Sadly this is true. The learning curve of any new language can't be that bad anymore - it has to be able to do the Turing stuff, usually does some math, some typing, some scoping - it's just a question of how to pronounce that in this one.
The library/module ecosystem now, that's amazingly complex, different, variably maintained; which of the 10 things in this repository that claim to parse some markup is the best for my task, and imposes what other design constraints - callbacks vs blocking, how does it handle poorly formed input?.
That's where the real pain resides. Evidently someone wrote wrappers around some C/C++ junk for ML (no one writes math intensive stuff in an interpreted language if they're older than 10) - and they are popular. If they'd been written in some older language, or some other "Fad of the year" one, the same would be true in ML - that would be the hot thing. It has little to do with python itself.