AI Can't Thrive Without This Stock (Hint: It's Not Nvidia)

Constellation Energy (CEG) is positioned as a direct enabler of AI data centers via nuclear power supply, highlighted by a 20-year power purchase agreement with Microsoft for Three Mile Island (restarting planned operations in 2024) and an additional 20-year PPA with Meta for power from the Clinton Clean Energy Center, expected to resume in 2027. In June 2025, Constellation’s acquired Calpine unit also secured a 380 MW connection agreement with CyrusOne for a new data center in Texas, complementing a prior 400 MW PPA. The article frames CEG’s nuclear capacity as immediately available versus advanced nuclear peers that require regulatory approvals, implying a supportive demand backdrop from AI-driven power needs.

Analysis

AI infrastructure is moving from a chip-scarcity narrative to an energized-capacity narrative. The economic moat here is not electricity sales per se; it is access to existing baseload, interconnects, and long-duration contractability, which should justify a scarcity premium for CEG versus less-flexible utilities and especially pre-commercial nuclear developers with multi-year regulatory lag. Second-order winners include gas-heavy merchant generators and grid-equipment vendors as hyperscalers scramble for any firm load they can secure.

Near term, the equity can rerate faster than cash flow because PPAs and partnership headlines are sentiment-positive while the earnings bridge is slow. The key catalyst path is 1-3 quarters of contract visibility and 6-18 months of proof that contracted power converts into margin expansion rather than just lower volatility. The main reversal risks are falling power prices, slower hyperscaler capex, interconnect delays, or policy/execution issues around nuclear restarts; if forward power spreads compress, the scarcity premium should mean-revert quickly.

Contrarian view: the market is still underpricing how binding power constraints are relative to semiconductor supply, but it may already be partially discounting that insight in CEG. For the hyperscalers, these deals reduce schedule risk more than they improve cost economics, so the real upside is to project timing, not to margin structure. If customers shift toward self-generation, demand response, or more flexible architecture, the premium on pure baseload owners could cap out while chip names continue to compound.

Key Decisions for Investors

• Long CEG on pullbacks over a 3-6 month horizon; best risk/reward is owning the scarcity multiple ahead of additional PPA announcements. Falsify if forward power spreads soften or if no incremental contracted load appears over the next 1-2 quarters.
• Relative-value: long CEG / short NVDA for 1-3 months only if the market starts extrapolating AI demand faster than energized capacity can be delivered. This isolates the bottleneck trade; cover if NVDA commentary shows deployment is still outrunning supply and power access is not binding.
• Add MSFT or META on weakness as a slower-burn beneficiary of lower data-center schedule risk; this is a quality-duration expression, not a near-term earnings pop. Falsify if capex guidance is cut or data-center build timelines slip again.
• Watch MU for estimate risk if power constraints start delaying GPU deployment rather than chip demand. If that gap widens, MU can trade better on inventory tightness but worse on revenue timing; use earnings revisions as the trigger rather than the headline AI theme.