IIUC, the performance of PyPy is still quite far behind normal compiled languages, due to limitations inherent in Python. And then it has limitations in its interop with Numpy and C libraries, at least last I checked.

That's the general picture when trying to overcome Python's performance wall: Sometimes you can use Cython, or Numba, or PyPy or Numpy vectorization, or call into C. But only sometimes, and they each come with their own set of awful restrictions and caveats.

It's such a breath of fresh air to switch to a fast language and just forget all those hacks and workarounds. I get why people just migrate to static languages