Quite a lot of ETL ends up being some minor transforms + a query or two. Not all of it is massive ML models doing a lot of computation, and I've had a lot of s

Quite a lot of ETL ends up being some minor transforms + a query or two. Not all of it is massive ML models doing a lot of computation, and I've had a lot of success using pandas and numpy in it (and gcp cloud functions). Serverless has its niche and is a great little tool to smooth the impedance mismatch between data stores.