An example of something that might be useful (or likely might not) of an idea that could have been done decades ago is my hobby project, the BitGrid.
The reason nobody has done it yet has to do with the fact that it's too far conceptually from everything that was done back then. The emphasis was on speed (1/latency), keeping the silicon area small, and utilization of the chips, in that order.
A BitGrid flips those assumptions, and would be a horrible way to build an FPGA. The idea of using cells of 100 transistors just to pass data would have been laughed out of the brainstorming meeting, and the person who brought it up derided.
However... one of the big problems with FPGAs is fitting designs into them, it can take days for a computer to figure out how to fit a design into an FPGA. Then there are timing validations that have to be run, etc.
The BitGrid uses the magic of graph coloring, and "slow is fast" to get around those things, focusing on ease of use, and aggregate throughput instead of latency.
It's likely a horrible idea... or it might be brilliant... without a real world trial, and the codebase to exercise it, we'll never know.
I suspect most world changing ideas are like this... such a big departure from "the way we've always done it", that nobody really wants to try them out for fear of wasting money, time, energy, and reputation.
Back in 1982, it would only have been someone with access to all of the resources required to get an ASIC made that could even attempt this. This might have been 1000 people.
Now it's a $150 bet using TinyTapeout for anyone to give it a go. ;-)