Study · Power & AI Infrastructure · August 2026
Inside PJM's Capacity Crunch
TLN, VST, and VRT all reported earnings within three days of each other this month, and all three reports pointed at the same underlying number: PJM Interconnection's capacity auction — the market that pays power plants to simply be available — cleared at $329.17 per megawatt-day for the 2026/27 delivery year, the maximum allowed under FERC's own price cap. Two years earlier, the same auction cleared at $28.92. That's not a typo. This is a look at the mechanics behind an 11x move in a regulated capacity market, and the three positions in my book — 6.54% of the total — that are directly exposed to it.
1. The auction, in three numbers
PJM runs a forward capacity auction called the Reliability Pricing Model: generators bid to be paid for having capacity available a delivery year in advance, and the market-clearing price is set where supply meets forecasted peak demand plus a reserve margin. It's meant to be a slow-moving, boring number. It has not been boring.
| Delivery year | Clearing price | Note |
|---|---|---|
| 2024/25 | $28.92/MW-day | Baseline, pre-AI-load-growth era |
| 2025/26 | $269.92/MW-day | +833% year over year |
| 2026/27 | $329.17/MW-day | Hit the FERC-approved price cap — the auction wanted to clear even higher |
2. Why: load growth PJM didn't see coming fast enough
PJM's own 2026 long-term load forecast projects summer peak demand rising to 253 GW by 2046 from 160 GW in 2025 — a 58% increase, driven primarily by data centers. The near-term number is the one that actually explains this auction: PJM's large-load additions (principally data centers) are projected at 35.1 GW between 2026 and 2031, against total system demand growth of only 34.6 GW over the same period — meaning data-center load accounts for effectively all of PJM's net demand growth, with every other demand category roughly flat to declining. DOM (Dominion, serving Northern Virginia's Data Center Alley), AEP, ComEd, and PPL together account for 74% of PJM's total annual demand growth.
On the supply side, the interconnection queue is responding, just slowly. 811 new generation projects totaling 220 GW have applied to connect under PJM's reformed interconnection process, and a Wood Mackenzie analysis puts utility-forecasted large-load growth within PJM's footprint at 55 GW by 2030 and 100 GW by 2037. But new generation — especially anything beyond gas peakers — takes years to permit, finance, and build. The capacity auction is pricing that lag directly: it is, structurally, a real-time bet that demand will keep outrunning new supply for the next several delivery years.
3. Three holdings, three layers of the same bet
This isn't an abstract macro story for my book — all three positions below reported earnings within the same week in early August, and each report cited PJM capacity pricing directly as a driver.
Talen Energy — 3.14% of book
Nuclear and gas generation serving PJM directly. Q2 2026 adjusted EBITDA of $374M, with guidance raised on the back of "PJM capacity auctions clearing at price caps" cited explicitly in the company's own release. See the full Q2 2026 report.
Vistra — 0.86% of book
Multi-region generation with PJM exposure alongside ERCOT. Generation-segment EBITDA grew 68% YoY in Q2 2026, citing "higher PJM capacity revenues" as one of several direct drivers. See the full Q2 2026 report.
Vertiv — 2.55% of book
Not a power generator — sells the cooling and power-delivery infrastructure that sits inside the data centers driving this entire demand curve. A picks-and-shovels way to own the same buildout without direct commodity-power exposure. See the full Q2 2026 report.
TLN and VST monetize the same capacity crunch directly, through the auction price itself. VRT monetizes the buildout that's causing the crunch, one layer removed. All three go up if AI-driven data-center demand keeps outrunning new generation and grid capacity — and all three would likely reprice together, not independently, if that demand growth decelerates.
4. The bull case and the bear case, stated honestly
This is a multi-year structural gap, not a one-time spike
New generation capacity takes years to permit and build, while data-center load is arriving now. Even with 220 GW in PJM's interconnection queue, only a fraction will actually get built on a timeline fast enough to close the current gap — meaning elevated capacity prices could persist for several more delivery-year cycles, not just this one.
A price this high is its own correction mechanism
$329/MW-day at the regulatory cap is an extraordinarily strong price signal — strong enough to pull forward exactly the new generation investment (including from competitors) that eventually closes the gap. Prices this elevated rarely persist for long once the market has this much incentive to add supply; PJM's own reforms to speed interconnection are explicitly designed to accelerate that response.
The math already shows early cracks in the clean story
TLN's own Q2 2026 report flagged a widening PPL zonal basis discount (~$20/MWh) eating into realized power prices even as headline capacity pricing stayed strong — a real, name-specific cost showing up inside the tailwind. The bull and bear cases aren't mutually exclusive: the multi-year structural gap can be real at the system level while individual zones and individual generators see uneven, sometimes disappointing, realized economics along the way.
5. What would actually change my mind
The clearest falsifiable signal is the next capacity auction result itself — if the 2027/28 auction clears meaningfully below the $329.17 cap, that's real evidence supply is catching up faster than the current setup suggests. Nearer-term, I'm watching two things already visible in this quarter's reports: whether TLN's PPL zonal basis discount keeps widening (a name-specific erosion of the broader tailwind) and whether VST's flagged ERCOT forward-price softness spreads into PJM commentary too. Either would suggest the capacity-price story is stronger in the auction mechanics than in generators' actual realized cash flow — a meaningful distinction this piece doesn't have enough data yet to resolve either way.
Related research
The grid constraint behind this piece is also the subject of Data Center Alley's section on Northern Virginia's own multi-year interconnection wait times — the same structural bottleneck, viewed from the physical corridor rather than the auction mechanics. AEA's own thematic concentration in this exact buildout is quantified on the Thematic Exposure page.
Methodology & sources
Capacity auction clearing prices. 2024/25 ($28.92) and 2025/26 ($269.92) figures, and the 2026/27 result ($329.17, hitting the FERC price cap): Utility Dive, "PJM capacity prices hit record high." Cross-referenced against Enel North America's auction summary and PJM's own auction results release.
Data-center share of cost increase (63%, $9.3B). IEEFA, "Projected data center growth spurs PJM capacity prices by factor of 10."
Long-term load forecast (253 GW by 2046, 58% increase). PJM Inside Lines, "PJM's Updated 20-Year Forecast."
Near-term large-load additions (35.1 GW vs. 34.6 GW total growth, 2026–2031) and zonal demand-growth share (DOM/AEP/ComEd/PPL, 74%). Modo Energy, "Data centers define PJM's 2046 load forecast."
Interconnection queue (811 projects, 220 GW) and utility large-load forecasts (55 GW by 2030, 100 GW by 2037). EPSA, "800+ Projects Enter PJM Queue."
TLN, VST, VRT Q2 2026 figures. AEA's own earnings reports for TLN, VST, and VRT, each sourced from the respective company's own investor-relations release.
Chart. Hand-built SVG; bar heights and dollar labels are the actual cited clearing prices, not estimates.
Limitations. PJM capacity-auction dynamics involve zonal variation this piece doesn't fully model — the cited clearing prices are for "most zones," and some zones (including areas relevant to TLN's PPL exposure) clear differently. This is a single-market study; ERCOT, MISO, and other regional markets face related but distinct dynamics not covered here.
Not investment advice. Nothing here is a recommendation to buy, sell, or avoid any security.