I don't know if you can point to a single feature, a lot of what makes a language successful is around context and placement (javascript being a great example of this).
I think there's a few key features for python that have definitely been a big help for it:
It's highly dynamic meaning you can get running with it before understanding every detail (no compile step, no static typing etc).
It may not be the best at everything, but is a decent choice for most things, making a good generic intro language for programming.
It has good interop with low level languages - this is a big deal for data science where a less technical (from a CS perspective) user still needs the performance of a language like C/Fortran/Rust/etc. Python's pattern of "use python as a high level api for libraries in those other languages" (as in numpy, pandas, polars, pytorch etc).
That last one has been a big deal recently, python has become the de-facto language for data science as a result of this.
I think there's a few key features for python that have definitely been a big help for it:
It's highly dynamic meaning you can get running with it before understanding every detail (no compile step, no static typing etc).
It may not be the best at everything, but is a decent choice for most things, making a good generic intro language for programming.
It has good interop with low level languages - this is a big deal for data science where a less technical (from a CS perspective) user still needs the performance of a language like C/Fortran/Rust/etc. Python's pattern of "use python as a high level api for libraries in those other languages" (as in numpy, pandas, polars, pytorch etc).
That last one has been a big deal recently, python has become the de-facto language for data science as a result of this.