Hi Aleksei,
As I understood it there is a self hosted pro version (Enterprise), but that is not 15 USD / Month.
That being said we build a custom solution in AWS because in the beginning, we where not aware of the autoscaler (and after testing the autoscaler I am still not sure whether we will stick to our solution).
Basically we build a solution outside of the server Instance. A seperate Instance is polling the ClearML queues, spinning up Instances, installing the ClearML Agent on them which then registers to the queues and fetches the experiments. If the instance is Idle for a defined time it will be shut down again. This should work on any cloud and is mainly written in Terraform.