Ten years ago, Python used to be excellent as a second choice for everything. So if you dabbled in machine learning, web, statistical analysis, text processing, system scripting, computer algebra, etc, it made sense to learn some Python even if your main job was in something else.

So it grew a critical mass of people who know Python but not for the sake of Python itself, but for something else. And suddenly these people are invested. They want more, better, and faster. And then the demand pulls the proposition, and a decade later, Python is a big thing.

I work in applied geometry. Numpy is good. Pandas is of little relevance for what I do. But what became a gamechanger is Numba. It's a run-time compiler that brings Python into native code for the target architecture, Inter, ARM, or even NVidia GPGPU.

We used to do research in Python, invent algorithms there, and then rewrite them in C++ "for performance". The last time we did that, C++ version became 3 times as slow as the Python one when translated literally. We hid the fact by rewriting the algorithm itself, but that was the last straw. We now run Python in the cloud directly, and it rebuilds the native code for every piece of hardware we procure for the task automatically and I think this is beautiful.