I think it was Scott Wasson from AMD, a few years ago (like… 2016 or so?).
The rationale is that humans have a roughly 180 degree FOV horizontally (not sure what it is vertically), and you want to be able to slightly exceed that so the user can’t see the edges of the display.
Given the typical headset distance and FOV angle calculations (i.e., similar calc to how apple determine their “retina” rating - and despite the nay-sayers/haters it IS based on maths - essentially the minimum angle between pixels from your eyes that you can discern - for a typical person) i believe he worked out that 16k is the number that will provide this. Per eye, for stereoscopic 3d.
Sure, you can likely get really good results below that, but he was talking about the “end goal” where we have VR that is essentially indistinguishable from “real” vision.
So that’s what we’re shooting for eventually as an end-game, but for sure, things will be “much better” and even “really, really good” long before that point.
I’m not sure if he mentioned frame-rate, but i’d hazard a guess that would have to be at 100-200+ FPS as well to feel truly “real”.
So there’s definitely a lot of power we need on top of what we have today to get there.
But as with anything like this - as proven by the PC demoscene since the 90s… if you can fake it fast enough you can get away with much lower res and still “look good”.
My memory is hazy with the specifics a bit here so it may have been 16k instead of 16k per eye (but it was definitely 16k), but either way - the appetite for more GPU power for VR isn’t slowing down any time soon.