Flamanville, at 10.5 billion euro for 1.6 GW is not nearly as good as watts bar 2. This has had serious delays due to quality issues detected by its regulator. As had the finish and Chinese units build to the same design.

Sure these are supposed by GenIII, but are currently not competitive in any way with other power sources in economic terms. 10.5 billion buys you about the same in offshore wind e.g. 3 london array installations ( 2 billion per 600 MW * 50% capacity, compared to 10.5 billion per 1600mw * 90% capacity) = 10/(1.6*.9). Big difference is london arrays take 1 year to build instead of 10 for nuclear. That alone gives a mayor difference in real project costs.

There's not much surprise if the first version of a new design runs over budget; I'm interested in the costs of routine, serial production once those unknowns have been worked through. What about the 59 reactors already in operation, including Flamanville 1 and 2?

That is really hard to determine, mostly because project costs are part of the greater EdF numbers and not broken down in details for those years. So someone with better French and access should be able to answer you there. This is true not just for nuclear but all power generation before the late 90's in europe. i.e. same difficulty to getting numbers in the UK and Germany. Once these national power companies went to the stock exchanges they started to report in more detail how much projects cost. Anyway there are also number of existing reactors that are no longer running even if in the same generation as some other ones.

The, unfortunate, reality is nuclear reactors are not build in serial production. Therefore learning costs do not decrease as much as you might expect. Main reasons are distance between sites, meaning many construction workers do not work on all sites. You see at most 10 units for one design. Time between forgings for mayor means parts are not made in large series so relatively little is learned. All 4 EPR sites seem to have suffered from the same issues with concrete pouring so even there learning might not have happened.

Economically nuclear is in a real bind: for regulatory reasons you want big plants (1gw+), but for economic reasons you want many small ones (about 100mw). The current 1.6GW units are in many ways to big and smaller reactors would be much better project/knowledge wise. Yet the regulatory overhead for each new unit means that the firms went towards designing larger units.

Bechtel marine nuclear powerplants could actually be produced in series as they are small enough to be build off site and could be made economically viable. Yet I don't expect that Bechtel ever will.

But lest assume Arreva learns how to build EPRs for 5 billion and in 5 years. Which gives you

EPR done right 5 billion / (1.6 GW * .7 historic capacity factor) = 4.4 to 5 billion / (1.6 GW * .9 best capacity factor) = 3.2 London array 2 billion / (.63 GW * .50 historic capacity factor) = 6/2

Even in this very optimistic scenario, offshore wind is between 100% and 60% more expensive on construction costs. Offshore wind however, still has lower opex and a much better learning curve. i.e. if ordering a london array today costs will be less than 2 billion. (big difference in reporting is that wind projects include interconnect costs which current nuclear projects do not due to sitting next to existing plants).

Wind farms, also have positive decommissioning costs (those steel towers have real economic value at end of life).

Both types of projects have issues with nimbys.