The MLAgents docs have docs on:
- training on AWS (marked ‘we dont use this’)
- training on Azure (marked ‘we dont use this’)
- training on docker (marked ‘we dont use this’)
So… given all these training approaches say you don’t use them, what approach are you using to train?
training model != cloud service 
Said that, I don’t know details, but with mlagents-learn they start a local Server on your machine to which then Unity Editor/an instance connects. So you could rent a virtual cloud pc (I think Linux would work) and then upload your files and let it train their. AWS and stuff needs some know-how to setup.