I don't think there's a good answer to that. It depends on how much compute and bandwidth does every request need :) As you're using Azure you can easily setup

I don't think there's a good answer to that. It depends on how much compute and bandwidth does every request need :) As you're using Azure you can easily setup rules to autoscale, so you can just leave it to do the magic for you or you can learn from its behaviour and setup your own desired architecture.