First, the obvious that if a massive surge happens (attack, bug, ...) on prem stop serving, aws surprise-drain your bank account. Yes it's possible to have fail

First, the obvious that if a massive surge happens (attack, bug, ...) on prem stop serving, aws surprise-drain your bank account. Yes it's possible to have fail safe to avoid it, it's also very common to not have them or that they're not tested or working. Second, the "in-between" costslike bandwidth from service to service, iops on your disk, ... They mostly don't exist on prem, you either have enough of it and it works or you don't and it doesn't work and you need to go buy a bigger pipe. They're one of the biggest source of surprise cost on AWS. Eg take a basic data storing application, the cpu and storage space are easy to estimate and are the same as on prem, but suddenly you also need to estimate your iops, and you realize you have no idea about it. Do you have an on-prem database running somewhere ? How many iops does it need ? Unless you've had to figure it out, you have no idea and will never know on prem.