So... geometry is now a solved problem? What next?
Yes, I honestly, truly think this is the future of rendering. I don't think it's a gimmick, or just a demo turned to max. As far as I can tell, this is real.
Once you have a triangle per pixel, where can you go?
First, performance. From a baseline of 1440p30, DLSS should take us to 4k30, and a 4x GPU performance boost should get us to 4k120. That's that done.
Then, well, I guess you have to find other stuff to do. Full ray tracing, subsurface scattering, better physics, contact deformation. There are still branches to pick. But, really, what would we even do with 200 teraflops of compute? I guess that's for the future to figure out.
Oh, and a quick point about VRAM: it's clear we have enough now. SSDs remove a huge amount of memory pressure, so what we've got seems enough for universal photorealism. Sure, ray tracing or other techniques might turn out to be memory hungry enough to push at the edges, but we're talking maybe factors of two, not 10.