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