The Hidden Winners of the AI Power Crunch: 3 Utilities to Watch

The article argues AI-driven power demand is driving a power crunch: AI GPU clusters can require 20–100 kW per rack vs. 5–10 kW for traditional cloud, creating multiyear interconnection backlogs. It highlights Constellation (22 GW nuclear) supported by 20-year PPAs with Microsoft and Meta, noting its stock is down ~35% from the 52-week high and trades around 20x next-year forward earnings with analysts projecting ~22% EPS CAGR. It also flags Vistra’s 20-year Meta PPA for 2,600 MW nuclear power and NextEra’s 35.1 GW renewables/storage pipeline and a partnership on a $100B AI data center campus in Kentucky, while warning merchant-power providers face regulatory and price-cap risks in PJM.

Analysis

The market is underestimating how concentrated the upside is in already-permitted, dispatchable generation. Scarcity of clean baseload in constrained ISOs should keep merchant pricing firm, but the bigger second-order winners are grid equipment, transformers, and interconnection vendors: hyperscalers can buy power, but they cannot buy time-to-power quickly.

CEG has the best mix of scarcity and contract visibility; VST has more operating leverage but also more regulatory beta; NEE is the lower-vol way to own the theme, though its regulated structure makes it a slower earnings catcher and a likely relative laggard if power prices stay elevated. Over the next 1-3 months, the catalyst is incremental PPA disclosure and guidance revisions, not the buildout story itself.

The main tail risk is political intervention if data-center load is seen as shifting grid costs to households or if PJM/state regulators cap behind-the-meter economics. The contrarian miss is that AI demand can be deferred, geographically shifted, or self-built, so power scarcity can stay bullish without translating into linear earnings upside; if interconnection rules ease or gas prices fall, merchant spreads could normalize quickly. Near term, the biggest loser may be NVDA’s deployment cadence rather than its end-demand, which would show up as timing risk, not a collapse in AI spend.

Key Decisions for Investors

• Long CEG on pullbacks for a 6-12 month hold; it is the cleanest scarcity asset, with upside if hyperscaler PPAs continue to re-rate contracted cash flows. Falsify if PJM or state regulators move to cap merchant nuclear economics or if CEG stops raising forward EPS.
• Pair long VST / short NEE over 1-3 months. VST has higher operating leverage to tight power markets, while NEE is a steadier but slower AI proxy; this is the better relative-value expression if power prices stay firm. Exit if NEE’s regulated/renewables pipeline starts converting faster than expected.
• Avoid chasing a broad utility basket for AI beta; prefer merchant power names over regulated utilities until the market proves the bottleneck is demand, not interconnection. If you need lower volatility, keep NEE as a defensive partial hedge, not the core expression.
• Set a watch item on PJM/FERC and Virginia/Illinois regulatory headlines. Any ruling that pushes grid upgrade costs onto data-center owners or restricts co-location would hit CEG/VST first and likely compress the entire AI-power premium.