V
Power, grid & generation
Datacenter interconnect queues and transformer lead times are measured in years; gigawatts are the new wafer starts.
Compute scaled past the chip and hit the wall every industrial economy eventually hits: you cannot run a gigawatt of accelerators without a gigawatt of firm power, the transformers to step it down, and the wire to carry it. Interconnection queues and transformer lead times are now measured in years, and only about one-third of the 12–16 GW of US data-center capacity slated for 2026 is actually under construction — the rest is waiting on electrical equipment with lead times now extending to about four years. Gigawatts are the new wafer starts.
Who owns it
The choke is a stack, and ownership concentrates at each layer. At generation, the merchant nuclear/gas independents own the only firm, dispatchable megawatts hyperscalers can contract today. Constellation closed its $16.4B Calpine acquisition in January 2026 to reach ~60 GW and has signed Meta (1,121 MW at Clinton from June 2027) and Microsoft (Three Mile Island restart). Vistra has contracted ~3,800 MW of nuclear to AWS and Meta. Talen is the purest expression: up to 1,920 MW to Amazon from Susquehanna, ~$1.4B/year at full quantity.
At the prime mover, gas turbines are sold out. GE Vernova's turbine backlog plus slot reservations hit 100 GW in Q1 2026 with slots effectively booked through 2030 — and the same company owns the electrification layer, booking $2.4B of data-center equipment orders in a single quarter, more than all of 2025. Below that, the distribution choke runs from substation to rack: Eaton's Q1 data-center orders rose ~240% with electrical backlog up 48%, and Hubbell raised its 2026 data-center growth outlook above 25%. Connecting it all is the EPC labor choke — Quanta's record $48.5B backlog, $40.1B in electric power.
What breaks it
Three vectors. First, hyperscaler internalization: the buyers are increasingly going around the IPPs — contracting turbines directly, building behind-the-meter gas, and funding on-site SMRs. Every megawatt internalized is a PPA that never gets signed. Second, regulation: FERC's December 18, 2025 order directing PJM to write new co-location and behind-the-meter rules is unresolved, and it already forced Talen to restructure its Amazon deal from behind-the-meter to grid-connected. The economics of the entire IPP-datacenter model hinge on tariff language still being drafted. Third, the material floor: the transformer shortage is downstream of grain-oriented electrical steel, and no listed merchant US large-power-transformer pure-play exists — Cleveland-Cliffs is the sole domestic GOES producer but buries that exposure inside an automotive-steel P&L. The hardest physical choke is the least cleanly investable.
What forces the reprice
Capacity-market price discovery is the near-term trigger: PJM's last two Base Residual Auctions both cleared at the cap on data-center-driven load, and the 2028/2029 auction closes July 7, 2026, repricing capacity revenue for CEG, VST and TLN simultaneously. Then the summer prints — GE Vernova (July), Quanta, Eaton — test whether record orders are converting to recognized revenue or are a pull-forward. Structurally, every new named hyperscaler PPA and every transformer lead-time extension reprices firm-power scarcity.
Be honest about size and earliness
Most of the revenue here is narrative, not yet realized. The IPP PPAs are dated 2027–2032 — Talen's $1.4B is an at-full-ramp figure with escalators starting 2028, not run-rate; Constellation's Meta megawatts begin in June 2027; the TMI restart targets 2028 and carries NRC and construction risk. The equipment names show the cleanest acceleration, but data centers are still a minority of total revenue at Eaton and Hubbell, and the growth overlaps a broad electrification cycle that would be running regardless of AI. GE Vernova is the rare name where the data-center share is both disclosed-ish (~20% of backlog) and large — yet even there, backlog is not revenue and slots can be cancelled. This is a real, multi-decade physical chokepoint with genuine sold-out capacity, but in 2026 it is mostly a backlog-and-PPA story whose cash flows arrive at the back of the decade. Own the layer that is sold out and disclosed; treat the back-dated PPAs and the indirect material plays as the call options they are. This is structure and catalyst description, not investment advice.
Who owns the choke
Constellation Energy Corporation
Closed the $16.4B (≈$26.6B with debt) Calpine acquisition in January 2026, creating the largest US power producer at ~60 GW. Signed Meta to a 20-year, 1,121 MW PPA at Clinton (deliveries from June 2027) and is reviving Three Mile Island Unit 1 (~835 MW) under a 20-year Microsoft PPA targeted for 2028. Affirmed 2026 adjusted operating earnings guidance of $11.00–$12.00/share.
- Bull
- If firm-power scarcity reprices, Constellation's ~60GW post-Calpine fleet plus 20-year Meta and Microsoft TMI PPAs make it the largest contracted-nuclear supplier.
- Bear
- Meta megawatts begin June 2027 and the TMI restart targets 2028 with NRC/construction risk, while FERC's unresolved co-location rules threaten the IPP model.
Cleveland-Cliffs Inc.
Produces grain-oriented electrical steel exclusively in the US at its Butler, PA works — the core material constraint behind multi-year transformer lead times — and is converting its Weirton, WV facility to produce three-phase distribution transformers (opening 1H 2026), which would lift GOES output 30%–40%. Datacenter exposure is indirect and not separately disclosed.
- Bull
- If transformer lead times stay at four years, Cleveland-Cliffs' sole US GOES production and Weirton conversion lifting output 30-40% addresses the hardest physical choke.
- Bear
- Datacenter exposure is indirect and undisclosed, buried inside an automotive-steel P&L, making it the least cleanly investable name in the power layer.
Eaton Corporation plc
Q1 2026 data-center orders rose ~240% YoY and data-center revenue ~50%, with total Electrical backlog up 48%. Management raised 2026 organic growth guidance to ~10% and frames a 228 GW total US data-center backlog — about 12 years of demand at 2025 build rates — roughly 70% AI-tied.
- Bull
- If the 228GW US datacenter backlog converts, Eaton's ~240% data-center order growth, 48% electrical backlog rise and raised guidance ride distribution scarcity.
- Bear
- Data centers are still a minority of revenue, and the growth overlaps a broad electrification cycle that would run regardless of AI.
GE Vernova Inc.
Gas turbine backlog plus slot reservations reached 100 GW in Q1 2026 (guiding to 110 GW by year-end), with ~20% of backlog data-center driven and turbine slots effectively sold through 2030. Its Electrification segment booked $2.4B of data-center equipment orders in Q1 — more than all of 2025 — and the company raised 2026 revenue guidance to $44.5B–$45.5B.
- Bull
- If gas-turbine scarcity holds, GE Vernova's 100GW backlog booked through 2030 and $2.4B single-quarter electrification orders make it the rare disclosed-and-large data-center power play.
- Bear
- Backlog is not revenue and slots can be cancelled; ~20% data-center share rides a broad electrification cycle, and orders may be a pull-forward.
Hubbell Incorporated
Raised its full-year 2026 data-center growth outlook above 25% after data-center revenue grew 40% in Q1 2026, driven by balance-of-system components and modular power-distribution skids; Grid Infrastructure sales rose to $727.1M from $617.7M YoY. Raised full-year organic growth guidance to 6%–9%.
- Bull
- If datacenter power distribution stays tight, Hubbell's >25% data-center growth outlook on balance-of-system components and modular power skids rides the buildout.
- Bear
- Data centers are a minority of revenue, and the growth overlaps a broad electrification cycle that would be running regardless of AI demand.
Quanta Services, Inc.
Reported a record $48.5B backlog in Q1 2026, $40.1B in Electric Power Infrastructure, with management citing expanding data-center, 765 kV transmission and generation opportunities. Raised full-year 2026 guidance to $34.7B–$35.2B revenue and adjusted EPS of $13.55–$14.25.
- Bull
- If grid modernization accelerates, Quanta's record $48.5B backlog ($40.1B electric power) and raised 2026 guidance ride the EPC-labor choke connecting datacenters to the grid.
- Bear
- Backlog is not revenue, data-center work is one driver among transmission and generation, and the growth overlaps a broad electrification cycle.
Talen Energy Corporation
Expanded its Amazon relationship in June 2025 to supply up to 1,920 MW from the Susquehanna nuclear plant, guiding to ~$1.4B of annual revenue at full contract quantity (2% escalators from 2028). Restructured the arrangement from behind-the-meter to grid-connected front-of-the-meter during the spring 2026 refueling outage and signed an SMR letter of intent with X-energy for PJM deployment.
- Bull
- If PJM capacity reprices firm power, Talen's up-to-1,920MW Susquehanna supply to Amazon (~$1.4B at full quantity) and X-energy SMR LOI monetize merchant nuclear scarcity.
- Bear
- The $1.4B is an at-full-ramp figure with 2028 escalators, and FERC's co-location rules already forced a behind-the-meter-to-grid-connected restructuring.
Vistra Corp.
Signed a 20-year PPA with AWS for up to 1,200 MW of carbon-free power at the Comanche Peak nuclear plant (deliveries from late 2027, full capacity by 2032), part of ~3,800 MW of nuclear contracted to AWS and Meta. Guided 2026 ongoing-operations adjusted EBITDA of $6.8B–$7.6B and is acquiring ~5,500 MW of gas via Cogentrix.
- Bull
- If PJM capacity reprices, Vistra's 20-year AWS Comanche Peak PPA, ~3,800MW contracted nuclear and Cogentrix gas acquisition monetize firm-power scarcity.
- Bear
- AWS deliveries begin late 2027 reaching full capacity by 2032, so much of the contracted revenue arrives at the back of the decade, not run-rate.
Catalyst calendar
- 2026-07-07PJM 2028/2029 Base Residual Auction closeshighReprices capacity revenue for every PJM generator (CEG, VST, TLN) at once; the prior two auctions cleared at the cap on data-center-driven load growth.
- 2026-07-22GE Vernova Q2 2026 resultsmediumCleanest read on whether 100 GW of gas-turbine slots and $2.4B of data-center electrification orders are converting into recognized revenue and margin.
- 2026-07-30Quanta Services Q2 2026 resultsmediumTests whether the record $48.5B backlog is translating into accelerating electric-power and data-center connection revenue, the EPC choke on grid build-out.
- 2026-08-05Eaton Q2 2026 resultsmediumWatch whether the ~240% data-center order growth and 48% electrical backlog build sustain, confirming the distribution-equipment layer as a durable chokepoint rather than a pull-forward.