The Cliff Nobody Is Modeling argued that the US grid isn’t facing a managed slowdown — it’s facing a step-down, concentrated in 2028, as the pre-OBBBA pipeline exhausts and supply chain restrictions block meaningful replacement. Here’s what that implies if you’re watching markets.
What the thesis suggests watching
The mechanism is sequential and physical, which matters for timing. The record 86 GW expected in 2026 — and a comparable surge in 2027 as developers race to beat credit deadlines — is already permitted, financed, and under construction. Those projects clear the FEOC supply chain restrictions automatically. What the thesis is actually claiming is that once this pipeline burns down, the replacement rate falls off a ledge. And the grid’s own accounting is already saying so.
PJM Interconnection — the grid operator covering 65 million people in 13 states — ran its 2027/2028 capacity auction last December and came up short for the first time in its history. That’s the first edge of the cliff showing up in planning data before the pipeline has even exhausted. The subsequent PJM auctions covering 2028 and 2029 delivery years are the primary signal to watch. If they clear at prices near historical norms, the thesis is wrong. If they come in short again — or if clearing prices spike — the grid’s own accounting is confirming the gap before a single reliability event occurs.
The second thing worth watching is GE Vernova’s (NYSE: GEV) backlog disclosures. Gas turbines can’t close a 2028 gap — lead times are five years, and GEV’s backlog roughly doubled in 2025, with delivery slots already booking into 2030. GEV’s earnings calls have become a real-time indicator of how seriously utilities are trying to fill the gap with gas and failing on timeline. A backlog that plateaus or contracts would be the signal that either the cliff isn’t materializing or utilities have found another path.
The third watch item is the hyperscalers. Microsoft (NASDAQ: MSFT), Amazon (NASDAQ: AMZN), and Google (NASDAQ: GOOGL) are the largest and most creditworthy power buyers in the country — they have procurement options that smaller buyers don’t. Watch for behind-the-meter generation announcements, long-duration power purchase agreements with non-utility generators, or moves toward co-location with dispatchable assets. The frequency and urgency of those announcements is a proxy for how seriously large buyers think the gap is real.
The bull case
If the thesis holds, the most direct market expression is capacity-constrained power in PJM’s footprint — and the clearest beneficiaries are companies that own dispatchable generation that can’t be replicated on any relevant timeline.
Vistra (NYSE: VST) and Constellation Energy (NASDAQ: CEG) are the obvious names: large owners of nuclear and gas in the mid-Atlantic and Midwest, already re-rated as capacity market prices have moved. The bull case is that the 2028/2029 auctions produce another shortfall, locking in elevated capacity payments for years and making existing nuclear structurally precious in a way the market is still partially discounting. Constellation’s Three Mile Island restart — contracted directly to Microsoft before a single capacity auction cleared — is the template: a hyperscaler paying a premium to own the output of an asset the grid can’t replace, years before the shortage becomes visible in retail prices.
Talen Energy (NASDAQ: TLN) is the most concentrated expression of this thesis. Talen owns the Susquehanna nuclear plant in Pennsylvania — one of the largest nuclear facilities in the US — and the Amazon co-location deal, though voided by FERC on interconnection grounds, established the underlying logic clearly: around-the-clock carbon-free power inside PJM’s footprint is exactly what hyperscalers are willing to pay a premium to access. If the capacity gap materializes as the thesis describes, Talen’s positioning becomes less about one contested deal and more about a template the sector is trying to replicate at scale.
That replication is where the opportunity gets specific. The procurement ladder the hyperscalers are climbing — standard utility contract, then long-duration PPA, then co-location agreement, then equity stake in generation — represents an escalating willingness to pay for certainty. Each rung is a higher premium for the asset owner. VST, CEG, and TLN are all sitting on assets that large buyers are actively trying to contract or acquire. The bull case isn’t just elevated capacity payments — it’s that the scarcity premium gets competed into asset valuations directly, as hyperscalers move from buying power to buying the plants.
The second leg is infrastructure adjacent to the gap: high-voltage transmission, grid software, and load management. Quanta Services (NYSE: PWR) and MYR Group (NASDAQ: MYRG) are positioned on the build-out side regardless of which generation technology eventually closes it.
The bear case
Two things could break the market expression of this thesis.
The first is the FEOC safe harbor tables. Treasury’s formal December 2026 guidance is the real test — the February 2026 interim guidance came in more workable than feared. If the formal tables provide certification pathways that allow post-2025-start projects to reach financial close at scale, the replacement pipeline refills faster than the step-down thesis requires. The signal to watch: developer announcements of post-2025-start projects moving to financial close after the tables are published. That would tell you the replacement rate is recovering before the capacity auctions confirm it.
The second risk is AI efficiency compressing data center load faster than adoption expands. DeepSeek-style architectural gains are real, and the trajectory is genuinely hard to model over an 18-month horizon. The Jevons dynamic cuts against a clean efficiency story — cheaper inference tends to expand the market rather than shrink consumption — but the uncertainty is honest. A sustained downward revision in hyperscaler capex guidance, across multiple quarters, would be the signal this risk is materializing.
Watch OEM power procurement disclosures. The base case for what routing around the grid looks like is already partially visible: Microsoft’s agreement with Constellation to restart Three Mile Island Unit 1, Amazon’s attempted co-location deal at Susquehanna, Google’s nuclear PPA with Kairos Power. These aren’t one-off experiments — they’re large buyers solving the same problem independently and arriving at the same answer: own or contract the generation directly, bypass the capacity auction entirely. If the cliff thesis is right, the next phase looks like more of this, faster: hyperscalers moving from PPAs into equity stakes in generation assets, data center campuses announced with on-site power already contracted rather than grid-dependent, and deal timelines compressing as competition for dispatchable megawatts intensifies. The signal that the thesis is breaking isn’t the absence of these deals — it’s a return to standard utility contracts, which would mean the grid has enough headroom that large buyers no longer need to route around it.
Tracked call
PJM’s delivery year 2028/2029 Base Residual Auction clears at a price more than 2× the 2027/2028 clearing price, reflecting a persistent capacity shortfall.
Falsification window: PJM BRA results for delivery year 2028/2029, expected Q4 2026.
Confidence: Medium
What would change my mind: Treasury’s December 2026 FEOC safe harbor tables are permissive enough to produce credible developer announcements of post-2025-start projects reaching financial close before the auction date — signaling the replacement pipeline is refilling faster than the cliff thesis requires.
What would change my mind
- Treasury’s December 2026 FEOC safe harbor tables provide workable certification pathways for post-2025-start projects, producing developer announcements of new projects reaching financial close before the PJM BRA auction date.
- Hyperscaler capex guidance revises downward across multiple consecutive quarters, signaling that AI efficiency gains are compressing data center power demand faster than adoption is expanding it.
Working Hypothesis tracks every thesis publicly. The main piece’s scorecard call and this companion call are scored independently. Prior calls and resolution notes are at working-hypothesis.com.
This is not investment advice. Working Hypothesis is an analytical publication. All positions and calls are tracked publicly on the scorecard.