2026 08 16 Feed Semianalysis 12B OF US Ratepayers Money Wasted ON A Modeling Mistake
PJM's 2024 Reserve Requirement Study understates winter gas capacity (cold-air uplift + post-Elliott winterization), inflating the capacity demand curve and overcharging ratepayers ~$11.6B across two auctions — which means the record PJM capacity clearing prices underwriting the nuclear-premium thesis are partly a modeling artifact, and PJM has hired E3 to review the methodology.
view source ↗Summary
The load-bearing causal claim: PJM's record capacity-auction clearing prices are inflated by a model error, not purely by scarcity. PJM's 2024 Reserve Requirement Study rates gas plants at summer capability year-round — ignoring that cold dense air lifts gas-turbine output 8.4–25% in winter — and still embeds Storm Elliott (Dec 2022) failure rates even though ~400 of PJM's ~700 gas units have since winterized. Because PJM is "operating at the limit," a small understatement of available capacity shifts the Variable Resource Requirement demand curve right and lifts the clearing price for the entire fleet (~93% of capacity settles at the clearing price). The article computes the resulting ratepayer overcharge at ~$11.6B across the 2025/26 and 2026/27 auctions (sensitivity $8.0–$14.5B), and argues PJM is about to repeat it with a 6.8 GW "emergency" backstop auction that has no committed counterparties.
For this project this cuts two ways and both matter: it is a falsifier-grade challenge to any thesis that treats PJM capacity prices as a durable structural signal (the price is ~9–11× the pre-2024 baseline and partly artifactual, with PJM's own consultant E3 now reviewing the methodology), and it is confirmatory on the interconnection bottleneck — a 23-month lead time makes new generation near-impossible to build, and datacenter developers are routing around PJM behind-the-meter.
⚠
partial: true. The public body is complete for the analysis above; the methodology annex (the reverse-engineered model reconstruction) is behind "This post is for paid subscribers." Figures below are from the free portion.
Article
The modeling mistake
Two factors PJM's Reserve Requirement Study (adopted 2024) fails to credit:
- Cold-air uplift. Gas turbines operate 8.4–25% more efficiently in winter because cold, dense air improves mass flow. PJM models plants at summer ratings year-round.
- Asset winterization. By late 2025, ~400 of PJM's ~700 gas units had invested in freeze protection after Winter Storm Elliott (Dec 2022). PJM's historical failure rates still embed that 2022 event without adjustment.
The transmission mechanism: "PJM has forced itself into operating at the limit, so inaccurately modeling power plants' capacities has a massive impact on auction costs." The capacity auction clears against a demand curve (the Variable Resource Requirement); overstating perceived shortage shifts that curve right and forces prices up across the whole stack, and roughly 93% of capacity settles at the clearing price — a windfall to existing generators.
The numbers
| Metric | Value |
|---|---|
| Capacity procured 2025/26 | 135.7 GW, cleared $270/MW-day |
| Capacity procured 2026/27 | 134.2 GW, cleared $329/MW-day (political price cap) |
| Pre-2024-model-change auction price | $28.92/MW-day |
| Current price range | $270–$333/MW-day (~9–11× the historical baseline) |
| Total spend, four auctions 2025/26–2028/29 | $63.6B |
| Modeled excess capacity (cold + winterization, 2026/27) | 3.8 GW (~8 large gas plants; ~$10B of avoided new-build) |
| PJM system peak | ~155 GW summer; lower in winter |
| PJM population served | 66 million across 13 states + DC |
| New capacity procured across four auctions | 4.8 GW vs a 134–138 GW annual requirement |
Overcharge attribution: the 2025/26 auction over-procured 0.014 GW and would have saved "$6.7B with only…14MW less power"; the 2026/27 auction over-procured 0.8 GW and would have saved "$4.9 billion." Combined ~$11.6B.
Capacity-requirement reduction if modeled correctly: 2025/26 −2,879 MW; 2026/27 −3,396 MW; 2027/28 −4,513 MW; 2028/29 −3,824 MW.
What PJM wants to do again — the emergency auction, with dates
PJM plans to "sign contracts…that are supposed to be paid for by new large loads" but has "no committed counter-parties." Timeline as filed:
- 31 July 2026 — filing
- 21 August 2026 — comment deadline to FERC
- 30 Sept – 21 Oct 2026 — auction window
- 2 December 2026 — results announced
- 9 December 2026 — next regular auction opens
- Contract duration through 2043 (~15 years)
Target 6.8 GW at a $555/MW-day average cap — maximum liability ~$21B. Under a corrected model the target would be ~3.0 GW, a 56% reduction. The PJM Board "rejected the subscription framework" in favour of a non-binding model, overriding members who had endorsed a voluntary prepayment structure. The risk the article names: "If no other counter-party emerges, then once again those left holding the bag will be the residential ratepayers."
Datacenter load, behind-the-meter, and the interconnection wall
- Datacenter demand is the primary driver of 2025–2027 capacity growth.
- "Many datacenter developers have written off PJM regardless, going elsewhere or planning behind-the-meter power configurations."
- The emergency auction is characterised as "an emergency mechanism hoping for datacenters to pay for their own capacity."
- Interconnection: "From October 2021 there was no study path while PJM cleared its backlog; a general application window did not reopen until April 2026, when 220 GW applied."
- Reliability Resource Initiative: 51 shovel-ready projects selected May 2025, now 41 (31.5% withdrawn); first output scheduled 2030. The Expedited Track (10 units/year) opened July 2026. "Neither has energized a single megawatt."
- Lead time for new plants was originally 36 months per the Independent Market Monitor; now 10–23 months. "Power plants take time to build and need long term visibility so giving only a 23-month lead time makes it near impossible to build new generation."
Market design
"PJM is also the only market in the world where the capacity auction price is for the whole generation stack…Japan was the last other market…has now concluded the design builds nothing and bolted on a separate 20-year auction for new entries."
Who gains and who loses
The article names no generator, utility, or asset owner by company name. It references the Data Center Coalition and the Electric Distribution Companies (as stakeholders backing the subscription model), PJM's Independent Market Monitor (recommending seasonal ratings), E3 (the consultant PJM hired to review the methodology), and Great Britain / Japan as comparison market designs.
Sectoral incidence as stated:
- Existing generators received "$325/MW-day" for doing nothing; the median existing combined-cycle unit "made 407% of its going-forward costs in the energy and ancillary services markets alone in 2025, before a dollar of capacity revenue."
- New generators / developers face 23-month lead times, cannot secure long-term contracts, and must bid $100–$350/MW-day to cover rapid-deployment risk.
- Residential and small-business ratepayers absorb the overcharge; bills "jumped 17–24%" in June 2025.
- Large datacenter loads can opt out via self-supply, direct contracting, or behind-the-meter, and face 50+ MW curtailment penalties (IRAS scheme) if they do not self-supply.