I think it might have something to do with the fact that the payment model isn't connected to a server instance, but how often your code is run. Heroku, GAE etc

I think it might have something to do with the fact that the payment model isn't connected to a server instance, but how often your code is run. Heroku, GAE etc. are server-oriented. Your code is running on a dyno or an instance, and you pay based on the number of instance-hours you use. In AWS Lambda, or Azure Functions, you pay for the number of requests you process and the time spent processing those requests, how many instances are spent processing those requests are irrelevant. It's a serverless payment modell.