Autoresearch: Does PJM's capacity-model correction threaten the nuclear capacity-price premium?

Generated by /autoresearch on 2026-08-17. Synthesized across 2 rounds (early exit — the cap-vs-uncapped spread settled the question) from 2 web pages, plus the same-day SemiAnalysis feed clipping as the prompting source. No Grokipedia anchor (live market-design question). Treat as raw material. Context: vault/projects/stock-market

Priors capture skipped — headless run. The standing prior being tested is the project's own pjm-capacity-prices-to-nuclear-premium thesis (conviction med-high, CEG).

Summary

Today's SemiAnalysis post argues PJM's capacity clearing prices are inflated by a modeling error worth ~$11.6B to ratepayers across two auctions, and that a corrected model would cut the capacity requirement by ~3.8 GW for 2026/27 (and 3,824 MW for 2028/29). Taken at face value that reads like a falsifier for any thesis that treats high PJM capacity prices as durable — including this project's nuclear-premium chain.

On the numbers, it isn't one — not for the delivery years already cleared. PJM's 2028/2029 Base Residual Auction cleared at the FERC-approved $325/MW-day cap across the entire footprint, and PJM's own estimate is that without the collar the auction would have cleared $554.72/MW-day, taking total cost from $16.4B to nearly $30B (Power Engineering). That is a ~$230/MW-day gap between the uncapped price and the cap. A requirement correction of ~3.8 GW against a ~138 GW procurement moves the demand curve, but it has to eat through that entire gap before the realized clearing price — the number that shows up in a generator's revenue — moves at all. The modeling error's incidence, as the article itself frames it, is on quantity over-procured at a capped price, not on the price the marginal nuclear MW receives.

The thesis is therefore refined, not falsified, with the risk relocated to a specific date: the collar covers 2028/2029 and 2029/2030 only. If the model is corrected and the collar lapses together, the two effects compound rather than offset.

Findings

The cap is binding, and it binds against generators

FERC approved the price collar on April 28, 2026 — a cap of approximately $325/MW-day and a floor of $175/MW-day — applying it to the 2028/2029 and 2029/2030 delivery years (auctions closing July 7 and December 15 respectively), having previously applied it to 2026/2027 and 2027/2028. PJM proposed the extension, "supported by all 13 governors, the White House National Energy Dominance Council, and the U.S. Department of Energy." FERC's stated rationale: "these circumstances support the continued application of a time-limited price collar to narrow the range of potential price outcomes, which will reasonably address concerns about excessive price volatility" (PJM Inside Lines).

The 2028/2029 BRA then cleared at the ceiling, $325/MW-day, in every zone — the third consecutive auction to hit the cap (Power Engineering). For a nuclear generator, a binding cap means realized capacity revenue is below the market-clearing level, not above it.

Constellation's exposure, quantified

Constellation cleared 18,875 MW in the 2028/2029 auction at $325/MW-day, of which 15,700 MW is nuclear (StockTitan / CEG 8-K). That is the concrete revenue line the thesis rides, and it is struck at the capped price.

Supply is not responding, which is the part of the thesis that holds

The 2028/2029 auction procured 138,318 MW total (base residual plus FRR regions) with a mix of "46% natural gas, 20% nuclear, 18% coal, 5% demand response, 4% hydro, 2% wind, 2% oil, and 1% solar" — and cleared only 525 MW of new generation and uprates (Power Engineering). Three consecutive auctions at the cap with a rounding-error of new entry is the supply-side fact the nuclear-premium thesis actually depends on, and it is independent of the winter-rating dispute. It corroborates the interconnection wall described in the same-day SemiAnalysis piece (220 GW applied when the window reopened in April 2026; the Reliability Resource Initiative's first output is scheduled for 2030; "neither has energized a single megawatt").

Where the modeling critique does bite

Two places, both real and both dated:

1. After 2029/2030, when the collar lapses. If PJM has by then adopted seasonal/winter ratings (its Independent Market Monitor recommends them, and PJM has retained E3 to review the methodology, per the SemiAnalysis piece), the uncapped price would form against a lower requirement. The $554.72 uncapped estimate is the pre-correction number; the post-correction uncapped number is unknown and could be materially lower.
2. The political economy. Retail bills "jumped 17–24%" in June 2025 and the collar exists because 13 governors demanded it. A credible published claim that ~$11.6B of the cost was a modeling artifact is exactly the input that makes the collar permanent, tightens the cap, or forces a requirement re-study — a policy risk to the price, distinct from a supply/demand one.

Contradictions and open questions

• The two sources disagree in emphasis, not in fact. SemiAnalysis reports the 2026/27 auction clearing at $329/MW-day and characterises current prices as ~9–11× the $28.92 pre-2024 baseline; PJM and the trade press report $325 (2028/29) and $333.44 (2027/28) as cap outcomes with a much higher uncapped shadow price. Both are true; they describe different counterfactuals. A thesis that cites "record capacity prices" without saying whether it means the capped or the uncapped price is ambiguous, and this project's chain currently does not distinguish them.
• Unquantified: what the uncapped clearing price would be under a corrected model. Until someone runs that, the post-2030 risk to the premium can't be sized. SemiAnalysis's public portion gives the requirement deltas (−3,824 MW for 2028/29) but the price reconstruction is in the paywalled methodology annex.
• Timing of the E3 review and any FERC filing on seasonal ratings is not established by these sources — the nearest dated item is the emergency-auction comment deadline of 21 August 2026, which is four days out and worth watching as the venue where the modeling critique gets aired.
• Does a lower requirement even reduce nuclear's cleared MW? Nuclear is inframarginal — it clears regardless. The correction affects the price, not whether CEG's 15,700 MW clears. This argues the exposure is to price only, but no source addresses whether a corrected winter rating would displace nuclear UCAP accreditation specifically.

Provenance

Rounds run: 2 of 3 (early exit — the cap-vs-uncapped spread answered the question; a third round would only have chased the paywalled price reconstruction).

Sub-questions by round:

Round 1 (broad survey):

1. What is the status of PJM's reserve-requirement study / E3 review and any winter-ratings reform?
2. What did the most recent PJM capacity auction clear at, and what is Constellation's cleared position?

Round 2 (drill-down):

1. What exactly does the FERC price collar cap and floor, for which years, and who backed it? — targeted whether a requirement correction can reach the realized price
2. How much new supply cleared, and what would the price have been uncapped? — targeted the size of the buffer between a corrected requirement and the cap

Anchor source: none (live market-design question; no encyclopedic entry).

URLs fetched (2 successful, 0 failed):

Round 1:

• PJM capacity auction easily hits price cap… Again — Power Engineering — trade press — the $325 clear, the $554.72 uncapped estimate, $16.4B vs ~$30B, 138,318 MW, supply mix, 525 MW of new entry.

Round 2:

• FERC Approves Capacity Auction Price Collar to Next Two Capacity Auctions — PJM Inside Lines — grid operator (primary) — collar levels, delivery years, April 28 2026 approval, backers, FERC rationale.

Cited from search results without a separate fetch (figures appear in the source's own SEC-filing summary):

• Constellation Energy 8-K — 18,875 MW cleared, 15,700 MW nuclear

Also drawn on: the same-day feed clipping of SemiAnalysis, "$12B of US ratepayers' money wasted on a modeling mistake and PJM wants to do it again" (2026-08-16), which prompted this pass.

Source-reliability note: insidelines.pjm.com and www.power-eng.com both fetched cleanly (Reliable). www.pjm.com primary documents are PDFs on a non-whitelisted (non-.gov) domain and were not fetched, per the skill's PDF rule — figures above come from the HTML trade-press and operator-blog renderings instead.

Tools used: WebSearch, WebFetch. Generated: 2026-08-17