Learning Curve Exceptions

Learning curves are real, but not universal. Two conditions can invert or stall them: complexity escalation and model anchoring. Understanding which condition applies explains why nuclear costs rose while solar costs collapsed.

Solar PV and nuclear power are both low-carbon electricity sources that scaled up in the late 20th century. Solar followed a steep learning curve: each doubling of cumulative capacity cut costs by roughly 24%. Nuclear went the other direction.

If both technologies were being built at scale, why would one get cheaper and the other get more expensive?

Solar is modular. A factory making 1 GW of panels runs the same process as one making 1 MW, just repeated. Yield improvements, cheaper silicon, and manufacturing automation compound across every unit. No regulatory event resets the cost clock.

Western nuclear ran into the opposite dynamic. Three Mile Island (1979) triggered retroactive safety retrofits on plants already under construction. Standardized designs became individualized. Requirements grew mid-build. Construction stretched from 5 years to 12. Costs roughly tripled in real terms between early and late 1970s builds. This is negative learning: accumulated experience drove costs up because each new unit incorporated more required changes than the last.

Forecasters missed the solar story for structural reasons. The IEA anchored on slow historical energy technology curves and used engineering cost models that could not capture Chinese manufacturing scale effects. When reality diverged, modelers adjusted slowly.

The policy implication: learning curves justify deployment subsidies only if the technology architecture supports learning. Subsidizing a complex, custom-engineered technology without standardization may not produce expected cost reductions.

---