And local models are getting more accurate and efficient. And they really don't need a dedicated GPU for fast response, just really fast ram. New APUs such as the AMD 395 are going to be game changers for local users with 256 GiB/s bandwidth.
The new Mac Studio is great for this job too. With the Max chip you get 128GBs of unified RAM for $3.5k. Memory bandwidth is at 546GB/s which is good enough for most tasks.