Public cloud was sold as the answer to every infrastructure question. For some workloads it still is — but for steady-state, data-heavy, or sovereignty-sensitive workloads, the economics increasingly point back to a private cloud you control. This piece walks through how to evaluate repatriation honestly, what a Proxmox-based target architecture looks like, and the operating model needed to keep it boring.
When repatriation actually makes sense
Repatriation isn't ideological. It's a TCO and risk decision. The strongest candidates share a few traits: predictable demand, high storage or egress, regulated data, or workloads where elasticity is rarely used.
What a Proxmox target looks like
A modern Proxmox estate combines clustered VE nodes, Ceph or ZFS-replicated storage, an SDN fabric for multi-tenant networks, Proxmox Backup Server for deduplicated backups, and a clear automation layer (Ansible, Terraform) for provisioning. Done well, it's quiet, observable, and predictable.
The operating model
Hardware alone doesn't repatriate workloads. The operating model — patching cadence, capacity planning, backup and restore drills, incident response — has to be in place before cutover. We help clients design that operating model alongside the platform, then run it with them or for them.