Given slowing growth of computing power, and that ray tracing algorithms are already mature and well studied, I don't think high-sample real-time raytracing will come along for a long time.

Instead, I think de-noising based on machine-learning will be the crucial factor for real-time raytracing, as Nvidia has been showing.

I think a coming advancement will be in dispatching rays based on ML too. That is, economically deciding where to cast rays, based on feedback from the de-noiser.