I think a better question is why are they pushing the power that high that they need so exotic solutions? They are way past any reasonable efficiency curve, it's fun to have the greatest halo product of all time (for now) but there's a point where is it really worth anything?
We already saw and will continue to see this with every "flagship" product, like the 4090 at 450w, drop that to 270w (-60%, no undervolt either, just the power level) and you're only loosing a measly 8% in performance
The 4090 is already a lot more efficient per frame then the likes of the 3090ti, now that was a power hog.
As to why they are pushed so hard? Well, you have yields, with the price of wafers being so much higher they'll want every last piece of functional silicon you can find. Those undervolt and underclock settings are not universal, while each GPU can undervolt the level they can reach will very, just like every generation, and nvidia likely doesnt see a reason to reduce yields to lower stock usage seeing as anyone who wants to do that can do so on their own time.
There's simply no reason for them to do so. Also for dies other then the 102, any die that undervolts really well will be used for the more expensive mobile parts that operate in limited TDP spaces. That's always been true. For AD102 anything that undervolts well will end up as an A series GPU that sells for 5 figures.
I know, I figure that is what they are trying to avoid (Pump issues, leaks, etc) even though they have gotten much better. I just mean at this size (Full Quad slot) you are getting to the point you might as well consider it because you are going to need a tower that is pretty huge to support a card like this so its most likely going to have multiple radiator mount points. But I get wanting to stay on air cooling for the ease/reliability.
Based on sales from amazon, reviews for newegg, ece the ATX form factor is still the most popular, and most modern ATX cases will take a 4090 with no problem. Even the mini ITX space has plenty of cases for 3 slot cards, genuine small ITX cases are hard to find.
But you have to think about it. I mean if this is what Nvidia is going with, then this card uses a massive amount of power and this cooler is required to keep noise levels in check (Probably could run on the already massive RTX 4090 reference cooler but the fans had to spin up to keep it in check).
NVIDIA has been overbuilding their coolers for a few years now, and frankly at this level consumers want outright performance, having a big heatsink is a literal non issue. Anyone who wants a small build with water cooling will be buying waterblocks that are far higher quality then any AIO solution that NVIDIA would ship, and more serviceable.
AIOs on GPUs are just a total waste of time.