The Nuclear Renaissance Question
The Nuclear Renaissance Question
AI’s energy appetite is enormous and growing. Major tech companies are signing deals for nuclear power: small modular reactors, extending the life of existing plants, even restarting shuttered facilities.
This represents a significant shift. After decades of nuclear decline (cost overruns, safety concerns, waste issues, public opposition), AI may be driving a nuclear renaissance.
Is this good?
The Case For
Climate: Nuclear is carbon-free once operating. It provides reliable baseload power that renewables struggle to match.
Density: Nuclear delivers enormous energy from small footprints — unlike solar/wind farms requiring vast land areas.
Reliability: AI data centers need consistent power. Nuclear provides 24/7 availability without intermittency.
Additionality: Tech companies building new nuclear adds clean energy to the grid rather than competing for existing renewable capacity.
Innovation: Investment drives technology improvement. Small modular reactors may solve problems that plagued large plants.
The Case Against
Time: Nuclear takes decades to build. AI’s energy needs are immediate. This timeline mismatch undermines the rationale.
Cost: Nuclear consistently runs over budget. Cost overruns may ultimately fall on ratepayers or taxpayers, not tech companies.
Waste: Nuclear waste remains dangerous for thousands of years. No permanent repository exists. AI-driven expansion creates more waste with no disposal solution.
Safety: While modern designs are safer, accidents remain possible. The consequences are severe and long-lasting.
Proliferation: Expanding nuclear infrastructure creates more opportunities for weapons material diversion.
Opportunity cost: Resources spent on nuclear could fund renewables and storage, which may ultimately be cheaper and safer.
The Uncomfortable Alignment
AI companies want reliable, carbon-free power. Nuclear provides it. But:
- Tech companies benefit from nuclear’s upside
- Society bears nuclear’s risks and long-term costs
- The timeline for nuclear doesn’t match AI’s growth curve
- “Carbon-free” may justify ignoring other nuclear problems
There’s a risk that AI’s energy needs launder nuclear’s reputation without resolving its genuine concerns.
The Scale Question
Even with nuclear renaissance, can supply meet demand?
- AI energy demand is growing exponentially
- Nuclear takes 10+ years to bring online
- Existing plants are aging
- The gap between AI growth and nuclear capacity is large
Nuclear may be part of the answer without being sufficient. The question is what else fills the gap.
Living with the Tension
Like The Irony of AI for Climate, this tension resists clean resolution:
- Nuclear is neither simply good nor simply bad
- AI’s energy needs are real and growing
- The alternatives all have problems too
- Decisions must be made under uncertainty
The question isn’t “is nuclear good?” but “is nuclear better than the alternatives for AI’s specific needs, accounting for all costs and risks?”
Open Questions
- Should AI companies bear long-term nuclear costs and risks?
- Is nuclear’s timeline compatible with AI’s growth?
- What role should public input play in AI-driven energy decisions?
- Can small modular reactors actually deliver on their promises?
See Also
- Training vs Inference Footprint — the scale of AI’s energy demand
- The Irony of AI for Climate — the broader climate accounting question
- Stranded Assets Risk — what if nuclear investments don’t work out
- Dependency Lock-in — committing to nuclear creates long-term dependencies
- Embodied Carbon — nuclear plants carry massive embodied carbon in concrete, steel, and reactor components
- Geographic Inequality of Compute — nuclear siting replicates the same pattern: local communities bear risk while distant users benefit
- The One More Query Problem — aggregate demand is what makes nuclear “necessary” in the first place