There's a few key differences between S3 and the Pulumi hosted backend: 1. S3 has no concurrency control. This is a big deal for your use case as you run the risk of automated deployments colliding and potentially corrupting state/making simultaneous edits to your infra. If you want to use the S3 backend, you'll have to implement locking around your deployments to guard against this. The pulumi backend has locking and concurrency control built in. 2. The pulumi backend is designed for software teams, and has features like stack access contorl, RBAC, built-in concurrency control, etc. 3. The pulumi backend has some interesting state storage capabilities that aren't in the other backends. You can refresh config from the last deployment, meaning you don't have to worry about storing it yourself. I've created a temporary vm manager that queries that pulumi backend for all ephemeral stacks matching time based criteria. This just kinda works and doesn't require me to maintain any state myself. It will work either way, but the S3 backend will require more manual effort on your part.