I have managed a goodly number of AWS implementations and considerations ("should we do it or not?"). AWS was cheaper in most cases not because of bare metal vs AWS, but because of the sheer efficiency of scale, and not having to hire staff and expertise to manage all of the infrastructure.
Hmm... you just inspired a new post. Thanks, noelwelsh!
Once you get to a certain scale, you're hiring admins either way. The only question is whether you're hiring AWS admins or you're hiring bare metal sys|network|linux admins.
You definitely have to hire a lot less staff with AWS, ESPECIALLY when you're in that "in-between" phase where you're starting to build out a tech ops organization. You don't have those step functions where you have a huge need for some vendor-specific technology knowledge, but not enough work to justify a full time employee. AWS provides an abstraction layer that you need to have anyway: the difference is that you don't have to BUILD that layer, it was there from day one.
I have yet to find an AWS Infrastructure Admin job for under $120K/year (SF, LA, CHI, Boston, NY were the locales I looked in). I have seen quite a few traditional sysadmin jobs under $100K/year.
> You definitely have to hire a lot less staff with AWS,
This has traditionally been true of Linux admins in general. I'm not seeing that much of a difference in the marketplace between Linux admin and AWS admin compensation packages at the moment, except AWS admin positions are usually offered by startups and included equity you wouldn't get at a typical enterprise position.
Disclaimer: I do DevOps/AWS/Infrastructure/Etc.
AWS is a lot easier for developers to conceptualize as well - they're just logical objects that can be created/destroyed like programmatic objects can. Used to be that if you needed a new server, you'd have to talk to your sysadmin and get a VM provisioned, keys issued, etc. but with AWS all you have to do is issue a method to a class in your framework of choice (since AWS has SDK libraries for most languages/frameworks), and the VM is created, provisioned and integrated into your application like no work needed to happen.
I always found beyond a certain scale that AWS was not cheaper, but lately I am questioning even that. Next client who is considering going public cloud, I will need to redo my spreadsheets.
Remember computers in the hey-day of the PC? Every two years brought a huge tangible improvement. Programs made for new computers just wouldn't run on old PCs, or if they did would be unusably slow. That ended sometime around the release of the Core 2, except for niche users (gamers, developers, etc.)
It's the same thing for phones right now. A phone that is more than a few years old just feels like a piece of crap to use; it's slow, the screen is poor, and it doesn't support new features. Good times for manufacturers! (Though we are very rapidly getting to the end of this cycle, IMHO).
Same thing applies to AWS, brought up as an example in the post. The base price is not really the issue -- AWS prices fall slower than Moore's law and it is still more expensive that bare metal. It's the new toys you get every year that make AWS attractive. AWS keeps providing more and more pieces of your infrastructure at attractive price points with less admin overhead. At some point they'll have reimplemented everything the modern company needs and then they too will be subject to more serious competition.