I've heard people say this, but I haven't seen a detailed cost breakdown. I know recently an automated testing company announced they were moving to a private cloud, but they spin up a lot of servers for short test runs, and the minimum one hour billing was hurting them.
I have an M1 large that I am using for some prototyping, and if I leave it running for a solid month it costs about $130. That's with.... 10GB I think of EBS persistent storage. But the upside is that I can run my stuff, then shut the instance down and billing stops.
My company did some analysis (that I don't know the details of) and we ended up getting dedicated servers for half of the price of what it would cost to get similar capacity on EC2. These dedicated servers are in a managed facility, so we're leasing them and don't have to maintain them ourselves. We can get a dedicated server setup in a few hours after the ticket is put in, so we can still add capacity pretty quickly if we need. Due to some weekly cycles, we do have some situations where we need lots of extra capacity for several hours a week and we use EC2 to temporarily provide that extra capacity.
I use DigitalOcean for my personal server. To get some numbers into this conversation, I'll compare it to Amazon EC2.
-----
Digital Oceans's smallest plan
0.5GB RAM, 1 core, 20GB SSD
$5/mo
The comparable EC2 plan seems to be the t1.micro.
0.6GB RAM, 1 core, no local storage
$15/mo on demand or $10/mo reserved for 1 year
-----
A larger Digital Ocean plan
8GB RAM, 4 cores, 80GB SSD
$80/mo
The comparable EC2 plan seems to be the c3.xlarge
7GB RAM, 4 cores, 80GB SSD
$220/mo on demand or $164/mo reserved for 1 year
----
For an always-on server, like a webserver, something like Digital Ocean is going to be lot more affordable than EC2.
All that being said. EC2 is great for running short jobs. I'll sometimes spin up an AWS instance for a few hours to run some sort of job that would otherwise saturate my computer. Might cost me $5 for that privilege though. I've also used their Elastic MapReduce for when I need to run a map reduce job for a personal project (sadly, I don't have my own Hadoop cluster at home).