Server & DevOps
So the application does not become the client's problem after handover: reproducible environments, supervised operation, and a restore somebody has rehearsed.
An application is not finished when it works. It is finished when somebody can run it. That means reproducible environments, a deployment that gives the same result twice in a row, and a restore that has actually been rehearsed at least once.
We run both on owned hardware — Proxmox, Linux, HAProxy — and on rented infrastructure. The same rules apply either way: configuration lives in the repository, access is granted individually, and an outage should be visible before a user picks up the phone.