Nuclear power's data-center revival — is this time different?
Two years ago the nuclear conversation was mostly theoretical: aging fleets, massive cost overruns at new builds (Vogtle came in $17B over budget, Hinkley Point C in the UK is running at roughly triple its original estimate), and a general sense the industry needed heroic policy support to survive.
Now it looks genuinely different. Microsoft signed a 20-year PPA with Constellation for Three Mile Island's restart. Google has a deal with Kairos Power for SMRs starting around 2030. Amazon bought a data campus co-located with a nuclear plant in Pennsylvania. The buyers aren't utilities or governments — they're hyperscalers trying to source 24/7 carbon-free power for AI compute that runs continuously and can't be balanced with intermittent renewables.
The economics have changed because the demand signal has changed. AI workloads are uniquely ill-suited to intermittent supply — you'd need grid-scale battery storage that isn't there yet, or an always-on baseload source. Nuclear is the only always-on carbon-free option at meaningful scale. Hyperscalers will pay well above market rate for that guarantee, which unlocks project finance that couldn't work when the buyer was a regulated utility absorbing merchant risk.
The skeptical case is also real: SMR timelines have been "10 years away" for about 20 years. Vogtle was supposed to be a replicable template for new US nuclear and it nearly broke the utility that built it. First-of-a-kind projects in nuclear are chronically, systematically underestimated — and no SMR design has actually been built at commercial scale yet.
So: does the corporate PPA market change anything structural, or does it just fund a different flavor of the same project-management failures?
0 replies