AI is energy hungry. Training one large model uses as much electricity as 100 American homes use in a year, and Microsoft, Amazon and Google have signed nuclear deals to feed their data centres. This piece asks what nuclear really offers.
The cost paradox
Nuclear is expensive to build and cheap to run. Roughly 60 to 70% of its lifetime cost goes on capital, against 25 to 30% for coal and gas, but fuel is a small and stable cost.
Why small modular reactors
- Lower capital cost and modular build, which makes the economics more workable.
- Flexible deployment, a good fit for tech companies’ sustainability goals and site needs.
- Far lower emissions, about 98.5% less than coal-fired power for AI operations.
I also look at how reactor generations have evolved and why countries such as France and China have made different choices. The open questions are infrastructure, grid readiness, policy, waste and public acceptance. Whether nuclear can supply clean, reliable power is settled. How well we handle the rest is not.