3 Stocks to Buy if You Believe the AI Power Crunch Is Just Getting Started

Global data-center electricity consumption is projected to more than double from 415 TWh in 2024 to roughly 1,000 TWh by 2030, while power infrastructure build times of 3-10+ years lag AI data-center construction timelines. GE Vernova signed 20 GW of gas-equipment agreements in Q2, has 116 GW under contract, and grew its electrification backlog 69% year over year to $40.6B. Constellation has 22 GW of nuclear capacity and expanded through its $21.8B Calpine acquisition, while Quanta reported a record $53.4B backlog, raised midpoint revenue guidance to $39.5B, and lifted free-cash-flow guidance to $2.25B.

Analysis

The investable bottleneck is shifting from generation headlines to interconnection and delivered-power execution. GEV has the highest operating leverage to a sustained turbine/switchgear shortage, but its valuation will increasingly depend on whether incremental capacity converts into higher-margin service agreements rather than lower-margin equipment volume. PWR is the cleaner second-order beneficiary: utilities, independent power producers and hyperscalers must spend on substations, interconnection, transmission and onsite electrical scope regardless of whether the marginal electron comes from gas, nuclear, renewables or storage.

CEG's upside is less about incremental nuclear build and more about the scarcity premium on existing, dispatchable clean capacity in constrained power markets. The Calpine combination reduces pure nuclear exposure but creates a portfolio optimization opportunity: firm nuclear can support premium contracted load while flexible gas/storage manages hourly shaping and reliability obligations. The risk is that long-dated contracted pricing caps upside just as merchant power prices rise; investors should track contracted versus merchant EBITDA, regional forward power curves, and whether large-load PPAs include pass-throughs for transmission and capacity costs.

Consensus is likely over-extrapolating announced data-center load into near-term power demand. Projects can be delayed by financing, chip availability, local opposition, water constraints and interconnection queues; a 12-24 month demand deferral would hit high-expectation equipment multiples before it materially affects utility load. Conversely, the underappreciated risk is that grid constraints force hyperscalers toward behind-the-meter generation, favoring GEV and gas infrastructure while bypassing portions of regulated transmission spend. A meaningful decline in PJM/ERCOT forward capacity prices, slower hyperscaler capex guidance, or backlog conversion slippage would falsify the current scarcity thesis.

Key Decisions for Investors

• Prefer a 6-12 month long PWR / short GEV pair in equal dollar risk units after either reports a data-center-driven rally: PWR has broader technology-agnostic spending exposure and less dependence on turbine-cycle execution. Exit if PWR backlog growth decelerates materially for two quarters or GEV demonstrates sustained service-margin expansion and accelerating free-cash-flow conversion.
• Maintain CEG as a 12-18 month core long only if regional forward power and capacity curves remain firm; use a 10-15% pullback rather than strength to add. The principal risk is a rerating from lower merchant-price expectations or Calpine integration leakage, not nuclear operating availability alone.
• For a higher-beta expression, buy 6-9 month GEV call spreads only after confirmation that new orders are converting to revenue without margin dilution; cap premium at 1% of portfolio NAV. Avoid outright calls if management signals accelerated factory capex or working-capital pressure, as backlog alone does not ensure cash conversion.
• Set a quarterly watchlist on MSFT and META capex guidance, PJM/ERCOT capacity auctions, and utility interconnection timelines. A broad hyperscaler capex reduction or weakening power-market forwards is a signal to reduce GEV/CEG/PWR exposure rather than treat the theme as a long-duration secular certainty.