Bloomberg Businessweek Daily: Nuclear AI Revival (Podcast)

Microsoft’s 20-year agreement to buy all output from Three Mile Island’s remaining 835-megawatt reactor supports the plant’s planned restart as soon as mid-2027 under the new Crane Clean Energy Center name. The deal ties nuclear power to rising AI electricity demand, highlighting a notable infrastructure and energy supply shift. The article is largely factual, but the long-term offtake adds a constructive signal for nuclear power and AI-related power capacity.

Analysis

This is less a single-asset story than a signal that large-language-model capex is colliding with the physical power stack. The key second-order effect is that hyperscaler demand is starting to reprice long-duration, dispatchable generation assets as quasi-infrastructure rather than cyclical utilities, which should widen the valuation gap between firm-power providers and intermittent clean-energy exposure. That dynamic is supportive for Constellation’s merchant-plus-contracted cash flows, but it also creates a scarcity premium across the small universe of nuclear, gas-peaker, and transmission-heavy assets that can offer 24/7 load service.

For Microsoft, the strategic value is not just megawatts; it is securing a moat around AI uptime and latency, which reduces the risk of model-training bottlenecks and gives it more flexibility versus peers forced into smaller, more expensive power deals. The hidden cost is duration risk: if AI power needs plateau or shift toward more efficient inference architectures before the plant fully returns, the company could be locked into above-market economics for a long-dated obligation. Over a 6-18 month horizon, however, the market is likely to reward any hyperscaler that can prove it has locked reliable power ahead of competitors.

The main bear case on Constellation is execution: nuclear restarts are binary, politically sensitive, and vulnerable to schedule slippage, permitting changes, or any safety-related headline. That means the best risk/reward is probably not a naked directional bet on the project itself, but on the broader theme that dependable baseload will command a premium as AI load growth tightens reserve margins. A contrarian takeaway is that this could ultimately be more bullish for regulated utilities and gas infrastructure than for the more narrative-driven renewables complex, because the market is finally paying for firm capacity, not just decarbonization optics.

Key Decisions for Investors

• Long CEG vs. XLU for a 6-12 month relative-value trade: expect CEG to outperform regulated utilities as the market increasingly values merchant baseload optionality and AI-linked power scarcity; risk is restart delays or a broad de-rating of power multiples.
• Add MSFT on any post-rally consolidation, 3-6 month horizon: the equity can absorb long-dated power commitments because the strategic payoff is lower AI downtime risk and better competitive positioning versus peers chasing spot capacity; downside is limited unless AI capex is re-rated more aggressively.
• Pair long CEG / short ICLN or TAN for a 6-18 month thesis: firm power should outperform headline renewable exposure if the market keeps prioritizing dispatchability over pure clean-energy growth; stop-loss on any policy-driven green subsidy shock.