Re: i bet
I understand why you might say that, but take a look at some of the ReInvent videos from last year and I'm not sure that's true. One customer did a big preso on how they saved costs by ensuring they used an appropriate instance type. In my experience, much like in the days of virtualisation, folks hopelessly overestimate how much compute they thought they needed.
Anyway, the gist of this preso was that they'd rolled out a bunch of larger instances based on the assumption that the steady state compute would be pretty high. They experimented with T2 instances (bursty workloads) and found they were a lot more appropriate. This saved them something like 60%/70% per month on compute, which is generally the lion's share of the bill.
Properly monitoring what you roll out and sizing it appropriately should not be overlooked once you're live. If you remember it's a living thing that still needs care and attention (regardless of what cloud providers say) then that's half the battle.