Six alternative branches. These are subjective probabilities from the seed, not observed frequencies or investment recommendations.
Why this trend
It is the one trend in the sector that appears on three of the dossier's five lists, and the dossier records that recurrence as a finding rather than padding: capacity market design is simultaneously the most important current dynamic, the largest business cost shock and the main consumer bill driver from June 2028. It is also the sector's most precisely resolvable object — PJM publishes a single number on a fixed date and MISO provides a control case in the same year.
Load-bearing assumption
The reliability requirement PJM procures against is built on a load forecast that materialises within roughly half of its projected growth.
Preconditions
- Capability
clear — capacity is metered and auctions clear on a published schedule
- Economics
clear — utility capex is rate-based; the 2028/29 delivery year cost $16.4bn against $2.2bn for 2024/25
- Supply
fails — turbines committed through 2030, GSU transformers at 144 weeks, switchgear at 44 weeks
- Demand
partial — real at the core, speculative at the margin; Exelon discloses 22% expected materialisation of its 65 GW pipeline
- Permission
fails — interconnection is the binding gate at a five-year median, and large-load rules are unwritten
- Capital
clear — roughly $1.4T of announced US investor-owned utility capex through 2030
Cap-clearing persists, bills arrive 2028
PJM continues to clear at or near the administrative cap through the 2029/2030 auction because new entry cannot arrive faster than turbine slots and transformer lead times allow. Capacity costs flow into retail bills from June 2028 and residential prices keep rising 3-5% a year, above inflation. The decelerating element is the rate of load-forecast escalation: PJM trims near-term vintages while holding its long-run 3.6% growth rate, so the requirement grows more slowly than in 2026 without falling.
Mechanism
Only 525 MW of new resources cleared in the 2028/29 auction, of which 208 MW were uprates. New entry is metered by three turbine OEMs with output committed through 2030 and by 128-144 week transformer lead times, not by price. The price collar caps the signal, so the shortfall persists rather than being cleared by price. Demand keeps growing at the metered core even as the speculative margin of the queue is vetted out, which keeps the requirement above supply.
Preconditions, early indicators and assumptions
- Calibration Basis
base_rate
- Preconditions
- Turbine and transformer lead times not shortening materially before 2028
- The price collar remaining in place at or above $325/MW-day
- Load forecasts trimming at the margin rather than falling
- Early Indicators
- Indicator
PJM 2029/2030 Base Residual Auction clearing at or above $325/MW-day (forecast F-05-02)
- Source
S-05-05 PJM auction results
- Would Be Visible By
2027-12
- Indicator
PJM's 2027 Load Forecast Report trimming the near term while holding the long-run growth rate (forecast F-05-01)
- Source
S-05-05 PJM Load Forecast Report, annually each January
- Would Be Visible By
2027-02
- Indicator
EIA STEO showing a 2027 US average residential price at or above 19.0 cents/kWh (forecast F-05-03)
- Source
S-05-01 EIA STEO, monthly
- Would Be Visible By
2027-09
- Affected Industries
- 05
- 01
- 03
- 18
- 09
- 12
- What Businesses Should Do
A 10 MW industrial load in PJM should budget a capacity charge rising from roughly $6,000 a month in 2024 to roughly $70,000 in 2028. Hedge or contract now, and check whether your retail supplier passes capacity through at auction or at a lagged average — the difference is the whole exposure.
- Precedence Note
Distinguished from downside by the direction of the clearing price, not by the load forecast: base is clearing at or near the cap; downside is clearing below it.
- Would Change Our Mind
A PJM auction clearing more than 4 GW of new resources, which would show entry responding to price.
- Assumptions
- Text
New entry stays metered by equipment lead times rather than by price, so the collar cannot clear the shortfall.
- Confidence
high
- Load Bearing
true
- Basis
T-05-06 — power transformers at 128 weeks, GSUs at 144 weeks, switchgear at 44 weeks; T-05-03 — turbine slots committed through 2030
- If Wrong
Entry responds, prices fall as they did in MISO, and mass moves to the upside branch.
- Text
PJM's reliability requirement does not fall materially as the queue is vetted.
- Confidence
medium
- Load Bearing
false
- Basis
S-05-05; PJM cut 2027 by ~4 GW and 2028 by 4.4 GW in the 2026 vintage
- If Wrong
The shortfall narrows without any supply response, which looks like downside arriving through the demand side.
Entry responds, MISO-style
Supply arrives faster than the equipment constraint implies, as it did in MISO where annualised capacity prices fell 42% to a $116-126/MW-day range in the same year PJM's hit the cap. Uprates, restarts, storage and contracted flexibility clear in volume; the PJM auction clears below the cap with the reliability requirement met. Prices stabilise rather than continuing to rise, which removes the affordability politics before it constrains anything.
Mechanism
Roughly 100 GW of curtailment-enabled headroom on the existing grid, identified by Duke's Nicholas Institute and built into FERC's flexible-load proposals, is contracted rather than merely studied. Storage, compounding at roughly 70% a year and reaching about 52 GW by mid-2026, clears as capacity. Nuclear uprates and restarts add firm megawatts without new interconnection. Because none of these requires a large-frame gas turbine, the OEM constraint stops being the binding one.
Preconditions, early indicators and assumptions
- Calibration Basis
analogue
- Preconditions
- An RTO tariff with an operational flexible-load service class and contracted volume, not announcements
- More new capacity clearing than in the prior auction
- Storage and uprates qualifying as capacity resources at scale
- Early Indicators
- Indicator
The first RTO tariff with an operational flexible-load service class and at least 1 GW contracted under it — contracted megawatts, not announcements
- Source
S-05-04 FERC eLibrary and S-05-05 PJM
- Would Be Visible By
2027-12
- Indicator
EIA Form EIA-860M planned capacity additions in PJM states rising materially year on year
- Source
S-05-02 EIA Electric Power Monthly, monthly
- Would Be Visible By
2027-06
- Indicator
Palisades reporting non-zero net generation in EIA data (forecast F-05-07)
- Source
S-05-02 EIA EPM generation by plant, monthly
- Would Be Visible By
2027-10
- Affected Industries
- 05
- 01
- 03
- 18
- 20
- What Businesses Should Do
If you are a large load, the flexible-load pathway is the cheapest speed-to-power option available — negotiate curtailment terms now, before the tariff hardens. If you own existing firm capacity, this is the branch in which your scarcity rent disappears.
- Precedence Note
Requires contracted megawatts under an operational tariff. Announcements and pilots leave the outcome in base.
- Would Change Our Mind
Two RTOs filing flexible-load tariffs with zero contracted volume twelve months later.
- Assumptions
- Text
Large loads will accept curtailment in a contract, not only in a pilot.
- Confidence
low
- Load Bearing
true
- Basis
T-05-10; the ~100 GW headroom estimate is an academic study that shaped a rule-making, not a contracted quantity
- If Wrong
Upside collapses into base; nothing moves to downside on this assumption alone.
- Text
MISO's 42% price decline reflects a supply response that is available in PJM too.
- Confidence
low
- Load Bearing
false
- Basis
T-05-04 evidence, 2026-04-29; MISO had 141 GW offered
- If Wrong
The control case is not comparable and the analogue misleads.
Demand deflation clears the market below cap
AI capex retrenches, queue withdrawals accelerate, and the PJM auction clears below the administrative cap. Utilities are left holding committed capex and minimum-take contracts against load that does not arrive; the 85% minimum-take tariffs become litigation rather than revenue. This is not the thesis failing — capacity is still short in places — it is the thesis delivering less than the forecast that justified the build.
Mechanism
The speculative component of the queue withdraws: duplicate and optioned requests, which ERCOT's 474 GW against a historical peak of 85.5 GW shows can be several multiples of real demand. Hyperscaler capex guidance is cut rather than narrowed, which removes the marginal large load. The reliability requirement falls as vetting tightens, so the auction clears below cap without any new supply. Utilities that contracted equipment against the 2026 forecast carry it.
Preconditions, early indicators and assumptions
- Calibration Basis
base_rate
- Preconditions
- Two consecutive quarters of declining large-load queue volume in ERCOT after Batch Zero resumes
- A PJM auction clearing below the administrative cap
- A large-load minimum-take dispute reaching a state commission docket
- Early Indicators
- Indicator
Two consecutive quarters of declining large-load queue volume in ERCOT after Batch Zero resumes, combined with a PJM auction clearing below cap
- Source
S-05-06 ERCOT weekly board materials and S-05-05 PJM auction results
- Would Be Visible By
2028-06
- Indicator
A large-load minimum-take dispute appearing on a PUCT or PUCO docket
- Source
S-05-21 PUCT Interchange and PUCO docketing, daily
- Would Be Visible By
2027-12
- Indicator
A hyperscaler capex guidance cut rather than a narrowing (forecast F-01-01 monitors the same series)
- Source
S-05-17 EEI and S-01-03 hyperscaler IR, quarterly
- Would Be Visible By
2027-02
- Affected Industries
- 05
- 01
- 03
- 09
- 18
- 11
- What Businesses Should Do
If you are a utility or a developer, do not contract equipment against the unvetted queue. If you are a large load facing an 85% minimum-take tariff, price the exit fee — in this branch it becomes the negotiation, and in litigation it becomes the whole argument.
- Precedence Note
Requires both legs — queue decline and a sub-cap clear. One leg alone stays in base.
- Would Change Our Mind
Queue volume declining while metered commercial consumption keeps accelerating, which would show the queue is measuring developer optioning rather than demand.
- Assumptions
- Text
Queue volume is a leading indicator of delivered load rather than a lagging indicator of developer optioning behaviour.
- Confidence
medium
- Load Bearing
true
- Basis
T-05-05 — only 13% of capacity requested 2000-2020 reached operation and 75% was withdrawn; no like-for-like duplication analysis of current large-load queues has been published
- If Wrong
The downside indicator fires without the downside occurring — a false positive this register would have to publish and score.
- Text
No metered data-centre-only load series exists, so the deflation is observable only through queues and commercial-sector totals.
- Confidence
high
- Load Bearing
false
- Basis
Dossier 05 §13 item 1 — the largest single data gap in the sector
- If Wrong
If an operator begins publishing metered large-load data, this branch becomes measurable rather than inferred.
Firmness is redefined and procurement follows
Enhanced geothermal, long-duration storage and contracted flexibility displace the gas peaker as the marginal firm resource, and capacity procurement follows: the auction starts clearing against a resource mix in which the turbine backlog is a liability rather than a moat. Google's 396 MW Fervo contract and the eight banks underwriting Fervo's $421m Cape Station facility are the existing evidence that lenders, not just buyers, have moved.
Mechanism
Fervo delivers Cape Station Phase 1 on schedule and a second developer reaches non-recourse project finance for enhanced geothermal — one company is a project, two is a sector. The marginal firm megawatt stops being a gas turbine, so the ~116 GW of slot reservations converts poorly and the scarcity rent that supports cap-clearing prices erodes from the supply side rather than the demand side. Storage, already compounding at roughly 70% a year, takes the peaking duty.
Preconditions, early indicators and assumptions
- Calibration Basis
analogue
- Preconditions
- Cape Station Phase 1 delivered on schedule
- A second developer reaching non-recourse enhanced-geothermal project finance with named lenders
- Turbine backlogs converting poorly
- Early Indicators
- Indicator
A second enhanced-geothermal developer reaching non-recourse project finance with named lenders — named lenders are the verification signal, not capacity claims
- Source
S-05-25 developer and OEM press and IR, irregular
- Would Be Visible By
2028-06
- Indicator
GE Vernova declining to disclose a slot-to-order conversion rate while backlog quality commentary changes (forecast F-05-04)
- Source
S-05-12 GE Vernova IR, quarterly
- Would Be Visible By
2027-12
- Indicator
EIA Form EIA-860M showing geothermal and storage additions displacing planned gas in PJM and ERCOT
- Source
S-05-02 EIA Electric Power Monthly, monthly
- Would Be Visible By
2028-06
- Affected Industries
- 05
- 20
- 09
- 01
- What Businesses Should Do
Track project-finance closes and named lenders, not capacity announcements. If you own turbine slots, the value is in the slot's optionality, not in the machine — and that optionality decays if a second firm-clean path becomes bankable.
- Precedence Note
Requires the second developer. A single delivered project resolves to base.
- Would Change Our Mind
Cape Station Phase 1 slipping more than two quarters with no second developer reaching financial close.
- Assumptions
- Text
Enhanced-geothermal cost declines observed at one site generalise to a second geology.
- Confidence
low
- Load Bearing
true
- Basis
T-05-12; the corpus's rule that one company is a project and two is a sector comes from this branch
- If Wrong
The branch stays a project and its mass returns to base.
- Text
Capacity market rules accredit new firm-clean resources at values that let them clear.
- Confidence
low
- Load Bearing
false
- Basis
Accreditation methodology is an RTO design choice, currently unsettled for flexible and long-duration resources
- If Wrong
The resources exist but do not clear, so the price effect does not appear.
The collar is rewritten
The price collar is the regulatory object, and it is contested from both sides. Either FERC or a settlement raises or removes the cap — in which case the 2028/29 counterfactual of roughly $555/MW-day region-wide and $777/MW-day in ComEd becomes the actual, at $29.7bn instead of $16.4bn — or states and the Commission push the other way and suppress it. Both are regulatory branches; the direction is genuinely two-sided and the instrument is the same.
Mechanism
PJM, its market monitor, state commissions and load-serving entities litigate the collar in FERC dockets. A shortfall against the reliability requirement is the strongest argument for removing a cap; a $16.4bn annual cost arriving on bills in June 2028 is the strongest argument for keeping or tightening one. Whichever way it resolves, the clearing price stops being a market outcome and becomes an administrative one, and the FERC RM26-4 large-load framework determines who pays.
Preconditions, early indicators and assumptions
- Calibration Basis
base_rate
- Preconditions
- An active FERC docket on the PJM capacity price collar or the variable resource requirement curve
- State commissions or legislatures intervening on capacity cost allocation
- The June 2028 bill impact remaining scheduled rather than deferred
- Early Indicators
- Indicator
A FERC final rule in Docket RM26-4 (forecast F-05-05)
- Source
S-05-04 FERC eLibrary, daily
- Would Be Visible By
2027-12
- Indicator
A FERC order changing the PJM price collar or the variable resource requirement curve
- Source
S-05-04 FERC eLibrary, daily
- Would Be Visible By
2028-06
- Indicator
Two or more RTOs with FERC-accepted large-load class tariffs containing cost-allocation provisions (forecast F-01-04)
- Source
S-05-04 FERC eLibrary, daily
- Would Be Visible By
2027-06
- Affected Industries
- 05
- 01
- 03
- 18
- 12
- What Businesses Should Do
Read the collar dockets, not the auction results — the auction reports an administrative outcome. Large loads should assume cost allocation lands on them in some form, and price the 85% minimum-take and exit-fee terms into siting decisions now.
- Precedence Note
Regulatory outranks base when an order issues, even if the clearing price looks base-like. Litigation without an order stays in base.
- Would Change Our Mind
The collar surviving two more auction cycles without a substantive FERC order on it.
- Assumptions
- Text
FERC acts on the collar within the horizon rather than leaving it to expire on its own terms.
- Confidence
medium
- Load Bearing
true
- Basis
S-05-04; the collar is a settlement construct and FERC has an open large-load docket in RM26-4
- If Wrong
The collar persists unchanged and mass returns to base, with the bill impact arriving on schedule in June 2028.
- Text
The $555/MW-day and $777/MW-day counterfactuals are approximately right.
- Confidence
medium
- Load Bearing
false
- Basis
T-05-04 evidence, 2026-07-15, recorded as an estimate rather than a fact
- If Wrong
The magnitude of the collar's effect is misstated, though its existence is not.
Affordability politics breaks the construct
Capacity costs reaching retail bills from June 2028, on top of residential prices already rising above inflation, produce legislative intervention that suspends or replaces the capacity market rather than adjusting it. The construct that has allocated reliability investment in PJM for two decades stops functioning as an investment signal, and the states revert to bilateral procurement or integrated resource planning. Investment decisions made against auction revenue are stranded.
Mechanism
A $16.4bn delivery-year cost lands on bills in an election cycle, in states where data-centre cost allocation is already contested. Legislatures act crudely, because affordability politics does not distinguish between capacity, transmission and fuel. The UK's VAT removal on electricity is the live template for a fast political fix; the US precedent is state re-regulation after the 2000-2002 merchant collapse. Merchant projects underwritten at 2024 rates against 2026 capacity prices lose the revenue line they were financed on.
Preconditions, early indicators and assumptions
- Calibration Basis
analogue
- Preconditions
- Capacity costs arriving on retail bills as scheduled from June 2028
- Residential prices continuing to rise faster than inflation into an election
- At least two state legislatures opening capacity-market or data-centre cost-allocation bills
- Early Indicators
- Indicator
A state legislature in a PJM state enacting a capacity cost cap, a bill credit funded from capacity revenue, or a market exit study
- Source
S-05-21 PUCT and PUCO dockets and state legislative records, daily
- Would Be Visible By
2028-12
- Indicator
EIA reporting US average residential electricity prices above 19.0 cents/kWh while capacity costs are still to arrive (forecast F-05-03)
- Source
S-05-01 EIA STEO, monthly
- Would Be Visible By
2027-09
- Indicator
PJM's market monitor attributing a larger share of wholesale cost increase to data-centre load than the 9% / $10.48 per MWh already published
- Source
S-05-07 Monitoring Analytics, quarterly
- Would Be Visible By
2028-03
- Affected Industries
- 05
- 01
- 12
- 18
- 20
- What Businesses Should Do
If your investment case rests on PJM capacity revenue past 2029, model a scenario in which that revenue is administratively reduced. Political risk in this sector now moves faster than project risk — stop-work orders were issued and judicially reversed within six weeks, and 202(c) orders were issued and vacated within sixteen months.
- Precedence Note
Failure outranks all other branches on occurrence. Regulatory becomes failure at the point where the construct is suspended or replaced rather than adjusted.
- Would Change Our Mind
The June 2028 bill impact arriving with no legislative response in any PJM state within two quarters.
- Assumptions
- Text
Electricity affordability becomes a binding political constraint before it becomes a binding economic one.
- Confidence
low
- Load Bearing
true
- Basis
T-05-14; the corpus notes the 2026 electricity-affordability electoral politics was under-researched because the search budget was exhausted
- If Wrong
Bills rise, voters absorb it, and the construct survives — mass returns to base and regulatory.
- Text
Legislatures act on capacity markets specifically rather than on retail rates generally.
- Confidence
low
- Load Bearing
false
- Basis
The UK VAT removal template acted on the retail bill, not on the wholesale construct
- If Wrong
The intervention lands on retail rates and the capacity market survives with a different payer.
Additional scenario notes
- Probabilities Sum
1
- What Must Be True To Grow
- Load materialises at roughly the vetted rather than the requested rate — Exelon discloses 22% for its own pipeline
- Equipment lead times do not shorten enough for new entry to clear the shortfall before 2029
- The price collar stays in place at a level that keeps cap-clearing meaningful
- Cost allocation lands in a way that survives state political scrutiny when bills rise from June 2028
- What Could Stop It
- Demand deflation — the queue contains optioned and duplicate requests against a 13% historical completion rate
- Political licence — affordability becoming unacceptable before the build becomes uneconomic
- Physical constraint working the other way — manufacturing expansions landing 2027-2029 and creating a 2029-2031 equipment glut
- Regulatory reversal — FERC or a settlement rewriting the collar in either direction
- Measurement revision — NERC's ten-year peak figure moved 69% in one year, and CBRE and JLL differ by roughly 9x on data-centre capacity under construction
- Uncertain Assumptions
- That the PJM reliability requirement is built on a load forecast that materialises — no metered data-centre-only load series is published by any US system operator
- That the $555/MW-day and $777/MW-day no-collar counterfactuals are approximately right — both are estimates, not auction outcomes
- That turbine slot reservations convert to firm orders at a rate that constrains entry — GE Vernova holds that number and does not publish it (forecast F-05-04)
- That PJM's unpublished large-load vetting criteria are applied consistently between forecast vintages
- Authored
2026-09-15