PJM Interconnection, the grid operator powering 67 million lives across 13 states and Washington, D.C., has failed to secure its mandatory power reserves for the third consecutive year. Results for the 2028/29 delivery year reveal just 138,318 megawatts of cleared capacity against a 156,013 MW reliability requirement—a 6.8 gigawatt deficit. Prices hit the administrative cap of $325 per megawatt-day, established after a 2025 regulatory settlement with Pennsylvania Governor Josh Shapiro to rein in unconstrained forecasts topping $500.
Yet that headline deficit masks a steeper deterioration. Factoring in 10.9 GW from Fixed Resource Requirement entities, total supply sits at 149.2 GW. This leaves a 14.7% reserve margin—well below PJM’s 20% target for a one-in-ten-year outage risk—and exposes an actual 8.7 GW installed-capacity deficit. The nation's industrial heartland is pricing severe electrical scarcity at a regulatory ceiling, not a market clearing level.
The Procyclical Trap and Statistical Mirage
This supply crunch is the product of procyclical market engineering colliding with artificial intelligence demand. When PJM capacity cleared at a rock-bottom $28.92/MW-day for 2024/25, generating just $2.2 billion across the market, utilities rationally retired coal baseload and shelved new capital expenditure. But after Winter Storm Elliott knocked out 57 GW of generation in December 2022, PJM tightened its Effective Load Carrying Capability (ELCC) rules, wiping out dependable capacity ratings on paper.
Simultaneously, AI data centers flooded long-term forecasts—adding over 5 GW annually. Because hyperscale campuses enter utility demand projections via preliminary requests long before breaking ground, existing ratepayers absorbed an estimated $23.1 billion in capacity costs across the 2025/26–2027/28 cycles for unbuilt, unverified load. When capacity repriced—jumping to $269.92 for 2025/26 ($14.7 billion total) and hitting the $329.17 and $333.44 caps thereafter—physical metal in the ground could not respond.
Even the apparent 3.7 GW year-over-year capacity gain for 2028/29 is an accounting illusion. Of that increase, 86% derived from accreditation shifts, returning units, and coal-to-gas reclassifications (accounting for a 5.6 GW gas gain against a 2.9 GW coal loss). Actual new generation and plant uprates contributed just 525 MW—or 14% of the addition. Despite 220 GW across 811 projects clogging PJM’s interconnection queue, development timelines now stretch beyond five years, making the 23-month forward auction horizon useless for underwriting real iron. Flexible demand also evaporated: Price Responsive Demand and distributed energy resources yielded zero cleared offers.
The 15-Year Backstop Gamble
To stem the bleeding, PJM proposed a Reliability Backstop Auction offering 15-year, pay-as-bid contracts capped at $555/MW-day—with commercial operation permitted until June 2032. At PJM's 9.5% discount rate, a 1 GW award yields a gross present value of $1.6 billion.
However, this backstop is a long-term financing mechanism, not a 2028 reliability fix. Pay-as-bid structures lacking conventional market-power screens inevitably reward developers for guessing the grid’s maximum willingness to pay rather than bidding true marginal cost. Worse, because regulatory filings remained pending as of July 14, these contracts face protracted litigation, leaving near-term deficits to be managed via emergency imports, retirement delays, and industrial curtailments.
The Paradigm Shift: Permission to Consume
The most sobering revelation lies buried in PJM’s counterfactual simulations. Had regulators removed the price collar, the auction would have cleared at $554.72/MW-day grid-wide—and $776.69 in Chicago’s ComEd zone—swelling total auction cost from $16.4 billion to $29.7 billion.
Yet an extra $13.3 billion in ratepayer costs would have coaxed exactly 77 additional megawatts of capacity out of the market.
The near-term supply curve is vertical. Removing price caps does not manufacture physical generators; it merely transfers wealth. This exposes the core fallacy of the AI infrastructure boom: generation capacity is no longer the binding constraint. The true scarce asset is time-certain, locationally deliverable power coupled with contractual permission to consume it.
The public auction at $325 is now a socialized reference index for ratepayers, while bankable scarcity rents ($555 contracts and bilateral agreements) migrate to long-duration private deals. For capital allocators, brownfield sites retaining transferable Capacity Interconnection Rights and existing turbine slots are worth more than any greenfield acreage. You cannot run a foundational AI model without electrons, but more critically, you cannot buy your way through a five-year substation bottleneck with a one-year market price.
not investment advice