Mostly to have stronger separation, I'm sure the person who prefers VM-per-project also has their own reasons.

I just have a few large VMs, each a different environment with slightly different ways how I treat them - the prod ones get more due diligence and being careful, whereas all of the dev ones (including where I host Gitea, Woodpecker CI, Nextcloud, Kanboard, Uptime Kuma etc.) I mess around with the configuration in and do restarts more often. I personally used to run a Docker Swarm cluster, but now just use Docker Compose with Ansible directly, still multiple stacks per each of those servers, dead simple

So my setup ended up being:

  * VPS / VMs - an environment, since don't really need replication/distributed systems at my scale
  * container stack (Compose/Swarm) - a project, with all its dependencies, though ingress is a shared web server container per environment
  * single container - the applications I build, my own are built on top of a common Ubuntu LTS base more often than not, external ones (like Nextcloud and tbh most DBs) are just run directly