I see those $5/mo or $15/yr VMs as pretty expensive. In the penny-pinching mindset you lowball your RAM which is fine on a good day but it runs out when it gets a load spike. So you need a monitoring system of some kind, backup, and so on, ...
w/ AWS I get all kinds of charts or alarms free or very cheap. The lambda is about 20 lines of Python code, I was able to complete part of the project that I was uninterested in at the time very quickly. Later on it took maybe 2-3 hours to make a UI that would let me add, view and remove feeds from my reader's UI.
YOShInOn's recommendation engine and UI were a research project however that was high-risk and might not have worked so it wouldn't have made sense to develop a better head end.
A better head end is on the agenda today but that system has a lot of other problems such as a dangerously large database that needs to be pruned, O(N) algorithms that were fine for the first year, other problems on the tail end. And it competes with other systems.
My experience is I can set something like this up in AWS and just not think about it for years.