>Python is absolutely terrible to use as a stand alone language.
Follows from the others.
>Almost everywhere it is used (ML, Data Science, Scientific computing, etc.) it is used to call C or C++.
See the big libraries in those fields. Do you think tensorflow or numpy are native python. If you are using python for scientific computing you will be using libraries which call out to native code.
>The amount of serious Python software that doesn't makes extensive use of calling to other languages is very small.
Given the above and that Python is most prominently found in those fields it seems pretty certain. There is standalone Python software, but usually only as a replacement for shell scripts or for relatively small applications.
>Python alone has very severe shortcomings
To name a few: GIL, lack of common default types (e.g. C like integers), enormous interpreter overhead, no mandatory typing (which introduces a whole range of runtime errors impossible in strictly typed languages). These alone make it totally unusable for many applications, unless you call to an external library.
>that make it totally unsuited for stand alone development.
See above.
>Python is absurdly slow, for naive array task you can easily get a 100x to 1000x speedup compared to any native language.
Personal experience.
>The GIL doesn't really help either...
It really does not...
>Right now AI is big, Python is at the center of that, to glue C++ together...
See e.g. what numpy, tensorflow, pytorch are written in. Spoiler, the matrix multiplication isn't implemented in python.