Nuclear energy to power data centers in Ohio, Pa.

Four nuclear plants in Ohio and Pennsylvania are being restarted or extended to supply around-the-clock power for AI data centers, including 20-year agreements involving Microsoft, Meta, and Oklo. Constellation’s Three Mile Island restart is backed by a $1 billion DOE loan, targets 2028, and is expected to add 835 MW, while Vistra’s three-plant deal will deliver 2,600 MW, with 2,176 MW dedicated to Meta. The article highlights strong federal and state support for domestic nuclear expansion as a cleaner, reliable baseload source for data-center growth.

Analysis

The market is starting to price nuclear less as a “power commodity” and more as a contracted infrastructure layer for AI. That shifts the earnings quality question from spot power prices to long-duration offtake, plant life-extension capex, and regulatory execution—i.e., the real winners are owners of already-permitted baseload assets with low reopening friction, not greenfield developers. The second-order effect is that every incremental MW locked into data-center PPAs tightens available dispatchable supply in PJM, which should support local capacity pricing and make remaining merchant nuclear plants more valuable than the headline deals imply.

The hidden bottleneck is not demand, it’s schedule risk. Restarting or re-licensing plants is a multi-quarter-to-multi-year process with heavy dependence on NRC timing, equipment refurbishment, and grid interconnection; that means the equity reaction likely front-runs cash flow by 12-24 months. In the interim, turbine/generator, switchgear, cooling, and nuclear-services suppliers may capture a better near-term economic uplift than the plant owners themselves, because capex is immediate while revenue recognition is delayed.

Contrarianly, the consensus may be underestimating the political/operational fragility of “clean, reliable” nuclear headlines. Any safety incident, NRC delay, or cost overrun would quickly re-rate the theme, and the market is likely overconfident that AI demand alone can justify every restart at current valuations. The better trade is to own the assets with contracted downside protection and avoid paying full multiple for early-stage advanced nuclear concepts that still face financing, licensing, and commercialization risk.

Key Decisions for Investors

• Long CEG vs. short a basket of uncontracted utility/merchant power exposure for 3-6 months: CEG has the cleanest earnings transmission from long-dated PPA structure, while the pair isolates the value of baseload scarcity in PJM.
• Accumulate VST on pullbacks over the next 1-2 quarters, but size it as an event-driven infrastructure trade, not a secular compounder: the asset base is already in place, so the risk/reward is better than greenfield nuclear plays if NRC/grid execution stays on track.
• Avoid chasing OKLO after the announcement; if you want exposure, use call spreads 12-18 months out instead of stock. The asymmetry is attractive only if you can cap premium in a name where commercialization risk is still unresolved.
• Buy MSFT on any AI-led drawdown as a defensive way to own the demand side of the power bottleneck: even if nuclear schedules slip, Microsoft still benefits from secured compute capacity and can reprice infrastructure faster than peers.
• Pair long CEG / short a renewable-only utility basket for 6-12 months: the market is likely to reward dispatchable reliability over intermittent ESG purity as AI load growth exposes the limits of wind/solar without firm backup.