If I may
Before I started using pixinsight I was massively doing images in photoshop. Topaz AI was a go to solution but not just like that. I certainly took the time to either remove stars and topaz the starless image on nebulosities. Another technique was using masks to protect stars and focus topaz on whatever I wanted to denoise. For instance grainy background... but the again I was also using mask tools like tony kyper's tk7 luminosity mask tools.
Having integrated pixinsight into my workflows I now rend to do noise reduction on linear data early. I use many PI techniques to do so. Then when I get to stretched .. I flip in ps to finish and more often than I can think I will use topaz on a part of an image, a background or depending on the target I will mask and denoise... whatever I need to tweak at the end of the workflow.
What i never do is run topaz on the whole thing with stars. I most likely will be using the low light... I am also sometimes critical if topaz makes a plastic look when pushed too far.
Topaz is just a complex tool like all tools we use. It has learning curve but in my view it is worth using when used well.
as for can it be replaced by PI noise functions... not for me. I PI denoise early on linear... process the image, and flip in ps. topaz is always my end of process finishing touch if need be.
two noise reduction technique not the same place in the workflow. For me not a substitute but different processing.