Well, Google is competing a bit differently - they are offering to replace not just the servers, but also the software (mail, document sharing, calendar, etc.) that runs on the servers with their cloud/web services.

Still Java-Developer here, with all the bad habits: wouldn't you properly encapsule the use of something like S3? I haven't used it yet, but I imagine I would create some kind of EncodeResourceLink class, that creates a link to a resource an can be configured for a variety of storage solutions. Then some kind of open source clone of S3 could run on my own server. If the amount of data got overwhelming, I could easily shift everything over to S3, or whatever competitors emerge.

Or do I have the completely wrong idea here?

If you want to build your own S3 clone before you get started on your project, that makes perfect sense.

Although, in that case, it makes little sense to use S3 at all. The whole point is that you don't have to write your own clone.

I wouldn't clone S3, only implement the interface.

There is this already:

littles3 - Server implementation of Amazon S3 API

http://code.google.com/p/littles3

Thanks, exactly the kind of thing I was thinking about.