[biztos]

Merely anecdotal but: many people interested in machine learning start with Python; many people wanting to do "serverless" things choose Python; many younger devs find Python a nice place to run away to from Java; and all this together means it's pretty easy to find devs who either know Python or are quite happy to learn it; and that in turn makes it a "safe" language choice when you aren't otherwise committed to the features of any particular language. (And running in Lambda can be a pretty big deal.)

It's a virtuous/vicious circle, depending on what you think of Python. But I think "Python is for ML" is the unofficial motivation behind a lot of it, because if you're starting to build your career and Python is an option then for sure you're going to mess around with scikit-learn and put that on your CV.

[edit]: Oh and boto3 probably accounts for a fair amount of Python use these days. I actually prefer the AWS support in golang but boto3 usually "just works" despite some very weird quirks.