Why nuclear power became the AI data center trade
AI data centers have a power requirement that is uniquely suited to nuclear: they run 24 hours a day, 365 days a year, cannot tolerate outages, consume gigawatts of electricity, and operate under corporate sustainability mandates that make coal or gas politically difficult for hyperscalers to sign long-term contracts with. Nuclear capacity factors above 90% — meaning the plant runs at full output 90% of the time — match perfectly with data center demand profiles. When Microsoft, Google, and Amazon began signing nuclear-specific power purchase agreements (PPAs) in 2024-2025, Constellation, as the largest US nuclear operator, was the most logical counterparty.
Constellation's $16.4 billion acquisition of Calpine in early 2026 expanded its fleet to 60 gigawatts of generating capacity, adding natural gas peakers that complement nuclear baseload with dispatchable capacity. The combined entity can offer hyperscalers a full power solution: nuclear for always-on baseload, gas for peak demand response. Management projects the Calpine deal to be 20% accretive to earnings per share by end of 2026 and guided to 20%+ EPS growth annually through 2029.
- Nuclear capacity factors above 90% match AI data center demand profiles better than solar, wind, or even gas peakers.
- Hyperscaler corporate PPAs with Constellation lock in multi-year revenue at contracted prices, reducing earnings volatility.
- The Calpine acquisition added 60GW of combined capacity — scale that creates a moat against smaller nuclear competitors.