Skip to content

Proxmox VE

Summary

Debian-based open-source (AGPLv3) virtualization platform unifying QEMU/KVM virtual machines and LXC system containers under one web UI, with clustering, high availability, software-defined storage (ZFS, Ceph), SDN, and integrated backup via Proxmox Backup Server. It is the default landing zone for the post-VMware private-virtualization exodus. Current release: PVE 9.2 (x86-64 2026-05-21, arm64 2026-08-05).

Key Facts

Fact Value
Latest Version 9.2 (2026-05-21; arm64 build 2026-08-05); pve-manager 9.2.21 as of 2026-09-24
Base Debian 13.5 Trixie, kernel 7.0, QEMU 11.0, LXC 7.0, ZFS 2.4, Ceph Tentacle 20.2 (default) / Squid 19.2
Vendor Proxmox Server Solutions GmbH (Vienna); Proxmox North America Inc. (Kingston, Ontario) announced 2026
License AGPLv3; every feature free, subscriptions add support + Enterprise repo
Subscriptions EUR 120 / 370 / 550 / 1,100 per socket-year (Community / Basic / Standard / Premium)
Support lifecycle PVE 9: EOL tba (Debian 13 lifetime); PVE 8 EOL 2026-08
Architectures x86-64; arm64 (officially NVIDIA Grace Hopper / Vera, other ARMv9 best-effort)
Family Proxmox Backup Server 4.2 (2026-04-29), Proxmox Datacenter Manager 1.1 (2026-05-28)
Interfaces Web GUI + REST API on TCP 8006; CLI qm, pct, pvesh, pvecm, pvesm, ha-manager

Overview

Proxmox Virtual Environment (PVE) turns commodity servers into a hypervisor cluster managed through one integrated stack: web GUI, CLI and REST API. Everything lives in the same package set on every node — there is no separate management appliance to size. Cluster configuration is a corosync-replicated filesystem (/etc/pve), HA is watchdog fencing plus the Cluster Resource Scheduler, and storage is one plugin abstraction over ZFS, Ceph, LVM, NFS/CIFS, iSCSI and PBS.

The compact view below shows how the pieces fit; the per-node and backup-path diagrams live in Explanation.

flowchart LR
    PDM["Datacenter Manager<br/>(multi-cluster view)"]
    subgraph Cluster["PVE cluster (corosync + pmxcfs /etc/pve)"]
        N1["node: pveproxy :8006<br/>QEMU/KVM + LXC"]
        N2["node: pveproxy :8006<br/>QEMU/KVM + LXC"]
        HA["pve-ha-crm / lrm<br/>+ CRS scheduler"]
    end
    STOR["Storage plugins<br/>ZFS, Ceph RBD/CephFS, LVM, NFS"]
    PBS["Proxmox Backup Server<br/>dedup chunk store, S3"]
    PDM -->|"API"| N1
    PDM -->|"API"| PBS
    N1 <-->|"corosync"| N2
    HA --- N1
    HA --- N2
    N1 --> STOR
    N2 --> STOR
    N1 -->|"vzdump"| PBS

Evaluation

  • Why it's better: One product covers the whole stack competitors split across projects: hypervisor + cluster + HA + scheduler (CRS), hyperconverged Ceph, ZFS replication, incremental deduplicated backup (Proxmox Backup Server), SDN with EVPN and WireGuard/BGP fabrics, and RBAC/OIDC identity — all AGPLv3, no feature gates per subscription tier.
  • When it fits: On-prem private cloud and homelab alike: VM + container consolidation, VMware vSphere replacements (native ESXi import wizard), hyperconverged 3-6 node clusters with Ceph, GPU passthrough for AI workloads (NVIDIA vGPU supported), NVIDIA Grace-class arm64 hosts (9.2 arm64), and air-gapped enterprises (offline updates on Premium).
  • When it does not fit: Multi-tenant public IaaS at OpenStack scale (tenancy, quota, billing); teams standardized on vSphere-specific tooling (vMotion/DRS/NSX-based automation); strict per-workload commercial SLAs without budget for the support tiers (Community tier has no ticket support).
Pros Cons
AGPLv3, all features free; subscriptions only add support + Enterprise repo Multi-cluster management (PDM 1.0 Dec 2025, 1.1 May 2026) is young compared with vCenter for very large estates
Unified VMs (KVM) + LXC containers, templates/linked clones, live migration Cluster file system (pmxcfs) caps config at 128 MiB — thousands of VMs fine, exotic metadata not
Hyperconverged Ceph + ZFS built in; snapshots-as-volume-chain (9.x, tech preview) qcow2 internal snapshots on file-based storage can block large VMs for minutes
Dynamic-load CRS with automatic HA rebalancing and cluster-wide HA disarm (9.2) Cross-vendor (Intel/AMD) live migration not guaranteed; cross-architecture (x86/arm64) impossible
PBS: dedup, encryption, live restore, backup fleecing, S3 datastores (GA in 4.2) nftables firewall still tech preview; the stock firewall is iptables-based
Official arm64 support since 2026-08 (same feature set) Live migration of passed-through GPUs is experimental and NVIDIA-only
  • Common Use Cases: Homelab consolidation (LXC-first), enterprise VMware replacements, hyperconverged 3-node Ceph branches, VDI with GPU passthrough, backup targets via PBS, CI runners, air-gapped defense/regulated deployments (offline updates).
  • Licensing & Commercial Use: AGPLv3 open source. Subscriptions are per occupied CPU socket (cores irrelevant), all nodes in a cluster at the same level: Community EUR 120 (Enterprise repo, no tickets), Basic EUR 370, Standard EUR 550, Premium EUR 1,100 per socket-year, or the free no-subscription repo. Premium gets global 24/7 support from 2026-10-19; Standard follows in a Q4 2026 window. Subscription keys are bound to CPU architecture (arm64 keys via sales). See Reference.
  • Ecosystem & Data Connections: Proxmox Backup Server (dedup backup), Proxmox Datacenter Manager (multi-cluster management; free, enterprise repo tied to remotes' subscriptions), Terraform/OpenTofu and Ansible providers (community), REST API + API tokens, community helper-scripts ecosystem, SDN IPAM integrations (NetBox, phpIPAM) and PowerDNS.
  • Compatibility & Requirements: 64-bit x86 (Intel VT / AMD-V) or arm64 (ARMv8-A min, ARMv9 recommended, UEFI + ACPI only); IOMMU (VT-d/AMD-Vi/SMMU) for passthrough; ~2 GB RAM for the OS + guest RAM; ZFS/Ceph ~1 GB RAM per TB used storage. See Reference.
  • Latest Versions: PVE 9.2 (x86 2026-05-21; arm64 2026-08-05), PBS 4.2 (2026-04-29), PDM 1.1 (2026-05-28). Roadmap themes: volume-chain snapshots toward GA, nftables firewall graduation, balancing for non-HA guests, large-estate "Cattle and Pets" features.
  • Alternatives: OpenNebula (topic) and OpenStack (topic) for cloud-style orchestration; XCP-ng + Xen Orchestra; Harvester (Kubernetes-native HCI); oVirt (upstream of the discontinued Red Hat Virtualization); the incumbent being migrated away from: VMware vSphere.
  • Migration & Lock-in Risks: Config in plain text under /etc/pve (pmxcfs) + open API — exit paths are real (qcow2/raw/zvol images are portable; PBS backups restore anywhere with the key). Lock-in is low; the bigger risks are corosync network design and staying current (see How-to guides).
  • Community Health: Very active forum and pve-user list, frequent point releases (22 pve-manager 9.2.x builds between 2026-05-20 and 2026-09-24), large third-party script/library ecosystem, and enterprise adoption growing since the Broadcom/VMware licensing shock of 2024.

Lifecycle

PVE 8 reached end-of-life in 2026-08 and the Ceph Squid repositories reach EOL in 2026-09. Upgrade with pve8to9 and plan the Squid -> Tentacle move.

Position in the Landscape

The private-virtualization field sorts by ambition: PVE owns the "hypervisor appliance" layer (per-node simplicity, clusters of a few to ~50 nodes); OpenNebula adds cloud-style self-service and federation without OpenStack's weight; OpenStack stays the answer for genuine multi-tenant IaaS with quotas, billing, and many teams. PVE's bet is that most "private cloud" demand is really a well-run hypervisor cluster plus good backups — the 2024-2026 VMware migrations validated that bet operationally, and Datacenter Manager, 24/7 support, and the "Cattle and Pets" roadmap are how it defends the upper end. The full side-by-side is in Proxmox VE vs OpenNebula vs OpenStack.

Topic Map

  • How-to guides — install, repos, VMs/CTs, clustering + QDevice, HA/CRS/disarm, Ceph, PBS backups, passthrough, SDN, 8 -> 9 upgrade, ESXi import, troubleshooting
  • Reference — versions, release history, support lifecycle, pricing, requirements, ports, repos, storage matrix, CRS modes, limits, hardening checklist
  • Explanation — pmxcfs/corosync internals, HA fencing sequence and CRS, storage and backup pipelines, SDN/identity/firewall, PDM, threat model
  • Comparisons — pointer to the domain-level comparison pages

Sources

Questions

  • How quickly does PDM close the gap to vCenter for large multi-cluster estates after 1.1? (Track the PDM roadmap and changelog.)
  • When do snapshots-as-volume-chains leave tech preview on LVM/Directory/NFS/CIFS, and when does replication extend beyond ZFS? (roadmap: in progress)
  • When does the nftables-based proxmox-firewall leave tech preview and become the default? (still tech preview in 9.2 docs)
  • What is PVE 9's official EOL date (listed as "tba"; expected to follow Debian 13's lifetime)?
  • Real-world behavior of the 9.2 dynamic CRS rebalancer on 50+ node clusters — community data is still thin.