> wouldn't recommend Python for applications that are (or are likely to become over time) compute intensive (unless you are willing to move the speed bottlenecks into C)
The compute intensive parts of AI already always happen in CUDA/C++/C.
The difference is that Lisp doesn't require a two language syndrome.
Julia, Mojo, XLA, Triton,... are picking up speed, and the pressure for a JIT on CPython is increasing from Microsoft and Facebook, exactly because not everything is AI, and not everyone wants to write C++, C to speed up Python.
Ok the other hand, a good lisp implementation of neural nets would offer a specific set of macros dedicated to matrix operations. This set of macros is, in the lisp point of view, a domain specific language. So, you still have 2 languages.
The compute intensive parts of AI already always happen in CUDA/C++/C.