Clouds & Inverse Sunk Cost Arguments
Over the past couple of weeks I've had conversations with several folks about Gnip leveraging Amazon's Cloud services, and the fact that they are not.
The vast majority of the time it is clear that the other party is not using the cloud because they've already invested so much in their existing infrastructure, and that they're intellectually justifying sticking with their current solution. That said, if your back-end is mature, baked, and generally static, then the financial math likely does justify avoiding the Cloud, but if you're a relatively young application, the Cloud is your friend.
There are certainly times when someone truly has an application that requires physical hardware ownership/leasing, but they are the exception for sure.
Many times folks convince themselves, and their bosses, that they need their rigid, generally expensive, physical hardware (self, or third party, managed) infrastructure, because of their unique high performance needs. So few applications today require "on the iron" cross-machine latencies and processing speeds, and it's embarrassingly obvious at times; does your consumer facing website really need to be served from hardware you own? Don't kid yourself, it doesn't.
If you've already sunk mountains of money into leasing co-lo services, or buying and hosting hardware yourself, and staffing a team to manage them, acknowledging that such cost is now sunk, and that you could/should move your infrastructure into the cloud is a hard thing to do. However, the industry evolves, and it might be time for you to do so as well.
I'm clearly sold on "the Cloud." As a start-up, the flexibility in being able to setup/tear-down dozens of machine instances to support a bump in traffic, or to perform load tests, is a joyous thing. Tough conversations with a managed infrastructure provider about short/long term contracts, or the Operations team, are a thing of my past, and I'll never go back.
If you're avoiding "the Cloud" because you already spent so much on your infrastructure, do yourself a favor and re-evaluate.