Not "Best Practice" according to whom? Cloud vendors selling lock-in services? From an architectural perspective how is a bare VM at a remote data center worse than indeterminate/unknown intermedate layers of a proper "cloud" provider that is still in the same remote data center with the same hops and latency?

That "very small set of primitives" is enough to build the cloud vendors themselves so how exactly is that a limiting factor?

If I seem overtly hostile I don't mean it that way. I am just genuinely curious for an extended explanation. In it's current form it sounds down right marketing tautology to my ears.