I've converged on "it's a bad idea to _start_ a nuclear project in 2026, but the sunk cost fallacy doesn't quite apply, so once significantly invested it should be finished and run to its operational lifetime".