This post was originally a little too concise. (I posted it from my Google Pixel 2.) The referenced papers are Grubler (2010), Boccard (2014), and Escobar-Rangel and Lévêque, as well as a slide presentation by them.

In addition, there is new and important research about why nuclear plants have negative learning curves.

First, there is a rebuttal to a critique of the 2010 learning curve costs work, which of course also cites the critique, so you can find it there.

Second, the work by by Escobar-Rangel and Lévêque is especially important, because it is based upon an update of their data, and their working paper and presentation offers a statistically justified explanation of why the negative learning curve. In short, there are two explanations. First, because procuring utilities are rarely the engineering firms that build a nuclear power plant, the firms that do build them as cost-plus jobs. This means each one is different. Second, to improve margins and take advantage of lessons learned, the engineering firms that build reactors have proposed bigger reactors over time, with more safety and other features. More complex jobs are inherently more risky in terms of cost and completion time.

The implicit criticism is that nuclear power reactor procurement should have pursued developing a modular product which could be replicated, and achieved scale by adding a number of the units together. To the degree they did not do this, the lessons of the learning curve were squandered early in design rather than being realized for end customers. Moreover, it is possible that any procurement with a high price tags suffers this phenomenon: It was see on the B-2 bomber procurement and is seen on the large nuclear submarines with missile-launching capabilities. While these are supposed to be identical vehicles, they are not, because of their staggered delivery and the shortcuts taken to meet delivery deadlines.

Consequently, the conclusion is that the reason why nuclear power does not see the advantages seen especially by renewables is that the units for renewables are essentially commodities, and are replicated in large numbers. This appears to be true of some fossil-fuel-based plants as well:

It is critically important for units of production to be produced in cookie-cutter fashion.