> Today, late optimization is just as bad as premature optimization, if not more so.

I agree but I don't think this discredits the "premature optimisation is the root of all evil" thing, aside from the fact that it's a heavy exaggeration.

The trouble is, people read it as "don't optimise" which is an incredibly bad decision.

Especially in the data world though, I've seen lots of teams really struggle with problems caused by using technologies they don't need (normally spark or kubernetes or both) just because they might need them later.

I think that type of pitfall is what the original quote is warning against.