A few years ago when AWS was getting a lot of attention in the library world (and everywhere else!), it immediately seemed to me, based on some back of the envelope calculations, to be likely unaffordable to libraries. And not a very good value proposition compared to our standard self-hosting — especially for academic libraries which can benefit from their host universities IT infrastructure, but in general, it wasn’t cheap.
Here’s a blog post arguing that in general, EC2 indeed isn’t a great value proposition — if you mostly need 24/7 instances. Where EC2 shines, of course, is it’s ability to “elastically” (the “E” in EC2) spin up and down instances on the fly, and pay hourly, to quickly adjust your provisioning for to the moment demand.
That’s for architectures that are horizontally scaled out a lot, and need to scale up a lot — not most of our library things, but perhaps increasingly more and more (if we succesfully get more competent and succesful!), sure. Although the author notes and provides some calculations showing that even this can be a dicey value proposition.
In that blog post, he mentions other providers with better prices, but not by name.
In the Hacker News thread, someone mentioned Digital Ocean. They seems to offer a service roughly analagous to EC2, but with far better prices — including hourly charging and instantaneous provisioning, so in theory supporting on-demand load balancing for highly horizontally scaled services, just like EC2. It might still be hard for a university library to beat in-house hosting, due to having our existing university IT infrastructures where we probably don’t need to pay for or pay seriously reduced pricing for bandwidth, electricity, server room facilities, maybe even operations staff, etc. But if you do have a context where cloud hosting makes sense, it’s worth remembering that Amazon is not the only reliable/competent player in town, and is definitely not the cheapest — although it may be the most feature-complete.