The cloud vendors all have their version of the Well Architected Framework. I recommend that you read about it before you start implementing anything in the cloud.
For AWS, the Certified Cloud Practitioner exam will ask you several questions on the Well Architected Framework. The exam is one of the most basic in the AWS canon though I feel very strongly that its importance as a foundation stone is far higher than the its low complexity would suggest.
The cost optimisation pillar of each cloud vendor's Well Architected Framework is the one to pay attention to here.
The 37Signals story about Leaving the Cloud is a case in point. I can remember Buck Woody and Andrew Fryer talking at SQLBits describing the cloud as someone else's data centre (or rather data centres) and it is. A series of massive data centres where most of us are constrained only by our budget. And that is the problem. We don't have to ask ourselves whether an extra server will physically fit in our computer room, whether we need to upgrade the power, cooling and a whole host of things that go with running our own hardware. Many of the cloud facilities we just use rather than manage. This leads to architectures that look good as a logical representation on paper (Powerpoint) but cost a lot when implemented physically.
I feel that many companies have forgotten basic disciplines. It is easy to scale up and even easier to scale out but how often do we ask ourselves if these are the best options? Isn't this the equivalent of throwing hardware at the problem rather than solving the problem?