The science education at the University of Oslo puts much emphasis on learning other languages, in particular MATLAB, R, C, C++, and Fortran. We teach (a subset of) Python in a way that eases the transition to and from MATLAB. For high performance we emphasize reusing Fortran and C libraries from Python or migrating slow Python code to compiled languages. Our experience shows that users who gain a good knowledge of Python tend to prefer that language and do as much as possible in Python before the quest for high performance demands using compiled languages. It seems that many students and professors, when their knowledge of Python is at the level of their MATLAB skills, steadily drift to do most of their work in Python. The numerical functionality in Python is not superior to MATLAB, so this drift is more rooted in Python's clear syntax, powerful software engineering tools, easy integration with Fortran and C/C++, and rich set of libraries for non-numerical tasks.
The described trend among students and professors are also found in industry: Python steadily eats of the MATLAB market because people find the language stronger and more convenient. Our efforts in teaching and professional use of Python contribute to the observed exponential growth.