America's largest power market cannot buy enough capacity

PJM stands for Pennsylvania-New Jersey-Maryland, the states whose utilities first pooled their power plants nearly a century ago, and the initials stuck as the footprint grew far beyond them. The job remains what it was then – keeping enough power plants available to serve the region – but the tool is now a yearly capacity auction. In the auction, PJM lines up offers from power plants from cheapest to most expensive and buys down the list until the region's needs are covered. The last plant it has to buy from sets the price for every plant that cleared, so the price tells you what that final megawatt costs.

That reading depends on the offers reaching far enough to cover the need, and this year they did not. This year's auction did two things at once: it cleared at its legal price ceiling and still came up short of the capacity the region needs. Think of the message you get at an online checkout when a payment will not go through. The system tells you that it cannot complete your request. The USD 325/MW-day figure PJM printed looks like a price, but it is really the same kind of warning: the auction hit the rules’ maximum because the order could not be filled.

The auction cleared 138,318 MW, measured in what the market calls unforced capacity, meaning capacity discounted for the hours a plant is likely to be unavailable. Adding the 10,864 MW committed outside the auction by utilities that supply their own plants brings the total to 149,182 MW – against the 156,013 MW PJM judged it needed. The region finished 6,831 MW short, the second year running and almost identical to last year's 6,516 MW gap. That leaves a 14.4% reserve margin, the spare generation held above expected peak demand, against PJM's 20% target – thin enough that a prolonged summer heat wave or a deep winter cold snap would leave little room for error.

Supply ran out before the auction could reach a plant that would set the price, so every sub-region simply landed at the ceiling. The widely quoted USD 16.4 billion price tag is also not what customers pay, and PJM says so plainly, because most buyers are covered by their own plants or by contracts that never touch the auction price. Because the auction came up short, that bill is about USD 0.8 billion smaller than if PJM had bought everything it needed, so a worse shortage produced a smaller headline.

The decline in the published clearing price, from USD 333 in last year's auction to USD 325 in this year's, reflects a changed measuring stick rather than a cheaper market (see figure 1). PJM writes the ceiling and floor against a plant's nameplate size, the maximum output it is built to deliver, and those limits have been frozen since the 2026/2027 auction at USD 256.75 and USD 138.25 per MW-day. These amounts are simply the most and least PJM is allowed to pay per nameplate megawatt, calibrated to the cost of building the benchmark new plant PJM uses as its yardstick.

A second number informs the published price. This is the translation factor between a plant's nameplate size and what it is expected to deliver. It exists because no plant runs every hour. PJM discounts each plant's nameplate size down to what it expects to actually deliver when the grid is stressed and pays on that lower figure, so a ceiling written per nameplate megawatt has to be restated before it can be quoted as a price. That translation factor is the only thing that actually moved, from about 0.77 last year to about 0.79 this year. That is the entire explanation for the decline. The ceiling did not move, but the translation factor did, and the same frozen ceiling divided by a slightly larger factor produced a slightly lower published price. In the units that actually set it, the price has been sitting at its legal maximum for three consecutive years.

The obvious explanation for the lack of new capacity is that the ceiling held the price down and discouraged developers from bringing new plants into the auction. PJM tested this, re-running the auction on paper with the ceiling removed and the same offers in place. The price jumped to about USD 555/MW-day and the dollars changing hands ballooned by USD 13.3 billion to a total of USD 29.7 billion. In this scenario, no more than 77 MW of extra supply was ultimately pulled in.

Only about 525 MW of new plants and expansions cleared, down from roughly 774 MW a year earlier, and half a gigawatt in a region this size is implausibly small. The cap is the least of the reasons so little new capacity showed up. The most relevant factor is time. This auction ran about 23 months before the delivery year would begin, versus the regular period of at least three years, and nothing substantial can be sited, permitted, ordered and built that fast. Even PJM's fast-tracked projects are expected to come online by 2031 rather than 2028. Generation built behind the fence for data centers is a further drain.

For a business, the auction outcome manifests as a capacity charge on the power bill, and how heavily it lands depends on how many hours you run. Think of the capacity charge as your share of a giant insurance premium that keeps the grid reliable. Your share is set by how much electricity you were pulling during the few hours each year when the system hit its peak. PJM grosses that up slightly to cover reserves, then multiplies by the clearing price and by 365 days.

Let's put a 10 MW user through that math, a mid-sized factory, a cold-storage warehouse, or a hospital campus. In 2024/2025, the annual capacity charge for a 10 MW user came to roughly USD 115,000. By 2028/2029, the same charge has risen to roughly USD 1,115,000. That is roughly USD 1 million a year in additional cost for the same operation. This tenfold jump is a system-average comparison, so it does overstate the increase for eastern customers, whose congested zones were already priced far above the system average. For these customers, the increase is closer to five or six times.

A commercial or industrial bill splits into delivery charges for the local wires, transmission charges for the high-voltage network, and the supply charge for the electricity itself. Because the capacity is not usually shown as a separate line item but sits inside the supply portion of the bill alongside the cost of power itself, many customers may not notice it until contracts reset. For a customer running about 60% of the time, the new charge works out to roughly USD 21/MWh (megawatt-hour). Against all-in commercial and industrial rates of about USD 85/MWh in West Virginia to well over USD 200 in Washington DC, that makes capacity roughly 15% of a typical bill, and more than 25% of the bill in the cheapest jurisdictions. The burden is greatest in the lowest-share price jurisdictions, where the capacity charge takes the largest share of the bill and reaches customers through a market most of them never participate in directly.

The same charge of about USD 111,500 per MW of peak demand also becomes a very different cost per unit of electricity depending on how much of the time you run (see figure 2). A hyperscale data center running near 95% of the time pays about USD 13/MWh, while a stop-start industrial plant running closer to 40% pays about USD 32 for an identical charge. The data center that is blamed for the shortage absorbs it most easily, because the cost barely registers next to its real problem, which is getting connected at all.

None of this will wait until 2028: the price increase is already flowing into bills. Each auction runs about two years ahead of the power it buys, so what a business pays today was set by an auction held two years ago, and the steep prices of recent auctions are only now showing up on bills. Through May 2026, the capacity price flowing into supply contracts was USD 269.92/MW-day, from the 2025/2026 auction. On June 1, 2026, it jumped to USD 329.17, and USD 333.44 and USD 325.00 are already locked in for the next two years. For a 10 MW customer, that is roughly USD 200,000 of additional annual capacity cost arriving this summer, held at about that level through 2029. Anyone waiting to see whether the capacity crunch reaches their electricity bill has already missed it.

The timing of the hit depends on how you buy power, but almost nobody is exempt for long. Default utility service recovers capacity in periodically reset rates, so it arrives late and buried. A fixed all-in contract shields you for its term and then reprices at renewal, which is the cliff most likely to catch a business off guard. On a pass-through or index contract, you will see the increase immediately and in full.

The auction can only buy plants that already exist or are nearly finished, which is why it was never going to close a 6.8 GW (gigawatt) gap. Because it has come up short twice, a third shortfall would trigger PJM's Reliability Backstop, which lets PJM contract directly for new capacity instead of waiting for the annual auction. Stakeholders advanced the design at the end of June, and PJM intends to seek approval for a special procurement as early as September 2026. The target is about 14.9 GW able to come online by June 2031, bought first through a window where large users and utilities contract directly with new plants, then through a central PJM purchase for the remainder, at an average cost capped at USD 555/MW-day.

Although PJM has not published an expected split between firm generation, renewables, and batteries, and the program is technology-neutral on paper, the best guide is its fast-track predecessor PJM’s Expedited Interconnection Track: twelve new-build projects consisting of six gas plants, five batteries, and one nuclear unit, with no standalone solar or wind. Between a 2031 deadline and penalties for non-delivery, the practical effect is a gas and storage procurement, whether it is labelled as such or not.

This is why the backstop matters more than the headline. Nobody commits capital to a plant against a capped, two-year-forward price an administrator could change again next year, but a project can be built on multi-year contract with a fixed price and a named counterparty. The real capacity market is quietly moving out of the annual auction and into those longer contracts, a step toward the model California has used for years, where suppliers must contract ahead for resource adequacy and the state has increasingly moved to central procurement for whatever nobody buys voluntarily. PJM would arrive at the same place from the opposite direction.

Whether the large new users that drove the shortage pay for the backstop themselves, or the cost is spread across everyone's bill, is now being decided through PJM's stakeholder process. Meanwhile, the compressed timeline tells those users the wholesale market will not deliver firm power on their schedule, which continues to push them to build their own behind-the-meter (BTM) power solutions, the parallel system we described in an earlier article.

A shortage of about 6.8 GW closes in only one of two ways. Either supply rises, which it barely has, or the capacity required by the region drops, which happens when large loads such as data centers cut back, build their own supply, or never arrive.

That is where the shape of the charge becomes a policy problem. The customer facing the steepest per-unit cost, roughly USD 32/MWh, is the stop-start industrial plant, and it is usually the one with the least room to do anything about it. A hyperscaler can finance its own generation and has teams dedicated to the problem. A single-site food processor or metal fabricator generally cannot, and its options come down to moving production into off-peak hours, trimming a peak, or absorbing the cost. This means the cheapest way to solve PJM's shortage is for those industrial customers to shrink, relocate, or cancel an expansion, balancing the region's capacity books by hollowing out its manufacturing base. How PJM and the states allocate the backstop cost will decide whether the risk is reduced or locked in.

With a MW of avoided peak demand worth about USD 111,500 a year, demand-side offers should be flooding in. Instead, demand response fell 277 MW even after a May 2025 rule change made it more valuable. A business that wants to look at their demand response options can start with PJM's demand response page and its directory of curtailment service providers.

One hedge that does not work here deserves calling out. The virtual power purchase agreement (VPPA) many companies sign to fix their power price settles against only the energy portion of the bill. Capacity sits outside it entirely and lands on top, in full, so a company holding a VPPA has fixed one component of its supply cost, believes it is hedged, and will still absorb the whole capacity increase.

What does hold up is BTM equipment that delivers a measurable saving, a battery that discharges during peak hours, on-site backup generation run during the peak hours of the day, or controls that shift production away from the few hours setting your bill. The other is a new power plant backed by a backstop or bilateral contract, where a creditworthy counterparty pays a fixed price for a fixed term.

Three things will decide how this plays out. Whether a third short auction triggers the backstop, the resolution to the fight over who will pay the bill, and whether regulators extend or reset the temporary cap and floor when the current arrangement covering the next two auctions runs out. The price was never really the point. The order could not be filled, and what matters now is who moves first to fill it.