There's little reason to expect that node development will get easier going forward - we're rapidly approaching the hard physical limitations of silicon, and new process nodes will increasingly need to deal with issues inherent with feature sizes just a handful of atoms large. And GPU makers seem to have settled on there not being any limits to GPU power consumption as long as it makes them competitive. Architectural improvements are our best bet for improvements going forward, but due to the aforementioned approach to product segmentation, that efficiency will likely be cashed in through pushing performance even higher. There's also the inherent thermal disadvantage of thin heatsinks and fan, especially the relatively cheap generic fans used by most GPU makers. It's easier and cheaper to just make a huge honking heatsink (which you can sell for more money, as big = good for a lot of people) than to research and develop tightly optimized thermal solutions. Brute force wins out.
The potential upside of this: not all GPUs are flagship SKUs. This means that there is room for more efficient SKUs lower down, or just undervolting and underclocking otherwise hot-running GPUs - assuming they are launched with 2-slot coolers. But sadly due to how an ATX case can house essentially as big a GPU as you want, and SFF still being relatively niche, I doubt we're getting strict 2-slot cards back any time soon.
