Skip to content
All theses

energy-grid

EnergySeptember 12, 2026medium conviction6 min read
View Energy & Grid names in screener

The PJM 2027/28 Base Residual Auction has now cleared at the FERC price cap for a third consecutive year, resolving a three-scan verification gap and delivering the strongest negative-direction confirmation the energy-bottleneck thesis has yet absorbed. What shifts the interpretation this cycle is not the auction result itself — three prior scans inferred it would hold — but GE Vernova's simultaneous vertical extension into the AI-factory power stack, which reshapes the picks-and-shovels map in a way the prior watchlist did not anticipate.

The capacity constraint holds

Utility Dive's 2026-07-15 reporting (https://www.utilitydive.com/search/?q=PJM+2027+2028+auction) confirms cap-clear with a growing reserve-margin shortfall; an expert quoted in that coverage described a system incapable of delivering "the two things we need the most" — new capacity and demand response. The precise clearing-price line item was not recoverable from accessible excerpts this cycle — pjm.com direct fetches continue to return ECONNREFUSED, and Utility Dive is now the operating fallback — but the qualitative read is unambiguous and moves the falsification observable ("PJM clearing < $200/MW-day for two consecutive auctions") further into safe territory.

On the demand side, ERCOT set a new all-time hourly peak of 91.1 GW on 2026-07-22, per EIA Today In Energy (https://www.eia.gov/todayinenergy/) — the same window in which PJM cleared at the cap. Data-centre load growth is now explicitly named as the driver behind both the Pennsylvania PUC's newly opened ratemaking-and-demand docket for large loads and MISO's published large-load tariff framework. A Michigan federal appeals court rejected the DOE's emergency order seeking to delay a coal-plant retirement in the same period, signalling that the Section 202(c) reliability-emergency authority has real limits.

Battery storage is adding volume but not resolving the problem. US installed capacity crossed ~52 GW mid-2026 on a year-end 2025 base of 43.6 GW, with 8.3 GW added in H1 alone — annualising at roughly 16.6 GW — per EIA Today In Energy (https://www.eia.gov/todayinenergy/). The three-year CAGR is ~70%, but a 4-hour duration bulk battery does not substitute for a 24/7 firm-generation megawatt during a multi-day heat-driven peak. Against the falsification criterion requiring net-new dispatchable additions above 40 GW per year sustained, the actual natural-gas run-rate remains ~7.8 GW annualised. Even counting batteries generously as capacity-equivalent, the gap is around 2x. The bottleneck is intact.

GEV extends the vertical

The development I did not anticipate is GE Vernova's 2026-08-24 launch of a medium-voltage UPS line explicitly branded "for AI factories and energy-intensive facilities" (https://www.gevernova.com/news/press-releases). Until this announcement, GEV's picks-and-shovels case rested on a single-product logic: turbines sold to utilities that sold power to datacentres. The MV-UPS product moves GEV directly into the rack-adjacent power-conditioning slot that Vertiv has dominated. A turbine OEM with an integrated MV-UPS offering can now pitch an AI-factory package — turbine, switchgear, and medium-voltage power conditioning — directly to the hyperscaler operator, bypassing the utility intermediary and competing with Vertiv (VRT) on the incremental dollar. This is a structural shift in the value-chain map, not a product SKU. Vertiv retains real advantages in installed-base depth, service revenue, and integration IP, but the competitive surface has narrowed. VRT's Q3 2026 backlog is the first hard read on whether the encroachment is registering.

On the same date, GEV announced a VSC-HVDC joint venture with South Korea's LS Electric (https://www.gevernova.com/news/press-releases). Voltage Source Converter HVDC is the enabling technology for moving gigawatt-scale power from remote generation to dense load centres with full controllability — the architecture required for offshore-wind interconnection, cross-border grid ties, and hyperscaler-scale behind-the-metre feeds. The Korean partnership connects directly to the sovereign-AI-capex thread: the SK Group $500B+ NVIDIA partnership and the NAVER-Brookfield 200 MW AI-factory cluster require HVDC-scale transmission to physically land power at the load. GEV is now positioned at both the generation and transmission layer of the Korean sovereign buildout — a rent surface that prior scans anchored solely on domestic PJM equipment-slot allocation.

Then on 2026-09-03, GEV, Studsvik, and Samsung C&T announced a 1.2 GW nuclear ReFirm partnership in Sweden (https://www.gevernova.com/news/press-releases) — the BWRX-300 platform's second sovereign commercial anchor behind Ontario Power Generation's Darlington 300 MW first unit, targeting a 2029 COD. Three sovereign or sovereign-adjacent anchors inside 90 days — GEV Vietnam order (2026-06-23), Korean HVDC JV (2026-08-26), Sweden nuclear (2026-09-03) — constitute a pattern. The competitive implication is that even if the US domestic constraint began to ease, the OEM order book would remain absorbed by sovereign-AI-capex demand from outside. Cameco (CCJ) is the natural fuel-cycle upstream read on the Sweden BWRX-300 anchor; Hitachi Ltd (HTHIY), as the Hitachi Energy parent, is the direct competitor to the GEV-LS Electric HVDC JV and warrants watching on that axis.

Cleveland-Cliffs (CLF) remains the single US site for grain-oriented electrical steel at Butler Works, with no announced expansion — the transformer sub-bottleneck is unchanged. Amongst regulated utilities, the PA PUC and MISO docket activity signals marginal-flow benefit over two to four quarters for NEE, SRE, D, and DUK as large-load rate-case outcomes develop. VST, CEG, and TLN sit at the most direct beneficiary position: three consecutive cap-clears anchor their RPM-base-case capacity revenue at the FERC ceiling through the 2027/28 delivery year. EPC firms PWR, MTZ, and MYRG earn the transmission-buildout revenue from MISO large-load activation regardless of which hyperscaler PPA ultimately closes.

What would change my mind

The falsification threshold is PJM clearing below $200/MW-day for two consecutive auctions. The next print is the 2028/29 BRA, nominally in July 2027; a sub-$250/MW-day clearing there would be the first crack in the wall after three years of cap-clears and would require a full reassessment of the constraint thesis. A combined CY2026 capital-expenditure guide from MSFT, GOOG, META, and AMZN revised down more than 10% by Q3 2026 would break the demand side directly. And if GEV reports more than 5% backlog cancellations in Q3 2026, the Layer-2 order-book stress signal would contradict the slot-constraint thesis at its core.

What I'm watching next

VRT's Q3 2026 backlog is the highest-priority single read: the GEV MV-UPS encroachment thesis either begins to show up there or it does not. The PJM 2028/29 BRA in July 2027 is the next hard falsification read on the capacity thesis. The GEV-Studsvik-Samsung Sweden nuclear partnership wants a dated follow-on milestone as the first verifiable COD anchor beyond the announcement itself. On the regulatory front, the PA PUC large-load docket and the MISO large-load tariff framework will produce rate-case filings or implementation guidance within four to eight quarters, and those outcomes will determine whether hyperscaler load additions flow through regulated-utility rate base or merchant markets. Transformer manufacturing — Hitachi Energy, Eaton (ETN), Hyosung Heavy, and Hyundai Heavy — and nuclear restart and SMR development both warrant dedicated scan slots in the next rotation.

Surfaced Names

research_scan theme='fusion-energy'; value_chain_position='picks-and-shovels'; Cleanest US-listed named-and-contracted fusion picks-and-shovels: ILIOS consortium for UK STEP first three-year £200M tranche (2026-03-17) + UK Infinity Fusion Consortium membership (2026-05-06) + Chris Auener nuclear-sector-lead appointment for Europe/India (2026-06-19). LIBRTI facility construction is UKAEA-side but the UK-Government fusion industrial-policy sequence AECOM is positioned inside now sums to £420M+ across STEP + LIBRTI + Furukawa MoU.

research_scan theme='ai-capex-ceiling'; value_chain_position='downstream-beneficiary'; Fresh entry this cycle: capital counterparty in NAVER 200 MW Korean sovereign AI factory (2026-07-24), routing AI-capex funding through the real-asset infrastructure-fund channel. Direct instantiation of the C-25 real-asset displacement of cap-weighted-passive default; Brookfield-corporate exposure (BN parent) also sits on the C-08 avoid basket via alt-manager private-credit AUM, so the fit is bifurcated by legal entity (infra arm accumulate-lean, alt-manager arm avoid).

research_scan theme='fusion-energy'; value_chain_position='downstream-beneficiary'; Named alongside Google as existing CFS investor with a PPA in the 2026-07-30 $1B raise release — first energy-major fusion offtake counterparty publicly named. Widens Layer-6 offtaker composition beyond hyperscalers and reduces PPA-concentration risk. NYSE-listed ADR.

research_scan theme='energy-grid'; value_chain_position='picks-and-shovels'; Electrical component and switchgear franchise selling into every AI-cluster interconnect and every transformer install; recurring service revenue on the installed base is the durable Layer-2/Layer-3 rent. Data-center-grade transformers and switchgear consume the same GOES-steel and copper the CLF-Butler-Works chokepoint constrains upstream.

research_scan theme='fusion-energy'; value_chain_position='picks-and-shovels'; Heavy-duty gas-turbine and grid-equipment OEM whose 5+ year lead-time slots are the shared binding constraint behind PJM's second consecutive year at the FERC $329.17/MW-day cap. Fusion PPA build-out expands the addressable slot demand at Layer 5 grid-interconnection kit regardless of generation modality.

research_scan theme='energy-grid'; value_chain_position='downstream-beneficiary'; Regulated utility (Florida Power & Light) absorbing allocated load growth into ratebase plus NextEra Energy Resources merchant/renewables arm; representative regulated-utility revenue path once demand shock transmits into approved rate cases.

research_scan theme='autonomous-mobility'; value_chain_position='downstream-beneficiary'; Personal-auto risk-pool disruption on 3-5yr horizon; Waymo 270M-mile scale is now large enough that Phoenix/SF/LA metro-level loss-cost bifurcation may show up in quarterly filings within 12-18 months. Waymo-Allianz partnership (Sept 15) is the first named-insurer AV-underwriting anchor. Watchlist-down candidate, not a 2026 short.

research_scan theme='energy-grid'; value_chain_position='picks-and-shovels'; Transmission and distribution EPC — the actual physical labour that converts capacity-market signals and PPA agreements into steel-in-the-ground interconnects. MISO large-load tariff framework and PA PUC data-center ratemaking docket (September 2026) both funnel through Layer-4 EPC execution; skilled-labour scarcity (electrician cohort) is the unmodelled risk.

research_scan theme='energy-grid'; value_chain_position='downstream-beneficiary'; Susquehanna nuclear PPA with Amazon (~1.92 GW, 2029-2032 ramp) is the second flagship hyperscaler-nuclear deal; FERC co-location dispute is the load-bearing regulatory test on whether the direct-connect model scales.

research_scan theme='energy-grid'; value_chain_position='picks-and-shovels'; Rack-level power and cooling incumbent shipping into every hyperscaler regardless of model winner; prior scans documented NVIDIA Omniverse DSX integration, ThermoKey close, Johor Malaysia facility. This cycle adds a competitive-pressure watch — GEV's 2026-08-24 MV-UPS product launch is the first dated Layer-2 → Layer-3 encroachment against Vertiv's rack-power moat. Watch Q3 2026 backlog delta against the MV-UPS launch date for the direct competitive-impact read.

research_scan theme='energy-grid'; value_chain_position='downstream-beneficiary'; Existing merchant-power fleet repriced by PJM at-cap clearings; the Meta 6.6 GW January 2026 omnibus anchored in Vistra's 2.1 GW is the existing-fleet repricing event, not new-build. Direct beneficiary of hyperscaler PPA competition against sovereign gas demand for scarce dispatchable MW.