Plenty of examples of this in practice (where Engineering guesses at cloud configuration then never revisits).
The wrong CPU - 64 hypercores when 1 or 2 would do, or a single-core cheapie when a reserved larger CPU is appropriate etc.
The wrong support - AES support missing causing massive slowdowns.
The wrong connectivity - slow links between layers instead of co-located.
The wrong location - cheap US location instead of Hong Kong or someplace, causing massive latency issues
I've been at companies that did it all. Spending $X10 what a better configuration would have cost.
