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

Global data-center electricity consumption is projected to rise from about 415 TWh in 2024 to roughly 1,000 TWh by 2030, while power infrastructure construction timelines of 3-10+ years create a supply bottleneck versus 18-24 months for AI data centers. GE Vernova signed 20 GW of gas-equipment agreements in Q2, has 116 GW under contract, and reported a $40.6B electrification backlog, up 69% year over year. Constellation operates 22 GW of nuclear capacity and added 23 GW of assets through its $21.8B Calpine acquisition, while Quanta's backlog reached a record $53.4B and it raised midpoint revenue guidance to $39.5B and free-cash-flow guidance to $2.25B.

Analysis

The investable bottleneck is not aggregate generation demand but deliverable, time-stamped capacity at constrained interconnection nodes. That shifts pricing power toward owners of dispatchable assets and scarce grid equipment, but only where local capacity markets and transmission constraints permit merchant repricing. CEG has the cleanest near-term earnings torque to higher realized power prices; GEV and PWR monetize through multi-year backlog conversion, making their earnings less sensitive to a one-quarter hyperscaler capex pause but more exposed to execution and labor inflation.

The second-order beneficiary set extends beyond the named equities: Eaton (ETN), Hubbell (HUBB), nVent (NVT), and Powell Industries (POWL) have high-value exposure to switchgear, distribution and power-management equipment, where qualification cycles and transformer lead times support margin durability. Conversely, hyperscalers’ AI returns face a hidden cost escalation: securing firm power can turn data-center expansion from an asset-light compute purchase into a long-duration energy commitment, pressuring incremental AI margins at META and MSFT before it impairs demand for NVDA. The key watch item is whether these firms absorb power costs through higher cloud pricing rather than slow deployment.

Consensus likely overstates the immediacy of broad grid-build earnings. Permitting, interconnection queues and utility rate-case recovery can defer transmission spend even as data-center announcements accelerate; PWR’s backlog is valuable only to the extent it converts at protected labor and materials margins. For the next 1-3 months, power-price and PPA headlines can keep CEG bid, while GEV/PWR need bookings, margin and cash-conversion evidence. Over 6-18 months, a sustained shortage favors equipment suppliers with capacity and service revenue, but a sharp gas-price increase, data-center project deferrals, or federal/state opposition to rate increases would weaken the thesis.

Key Decisions for Investors

• Prefer a 6-12 month long GEV / short ETN pair only on a pullback: GEV has greater turbine and grid-backlog operating leverage, while ETN is the higher-quality but more broadly owned power-electrification proxy. Size for a 15-20% relative-return target; exit if GEV's next two quarters show backlog growth without conversion to segment-margin or free-cash-flow improvement.
• Maintain CEG as a tactical 1-3 month long only if regional forward power prices and incremental contracted load continue rising; use a defined-risk call spread rather than unhedged equity after power-PPA-driven rallies. Falsifier: lower 2027-28 realized-price outlook, adverse nuclear outage trends, or Calpine integration costs that dilute expected free-cash-flow accretion.
• Add PWR only after confirming that new awards include price-escalation protection and that labor productivity offsets hiring costs. A 12-18 month long is warranted by conversion of backlog into cash flow, but avoid chasing on backlog headlines; downside risk is margin compression from craft-labor scarcity and delayed utility approvals.
• Build a diversified equipment basket—long ETN, HUBB, NVT and POWL—rather than treating GEV as the sole grid trade. These names capture distribution-level upgrades required after bulk generation is procured; reduce exposure if order growth decelerates for two consecutive quarters or transformer/switchgear lead times normalize materially.
• Monitor META and MSFT capex commentary for evidence that power availability, rather than GPU supply, is limiting deployment. A disclosed project delay or materially higher energy-commitment cost is a relative negative for META/MSFT versus NVDA over 3-6 months, but remains an alert rather than a recommended short absent quantified guidance impact.