I don't think this is true. It's just that there aren't many optimization practitioners and the OR community has done a terrible job promoting itself. The value in scheduling and logistics is well established. So optimizers tend to congregate. But optimization could be applied to lots of other domains successfully. ML is just minimizing a loss function. I see simulation and post-hoc analyses being applied in numerous situations where optimization would be more efficient and produce a guaranteed optimal result. RL on simple, static problems is really an MDP but I've seen organizations fail to realize this. Etc.