No one said they don't. Just there still are and always will be parts which are computationally heavy but written in python. e.g. pandas map is 300x slower than numpy map[1] for the same operation. But not everything could be vectorized and even if some complex function could be vectorized it will involve multiple passes over memory for the same array.
[1]: https://www.exxactcorp.com/blog/Deep-Learning/how-to-speed-u...