I think you're right in the sense that they make it easy to have compute units in different geographic/availability zones, which they make much easier/cheaper than doing yourself.
But it isn't something built-in with regard to fail-over handling, etc.