Laws of physics apply always, it's just a matter of cost. As you go to smaller geometries it gets more and more expensive. One of the things that was driving Moore’s Law is that the cost per transistor was dropping. It hasn’t been dropping noticeably recently (going below 5nm), and in some cases it’s going flat. So yes, you can still get more transistors and lower wattage at smaller nodes, but the cost per die is going up significantly, so those two things balance out. I'd say 3nm is a sweet-spot for compute hardware for now, because it is great for compute density and relatively low leakage power. Below that we probably won't see retail GPUs and CPU shrinkage anytime soon as arch become very, very complex which means A LOT of R&D $$$ and abysmal die yields. But hey we're still talking about Samsung's 8nm here (aka 10nm in reality) and 7nm with Ampere/RDNA2 not below 3nm nodes, so there is still plenty power efficiency to gain simply by moving to smaller node. The problem Ampere has is that it was not build exclusively for gaming and Samsung's 8nm node that was never meant for big dies and it shows. It's Nvidia's "GCN 5 Vega" moment, trying to sit on 2 chairs at the same time & cheat on cheap inferior node. Luckily for them AMD is so far behind that they can pull it of without having to worry about competition too much. 3080 on TSMC 7nm euv process would be 250W TPD GPU, that's all NVidia had to do to obliterate RDNA2, but they've chosen profit margins over efficiency and maybe, just maybe that will bite them in the ass.
.
