction clears higher, and every load-serving entity pays that price on its full obligation, including the share attributable to load that never connects. There is no refund mechanism for a forecast that shrinks.
The scale of the forecast’s error is documented in the monitor’s own filings. In a September 25 answer in the docket over PJM’s Reliability Backstop Procurement, the monitor calculated that 4,016.9 MW of forecast data-center load — 42 percent of the growth priced into the two most recent auctions — had already been canceled or delayed as of September 21, based on project-status tracking by Industrial Info Resources, counting only projects originally scheduled to be in service by 2029. The monitor disclosed a limitation: it does not have access to the project-level data behind the large load adjustments that distribution companies submit to PJM, so its count is an independent check built from outside the forecast process.
The monitor translated the canceled and delayed load into the auction’s own terms. Start from the 2028/2029 auction’s actual result: a 6,831.3 MW shortfall against the reliability requirement, after clearing at the $325 per MW-day cap and procuring 138,318 MW of unforced capacity. If data-center load had stayed at the level forecast for the 2026/2027 auction — 11,993 MW instead of 21,474 MW — PJM would have had a 1,096.5 MW surplus rather than a shortfall. If only the 4,016.9 MW of canceled or delayed load is treated as never arriving by the delivery year, the shortfall falls to 3,055.0 MW, a 55 percent reduction. What the table demonstrates is that the shortfall is highly sensitive to a forecast that has already moved by more than 4,000 MW — with the caveat that delayed is not canceled, and the scenario assumes the identified load does not come online by the delivery year at all.
The forecast is also older than the auction it priced: it rested on large load adjustments PJM received before September 5, 2025 — more than ten months before the auction cleared — and PJM’s backstop filing contains no provision to revise the load forecast or the procurement target before the procurement runs. That is also why the backstop is so consequential: it would convert a one-year auction shortfall into 15-year commitments, locking in payments for capacity sized to a gap the monitor says is already more than half overstated. FERC accepted the backstop filing on September 29 but suspended it until February 28, 2027, pending a paper hearing. PJM refiled on October 29 with a far smaller target — roughly 630 MW, by the latest estimates — after utility forecast revisions shrank the gap.
The congestion ledger: $4.9 billion, and $7.6 billion not returned
The report’s second ledger is the one about transmission, and it is the piece of the nine-month findings most likely to be overlooked in the capacity debate. When transmission constraints bind, customers pay more for energy than generation is paid to produce it. The difference is congestion revenue. Total congestion rose $2,674.0 million, or 119.7 percent, from $2,233.8 million in the first nine months of 2025 to $4,907.8 million. Congestion revenue, the monitor says, is paid by customers, belongs to customers, and should be returned to customers. But only 72.8 percent of total congestion paid by customers for the first four months of the 2026/2027 planning period was returned through the auction revenue rights and financial transmission rights offset. Customers have received $7.6 billion less in congestion revenues than they should have since the 2011/2012 planning period, the monitor calculates, as a result of flaws in the ARR/FTR market design. That is a separate failure from the capacity market, but it is connected to the same load wave: data-center load is concentrated in a small number of transmission-constrained pockets, and transmission constraints were the largest single driver of the energy price increase this year.
The investment signal is working — which is the awkward part
There is a finding in the report that complicates any simple story of market failure. Theoretical energy market net revenues — the measure of the incentive to invest in generation — rose sharply in the first nine months of 2026: up 105 percent for a new combustion turbine, 76 percent for a new combined cycle, and 21 percent for a new solar installation. The energy market is paying new entry substantially more than it did a year ago.
That is the awkward part for the redesign debate: the two markets are doing different jobs, and only one of them is broken. The monitor’s recommendation follows: exclude forecast data-center demand from the capacity market, and require data centers to bring their own new generation. The monitor also accuses PJM of failing to implement the public policy mandate in the principles agreed to by all the PJM governors and the White House, and in the Ratepayer Protection Pledge, to ensure that data centers pay their own costs.
December is the next compounding event
The report’s warning about December is the sentence that should frame everything that follows. The Base Residual Auction for the 2029/2030 delivery year closes December 15 — the fourth and final auction under the price collar, capped at $337.83 per MW-day, against a reliability requirement of 158,712 MW, 2,699 MW higher than the year before. It is the first auction in which generators must weigh a one-year commitment at the capped price against holding their best offers for the backstop’s 15-year contracts at up to $555 per MW-day. If the auction clears at the cap again — and with a higher requirement and a forecast that still includes more than 20,000 MW of data-center load, the monitor’s expectation is that it will — the $29.4 billion grows, and the growth is irreversible in the same way the first four auctions were.
That is why the monitor’s report lands with such force on the day PJM convenes its workshop. Today, from 9 a.m. to noon, PJM staff and stakeholders work through more than 30 papers filed in response to the grid operator’s May white paper, which offered three broad pathways and recommended none of them. The monitor’s report is not one of those papers, but it supplies the number every paper will have to answer: $29.4 billion, irreversible, with more coming in December. The workshop’s redesign is a multi-year project by PJM’s own timeline, while the next auction closes in a month. The report does not resolve that mismatch. It prices it.
What to watch
December 15: the 2029/2030 Base Residual Auction closes — the last auction under the price collar, and the first test of whether the monitor’s warning about a significant increase in the $29.4 billion total is borne out. The clearing price, the shortfall against the 158,712 MW requirement, and the data-center load in the demand curve are the three numbers to read first.
January 31: the date by which Exelon has asked FERC to rule on its complaint seeking a prompt capacity auction, held two months before the delivery year, starting with 2030/2031. FERC has not ruled. February 28, 2027: the date FERC’s suspension of the Reliability Backstop Procurement expires, pending the paper hearing on cost allocation, collateral, and transmission-owner exit rules.
And the forecast itself: the Large Load Registry took effect October 12. Its first test is whether the December demand curve reflects the 4,017 MW of canceled and delayed load, or prices the old forecast one more time. The $29.4 billion is the price of the old forecast. What happens next depends on whether the institutions that run the market update the forecast before they charge for it again.