Both. Amazon has a habit of going down when you least expect it, and while the outages aren't usually system wide, you want an off-site backup site that can ha

Both. Amazon has a habit of going down when you least expect it, and while the outages aren't usually system wide, you want an off-site backup site that can handle some percentage of the traffic, even if it's just a "we'll be right back" page. Additionally, if all your data is on Amazon and it has a catastrophic failure, it's can be lost forever. Your databases should be hosted on your own network as Amazon can't guarantee that nobody will have access to your network segment but you. If you take credit cards, be especially careful as the service level you choose may not be PCI compliant, which is required by most credit card companies. Read this (may be out of date): https://forums.aws.amazon.com/thread.jspa?threadID=34960 https://forums.aws.amazon.com/thread.jspa?threadID=34960 If you use their payment system, they do claim to be PCI compliant. Regarding bandwidth, in your example, you show $8/mbit for bandwidth. I get 300mbps for $3/mbit from Cogent, and $6/mbit for a 100mbit commit. Additionally, the example shows reserved instances at 51 and self-hosted instances at 131. This implies that Amazon's reserved instances are somehow magically more than twice as powerful as self-hosted servers. EC2 Reserved instances are NOT an instance type, but rather a rate plan, and Amazon's servers are notoriously slow (and shared), and in my experience there's no way an Amazon EC2 instance is going to outrun a standard dual or quad-core XEON server in a self-hosted model.