Thanks for this feedback. Couple of points of clarification. The infrastructure interpretation happens at build/deploy time, not at execution time, so there isn't any additional tax. The deployment engine isn't open source, but we are evaluating ways to have optional eject paths. In terms of deployment across cloud providers, the abstractions we've built (and are working on) use all those fancy AWS managed services like DynamoDB. Our abstractions focus on use cases, not primitives, so when we eventually deploy to Azure, we'll be able to replicate the service using native services like Cosmos DB.