It really depends on the project. My company is hosting multiple ML/AI projects, some with datascientist that are, as you said, glorified data analysts. Usually MBAs or mixed cursus, but also CS guys (my favourite clients as they will never tell you "i can't ssh onto my server" after executing `chmod -R 777 /etc/`).

And some with genuine DS/statisticians. Also the first kind of project almost always end up hiring statisticians in the end, so realistically, having "glorified data analysts" that can sell to the consortium or kickstart project is enough.