In this context, lock-in is a function of the size of the AWS user's engineering team, and I don't think the NYT Author thoroughly understands this aspect of th

In this context, lock-in is a function of the size of the AWS user's engineering team, and I don't think the NYT Author thoroughly understands this aspect of the problem. It wouldn't be infeasible to, say, move from S3 to OpenStack Swift, but that project is much more feasible for either a large team or a small team with very little data. Similarly, one could use say RabbitMQ instead of AWS SQS, but that takes a bit of effort. The availability of Docker images (and, eventually, k8s- or Swarm-orchestrated cloud apps) helps here too. While AWS might have proprietary APIs, the semantics of those APIs (e.g. 'enqueue this blob') don't entail very much proprietary magic. Instagram's post about moving off AWS is enlightening here: http://instagram-engineering.tumblr.com/post/89992572022/migrating-aws-fb http://instagram-engineering.tumblr.com/post/89992572022/mig... The problem isn't as bad as, say, writing your app in Objective-C or CUDA only to later find you don't want to ship to iOS or NVIDIA cards.