In my research group specializing in computational materials science and thermodynamics, I've been advocating for using Nim roughly once every two weeks with some success, as we have already reimplemented two Python packages we use internally.
Some of my colleagues who almost exclusively use Python (like 90+% of my field since around 2018) and saw Nim for the first time were able to explain right away what it is doing without any docstrings and make simple modifications. That is until they reached some compile-time metaprogramming. We also mentioned it in research proposals, highlighting the combination of performance and syntax readability to the general scientific audience. In a classroom setting, we have yet to start using it; however, I've already convinced my boss to utilize a Nim package I wrote (not public yet) for a course he teaches next Fall (for ~3/4rd year PhDs). It will replace two-lectures-worth of a larger FORTRAN90 codebase, introducing genetic algorithms for solid-state physics Monte Carlo. Then, throughout the semester, students will be given an opportunity to rewrite the remaining FORTRAN parts in Nim in lieu of homework assignments requiring them to modify the original.
