Proxmox went ARM — and it made me defend my boring x86 VPS
Proxmox VE just shipped official 64-bit ARM support — the first time in its history it backs a second CPU architecture. The homelab corner of the internet lit up, and I felt the familiar tug to go read about it for a box I don't have. Then I remembered my actual setup: one x86 OVH VPS running Docker directly, no hypervisor in sight. And the ARM news, exciting as it is, mostly made me want to defend that boring choice.
What actually shipped
Proxmox VE 9.2 added official arm64 support on 5 August 2026 — Debian 13.5 base, Linux kernel 7.0, and Proxmox claims full feature parity with the x86 build across KVM, LXC, ZFS, and Ceph. But read the fine print on what it targets: the certified platforms are NVIDIA's Grace and Vera server chips (NVIDIA effectively sponsored the port to feed its AI-factory management stack). Hosts must boot via UEFI and describe hardware via ACPI, which means — sorry, homelabbers — Raspberry Pis and other device-tree-only boards are explicitly not supported. Everything outside the certified server platforms is "best-effort." So this is an enterprise-ARM-server release wearing a homelab-hype halo.
Why it's irrelevant to a one-box VPS
Here's the part that matters for people like me, and it's not a knock on Proxmox. Proxmox is a type-1 hypervisor. Its entire job is carving one physical machine into multiple isolated virtual machines. I have one box, one tenant (me), and my isolation boundary is already the container, not the VM. Stacking Proxmox into my world would mean running a hypervisor inside a VM I'm already renting — because a VPS is itself already virtualized by OVH. That's nested virtualization overhead, a second control plane to patch, and a second thing to break, all to buy isolation I don't consume.
The ARM-homelab hype and "virtualize everything" are cargo cult for a solo operator. On a single-tenant box where Docker already runs on the host, a hypervisor underneath it is a layer solving a problem you don't have.
Which silicon Proxmox can run on is a question for people who own bare metal. A VPS renter doesn't pick the host hypervisor or the host CPU — the provider does. Grace/Vera certification changes nothing about a rented x86 container box, and that's fine.
When Proxmox genuinely is the right tool
I don't want to be dismissive, because Proxmox is excellent at its actual job — and official ARM support is real, good news for the people it serves. You should reach for it the moment you're in any of these situations:
- You own or rent bare metal and want several isolated VMs on it — different OSes, hard tenant separation, separate blast radii.
- You need whole-OS snapshots and rollback — snapshot the entire machine before a risky upgrade and roll the whole thing back in seconds, not just container images.
- You run mixed workloads that can't share a kernel — a Windows VM next to Linux, or something needing a different kernel than the host.
- You've grown into a real homelab: multiple machines, a cluster, live migration, Ceph, HA failover. That's Proxmox's home turf, and it's genuinely great there.
The takeaway
Match the layer to the job. For a homelab with several boxes and mixed workloads, Proxmox is the right tool and arm64 support opens efficient new hardware to it — genuinely worth being excited about. For one Docker host serving one blog, a hypervisor underneath is abstraction you'll pay for in RAM and complexity and never use. The most useful thing the ARM announcement did for me was remind me that the boring stack — containers straight on the host — is still the correct amount of machine for the job I actually have.