To be fair, it is my impression that even in data science, python is more a less a scripting language, and all the libraries which do the heavy lifting computing-wise are written in something else, because speed. Has this changed?
Sure, but for every line of code in such a library there's a hundred (or hundred thousand) lines of code that use that library. A small bunch of people maintain some CUDA code that's underneath a Python wrapper, but great many people write code that uses it.
The underlying libraries may be in C, but the vast majority of data scientists never need to alter the C so it's fair to say most data science code is written in python. Python/Numpy/pandas has enabled data scientists to avoid writing C entirely.