But it looks like that attitude has finally caught up with them, especially since they are down to just 3 technical staff (from 5 last week), and two of them are brand new.
I tend to try to find at least 1 good local/regional datacenter for a good portion of a server stack. There is huge benefits (IMO) in being able to drive to your servers and have a face to face meeting if there are issues and/or take your toys and go someplace else if there is a massive outage. If you're in Green Bay, options might be limited, but in any semi-major metropolitan area there are usually enough datacenter options that you have multiple choices.
That being said, there are situations where you may truly need single-node performance that isn't available on AWS.
1. No traction? Just put it anywhere, 'cause frankly, it doesn't matter. Cheapest reputable VPS possible. Let's say, Linode.
2. Scaling out, high concurrency and rapid growth? DEDICATED hardware from a QUALITY service provider--use rackspace, softlayer et al. Have them rack the servers for you and you'll still get ~3 hour turnarounds on new server orders. That's plenty fast for most kinds of growth. No inventory to deal with, and with deployment automation you're really not doing much "sysadmin-y" work or requiring full timers that know what Cisco switch to buy.
3. Technology megacorp, top-100 site? Staff up on hardcore net admin and sysadmin types, colocate first, and eventually, take control of/design the entire datacenter.
I simply don't understand why so many of these high-traffic services continue to rely on VPSes for phase 2 instead of managed or unmanaged dedicated hosting. The price/concurrent user is competitive or cheaper for bare metal. Most critically, it's insanely hard to predictably scale out database systems with high write loads when you have unpredictable virtualized (or even networked) I/O performance on your nodes.