The real reason python won in my opinion is it has been a default install on most linux distros for decades. A ton of the groundbreaking ML research (especially in the 2000s and 2010s) was done in government labs and DoD settings where it is extremely difficult to get external software approved/installed, yet there was python sitting there already installed on every single DoD Red Hat machine. So a lot of research code ended up getting written in python as a prototype and then improved with C++ in the speed-sensitive bits, and before you know it, Python = ML.
Being a default install also has a lot of barrier-to-entry advantages in other settings, particularly in academia where there are similar (but usually not nearly as strict) bureaucratic restrictions on installing things, etc.
source: was a DoD computer science researcher focusing on ML for several years and witnessed people reach for python because they were still waiting for months, sometimes years for whatever software package or language to be installed / approved by security/IT and had no other dev tools
Being a default install also has a lot of barrier-to-entry advantages in other settings, particularly in academia where there are similar (but usually not nearly as strict) bureaucratic restrictions on installing things, etc.
source: was a DoD computer science researcher focusing on ML for several years and witnessed people reach for python because they were still waiting for months, sometimes years for whatever software package or language to be installed / approved by security/IT and had no other dev tools