However, the economies of scale to work out the chip, make prototypes, and put it on a board are such that only from a certain level of performance-over-CPU this becomes interesting. So on chains that will not go to the moon, CPU/GPU mining may hold out with a sufficiently sophisticated hash function. From the moment you hope to go to very high prices, and hence, very high block rewards, this investment will always become useful.
Point is, however, why try to get consensus over that with bitcoin ? If you want that, there are several alt coins *doing that already*.
Well, there's quadrillions of SHA256 hashes already protecting Bitcoin's holders, that's one reason.
The altcoins *doing that already* don't have the breadth and depth of developer talent that Bitcoin has. 2 major developers at Bitcoin Core (Peter Todd and Luke Dashjr) have already expressed support for a PoW change, when you look at the present situation it's not a case of if, but when a BU hardfork happens. More Bitcoin devs will follow their lead.
The other reason I've already stated; miners have started to try to push the users around. Segwit ready nodes are the majority of the Bitcoin network, have been for several weeks and it continues to rise.