Proxmox VE 9.2 brings official 64-bit Arm support with the same release lifecycle and enterprise support as x86-64. Useful if you are weighing Arm platforms, with some real constraints worth knowing first.
What was announced
- Proxmox VE 9.2 adds official support for 64-bit Arm (arm64/aarch64), sharing the same code base, package repositories and release lifecycle as x86-64.
- The arm64 build is based on Debian 13.5 Trixie with Linux kernel 7.0, and ships QEMU 11.0, LXC 7.0 and ZFS 2.4.
- Full support is limited to NVIDIA Grace Hopper and NVIDIA Vera platforms; other UEFI-based ARMv9-A or newer server hardware receives best-effort support.
- Hosts must boot via UEFI and describe hardware through ACPI, which rules out device-tree-only devices such as Raspberry Pi via the standard installer.
- AMD SEV memory encryption and Intel GVT-g mediated vGPUs remain x86-only; live migration is limited to nodes of the same architecture and mixed x86-64/arm64 clusters are not officially supported.
- Proxmox states arm64 is not a technology preview and carries the same support windows as x86-64, with enterprise repository access and subscriptions available on request.
- Proxmox says it validated systems with Supermicro as well as NVIDIA, and claims ZFS and Ceph reach feature and performance parity with x86 deployments.
The ETON view
This is the kind of unglamorous release that quietly changes what is buildable. Arm servers have had the silicon for a while; what they have lacked is a supported hypervisor that an ordinary operations team can run without becoming a distribution maintainer. Storage parity for ZFS and Ceph matters more than the CPU support itself, because it means an Arm node can be a storage node rather than only a compute node.
The constraints are the part to plan around. No cross-architecture live migration and no officially supported mixed clusters means an Arm deployment is a separate cluster with its own capacity headroom, not a few nodes dropped into an existing estate. Treat it as a new pool for a specific workload, size it so you can lose a node, and confirm your backup and monitoring stack is genuinely architecture-neutral before committing. Where it makes sense, and density-per-watt workloads are the obvious case, we can spec and supply the platform alongside the x86 estate you already run.
Related infrastructure
Category: Servers · Vendor: Proxmox, NVIDIA, Supermicro · Technology: Proxmox VE 9.2, arm64, Grace Hopper, NVIDIA Vera, ZFS, Ceph · Last verified: 02 Sep 2026 · ~2 min read
Sourcing this kind of infrastructure? Talk to ETON about availability, lead time and pricing.
