Why does the async AWS Java SDK use a thread pool?

Question

This is 2016 already and the Java AWS SDK does not have a really non-blocking API that avoids thread pools (I mean thread to api request scheme)! But why? What if I want to create a mass messaging application or, finally, a Pokemon Go killer based on DynamoDB, I will need to scale up to hundreds of generations of api VM instead of tens in the case of non-blocking IO for the same load (millions of requests per second) . What good is it for me to use something very powerful, like Akka or Vert.x, if the bulk of apis memory uses a thread to query?