AI’s power gap: why nuclear became 2026’s hardest second-order AI theme
Data centers are racing toward 12% of U.S. electricity (from ~4.4% at end-2023); solar/wind intermittency can’t meet AI’s baseload need, so nuclear is being repriced. Microsoft, Amazon, Google, and Meta have signed nuclear PPAs; uranium holds near $86/lb. We view this hardest second-order theme through a long-term, consensus lens.
Compute ends at electricity
Data centers are projected to move from ~4.4% of U.S. electricity at end-2023 toward ~12%; U.S. data-center demand is seen rising from 19 GW in 2023 to 35 GW by 2030. Grid-interconnection queues stretch eight years, wholesale power spiked as much as 267% near hyperscale sites, and nearly half of planned AI data centers face delays from transmission and transformer shortages — power has gone from supporting act to hard constraint.
Why nuclear
Solar and wind have a fatal flaw for AI: intermittency — a data center can’t go dark when the wind stops. AI needs 24/7 baseload power, which nuclear fits. Across 2025–2026, Microsoft, Amazon, Google, and Meta signed or announced nuclear power-purchase agreements; uranium holds near $86/lb on AI-driven demand. The nuclear investment narrative was almost completely rewritten this year.
A second-order theme, better for patient capital
Nuclear/energy is a classic second-order AI beneficiary: it doesn’t sell AI but is underpinned by AI’s power demand. Slower than chasing chips, but sturdier — suited to patient capital. The site groups it under the “AI energy & power” layer; for the cloud angle, the consensus data also surfaces the Microsoft/Amazon/Alphabet names that “buy both AI and power.”
Sources: The Motley Fool — AI is creating a nuclear power renaissance (2026) · MarketWise — Best nuclear energy stocks as AI drains the grid · Yahoo Finance — Nuclear power stocks set to flourish on AI data-center boom