Why Data Centers Are Turning Energy Stocks Into AI Plays

U.S. data center power demand is projected to jump from 540 kWh per capita in 2024 to 1,200 kWh by 2030, creating a supply bottleneck that is lifting energy utilities into “AI growth” roles. Constellation signed ~920 MW of long-term nuclear contracts averaging 18.5 years and locked in ~30% of clean baseload output, while adding a first-ever nuclear deal with Walmart covering ~176 MW across 15-year terms. Vistra meanwhile secured a 2,600 MW power purchase deal with Meta and up to 1,200 MW for AWS, reinforcing multi-decade, fixed-price revenue visibility (with inflation escalators) despite regulatory and AI capex rollover risks.

Analysis

The clearest winner is not “utilities” broadly but owners of scarce, dispatchable baseload with existing interconnects. That makes CEG and VST more like infrastructure monopolies than regulated rate-base names: the value is in contract duration, not spot power prices. A less obvious beneficiary is KKR, which can monetize the financing layer if power and data-center buildouts migrate toward private capital structures; by contrast, hyperscalers like AMZN, MSFT, META, and ultimately NVDA face a hidden ROIC tax as power availability becomes the pacing item, elongating the payback on AI capex and forcing more spending into less optimal sites.

The near-term move is mostly multiple re-rating, while the cash-flow evidence arrives much later. Over 1-3 months, the biggest catalysts are incremental PPA announcements, backlog disclosure, and any commentary that contracted MW is accelerating faster than expected; the biggest reversal is regulatory scrutiny of co-location/behind-the-meter arrangements or a pause in hyperscaler capex. Over 6-18 months, the key question is whether grid constraints force a durable scarcity premium or whether new supply, policy intervention, or demand normalization compresses it.

Consensus is likely underestimating how much of the AI boom’s marginal economics will be captured by power owners rather than chip/cloud vendors. But the market may also be overpaying for the story if it assumes contracted megawatts translate quickly into earnings; interconnection and permitting lag create a real timing mismatch. If the AI buildout slows, CEG/VST could still look optically expensive even with strong contract books, because the stock is front-running cash flows that sit years out.

Key Decisions for Investors

• Long CEG on 5-10% pullbacks, 6-12 month horizon; thesis is scarcity premium plus contracted cash-flow visibility. Falsify on weaker-than-expected backlog growth, softer nuclear contract pricing, or hostile regulatory language around co-location.
• Long VST vs short XLU as a 3-6 month relative-value trade; VST should outperform if power scarcity keeps favoring contracted baseload over generic regulated utility beta. Exit if the spread tightens after a regulatory event or if AI capex commentary deteriorates.
• Add KKR as a second-order beneficiary on a 6-18 month view; financing and platform ownership should capture more of the capex cycle than the market is modeling. Risk is that project economics shift back to balance-sheet utility funding, reducing fee/asset growth optionality.
• Avoid chasing AMZN/MSFT/META into the next AI-capex leg until there is evidence that power procurement is not constraining deployment. If you want exposure, prefer a smaller weight and treat it as a delayed-ROI trade rather than a pure AI upside play.