I'm curious about AI denoiser too but the way its been marketed for games seems like its trained just for that game. In my feeble mind that means the AI learns what images looks like in just that game since the variations in a closed sandbox like Battlefield there are only so many. For ArchViz the content is never the same and seems like AI just couldn't work the same way its intended for games.
I also assume this based on an article recently about AI taking the text graphics off of images by Nvidia. The AI can both understand what text is and separate it from the image but the trick is understanding what lies under the image to fill it back in with the correct color pixels. I don't know if this was learned from seeing the same images with out text or just similar images.