The problem with Dynamo DB is that one needs to provision the read and write capacity in advance, which means that one needs to over provision in order to be sure. Also even for things like exporting database which requires lots of reads in short time, one needs to provision the read capacity. This can make the back ups very expensive. However I am basing this from my study of their docs as we were building our application. (A chat server).
It will be good if somebody can share their actual experience about Dynamo DB especially from the perspective of cost. Yes, almost unlimited scalability is very attractive, but I found that I had to think about data access patterns a lot to ensure that we will stay in acceptable read write quota for Dynamo DB. This was causing a lot of delay in our MVP. So we decided to go with Mongo DB which we knew we could scale fairly beyond our MVP and may be rewrite the data access layer later on if we achieve a mass that requires a switch to something like Dynamo Db or manual sharding.