Knowing little about how the FSD chip works, is the fact that it runs locally and not on some super cluster a bottleneck in performance right now?

The fact that it's only in the last year or so that Stability/OpenAI and efficient hardware like before mentioned processors that can produce relatively fast and accurate results in the NN sphere makes me wonder if it's just "too early" to have self driving cars?

ChatGPT still requires a supercluster to run and as a layman one would think video analysis with near zero latency would be even more demanding.

Also how does it compare to to apples newer Arm based architecture and Neural Engine which as far as i know is a good example of in-house designing (not production off course).