A lot of conversation is about the cost of acquisition, but does not compare the prices of maintenance, hardware failure/outages, and recovery.
For example, with dedicated hardware and a guessed failure rate, how much does it cost, over time, to 1) replace the hardware, software, and configuration and 2) value lost (could be goodwill, could be $$$) due to outage.
With AWS using chef/puppet/salt/etc., it's trivial to boot up new nodes and down the malfunctioning ones. The value of that is super high.
You're acting like you can't do the same thing with dedicated hardware. On the post-install side, chef/puppet/salt/etc work just as well. On the setup/build side, dhcp, pxe, and the hands-free install setups built into every unix distro let you spin up and down machines with minimal hassle. Need a new db machine? Throw the MAC address of the new box in the db pool of your dhcp server and tell it to boot. Need to move that machine into the web server pool? Move that MAC into your web server pool and tell the box to reboot. Combine that with a mixture of hot spares, cold spares, and warm spares, and I'd still bet you'd be cheaper than AWS for any significant sized service.
A lot of conversation is about the cost of acquisition, but does not compare the prices of maintenance, hardware failure/outages, and recovery.
For example, with dedicated hardware and a guessed failure rate, how much does it cost, over time, to 1) replace the hardware, software, and configuration and 2) value lost (could be goodwill, could be $$$) due to outage.
With AWS using chef/puppet/salt/etc., it's trivial to boot up new nodes and down the malfunctioning ones. The value of that is super high.