Proxmox VE vs XCP-ng: Choosing an Open-Source Hypervisor

Answer first

Both are serious, actively developed, fully open-source hypervisor platforms with real companies behind them — nobody gets fired for picking either. The short version of where each wins:

• Proxmox VE is the stronger default for dedicated-server operators: everything integrated on every node (web UI, clustering, backups, firewall), first-class ZFS and Ceph, native LXC containers alongside VMs, and a modern Linux/KVM base that tracks current kernels. If you run one to a dozen hosts and want the platform self-contained, it's the shorter path.
• XCP-ng appeals where its XenServer heritage does: a deliberately thin hypervisor host with management centralised in Xen Orchestra — one pane for many pools and sites — a long-LTS release model, per-host (not per-socket) commercial support from Vates, and excellent VMware import tooling. Shops coming from ESXi + vCenter often find its shape familiar.

The rest is the reasoning — and the details both fan bases tend to gloss over.

Architecture: KVM in Linux vs Xen plus dom0

Proxmox VE is a Debian distribution (currently Debian 13-based, PVE 9.2, Linux kernel 7.0) whose hypervisor is KVM — a module inside the host's general-purpose kernel, with QEMU providing devices. The host is a full Linux system: standard tooling, drivers arriving with kernel updates, and the ability to run LXC containers natively next to VMs.

XCP-ng is a dedicated appliance built on Xen (currently Xen 4.17 in XCP-ng 8.3): a slim type-1 microkernel-style hypervisor boots first, and all management runs in a privileged control VM, dom0. It descends from Citrix XenServer — forked openly in 2018, now stewarded by Vates — and inherits that design's virtue (a minimal, appliance-like host you barely touch) and its baggage: the current dom0 userspace is CentOS 7-based with a 4.19-era kernel, kept secure by Vates' own backporting rather than upstream freshness. A modernised XCP-ng 9.0 is in development without announced dates — so evaluate the platform as it ships today. Practical consequence of the aging dom0: newer hardware sometimes needs added drivers, where Proxmox's current kernel tends to just have them.

One capability difference falls straight out of the architectures: Proxmox runs containers natively; XCP-ng is VM-only (containers go inside VMs).

Management: integrated everywhere vs one pane for everything

This is the philosophical split that decides most real adoptions:

• Proxmox: every node ships the full web UI; any node manages the whole cluster. Nothing extra to deploy, no management VM to keep alive — the trade being that multi-cluster/multi-site oversight is not the built-in strength.
• XCP-ng: the host offers only the lightweight built-in XO Lite for basics; real management lives in Xen Orchestra (XO) — a separate deployment that then manages many pools and sites from one place, including the whole backup system. Know the two XO flavours: the turnkey appliance (XOA) is feature-limited on the free tier, while building XO from source unlocks everything, free — same code, no QA/support. (The rewritten XO 6 interface shipped in late 2025 and is still maturing; the XO 5 UI remains the daily driver for some operations.)

Neither is wrong: integrated-per-node is simpler at small scale; hub-and-spoke scales oversight better. Match it to how many hosts and humans you have.

The comparison table

Two table rows deserve the footnotes vendors skip. The 2 TiB disk limit that dogged XCP-ng for years was genuinely lifted only in May 2026, when QCOW2 support went GA (new ceiling 16 TiB) — recent enough that much advice online still treats it as current, in both directions. And SMAPIv3, XCP-ng's next-generation storage stack, remains a tech preview — today's production storage runs on the mature-but-older stack the QCOW2 work extended.

Storage and backups: the depth difference

For a dedicated-server operator this is Proxmox's strongest suit: ZFS at install time (checksums, snapshots, replication) and Ceph fully integrated for hyperconverged clusters — both free, both first-class in the UI. XCP-ng's SR model covers the standard backends well, but ZFS/Ceph are not first-class citizens, and Vates' hyperconverged answer (XOSTOR) is a paid add-on.

Backups are closer to parity, differently shaped: Proxmox pairs built-in vzdump with the separate free Proxmox Backup Server for deduplicated, verified, encrypted backups; XCP-ng's equivalent lives inside Xen Orchestra — delta backups, continuous replication, mirror-to-cold-storage, S3 remotes — genuinely capable, with the same from-source-or-paid access nuance as XO itself. Ecosystem note that says something about both platforms' maturity: Veeam added official support for Proxmox and then XCP-ng in its recent releases — the traditional-enterprise backup world now treats both as mainstream.

Pricing and support: per socket vs per host

Both platforms are fully functional with no subscription — you pay for stable/tested repositories and humans, not features. The models differ in shape, which matters at different fleet geometries:

• Proxmox charges per CPU socket: €120 (Community, enterprise repo access) to €1,100 (Premium, 2-hour response) per socket per year. Dual-socket servers count double; single-socket boxes are cheap to cover.
• Vates VMS charges per host (or flat): Essential at $2,000/year covering up to 3 hosts, Pro $1,000/host/year, Enterprise $1,800/host/year with 24/7 — dense dual-socket hosts don't cost extra, small three-host shops fit the flat tier neatly.

Run your actual topology through both price lists; the winner flips with socket counts. Both companies' free tiers are honest enough that many production fleets run unsubscribed with community support — a legitimate choice if your team can self-rescue.

For the VMware refugees

Both projects have spent the post-Broadcom years polishing the runway. Proxmox ships an ESXi import wizard (point it at a host, import VMs with disks). Xen Orchestra's V2V import streams from ESXi/vSphere via VMware's own VDDK — only allocated blocks, warm-migration support, vSAN sources included. Neither migration is the hard part anymore; the hard part is re-mapping operational habits (vCenter → XO comes naturally; vCenter → Proxmox's per-node model is a bigger mental shift, softened by better storage options at the end of it).

Verdict, with conditions

• Choose Proxmox VE if you're a dedicated-server operator who wants one platform per box with everything aboard — ZFS/Ceph, containers, backups via PBS — and current-kernel hardware support. The default recommendation for most readers of this blog, and why our Proxmox line exists.
• Choose XCP-ng if you're centralising management of multiple pools/sites in one XO, you value the thin-appliance host and long LTS windows, you're coming from XenServer/ESXi and the pool model fits your team, or per-host support pricing suits dense hardware.
• Either way, you're getting a real platform, not a compromise: active development, real VMware import, real backup stories, and companies (Proxmox GmbH, Vates) whose incentives align with the open-source products.

Frequently asked questions

Is XCP-ng's CentOS 7-based dom0 a security problem?

Not in the way the words suggest — Vates' security team backports fixes to the dom0 userspace and Xen itself is patched actively, so "CentOS 7 is EOL" doesn't translate to "XCP-ng is unpatched". The honest costs are elsewhere: older kernel means newer hardware occasionally needs driver work, and the platform carries technical debt that the in-development XCP-ng 9.0 exists to pay down (no release date announced). Judge the product shipping today: secure, somewhat dated underneath, with modernisation visibly in progress.

Which is faster, Xen or KVM?

There's no honest blanket answer — both are mature type-1 designs within a few percent of each other on typical server workloads, and results flip with workload, guest OS, storage, and tuning. Anyone quoting a decisive cross-hypervisor benchmark is usually measuring storage stacks or driver freshness, not the hypervisors. Choose on management model, storage integration, and ecosystem — the dimensions above where the differences are real and durable — not on microbenchmarks.

Do I need to pay to get Xen Orchestra's full features?

No — that's the most misunderstood fact in this comparison. XO's source is fully open with every feature (including the complete backup suite); building it from source (community scripts make this a few commands) unlocks everything, free. What Vates sells is the prebuilt XOA appliance with QA and support behind it. Budget honestly though: "free" from-source XO means you're the one updating and rescuing it — the same self-support calculus as Proxmox's no-subscription repo.

Which should I run on a single dedicated server?

Proxmox, in most cases: the integrated UI means the box manages itself with nothing extra deployed, ZFS from the installer gives single-host storage resilience, LXC lets small services skip VM overhead, and the single-node story is polished. XCP-ng's architecture — thin host, management elsewhere — shines with multiple hosts; on exactly one server it means either running XO as a VM on the host it manages or living with XO Lite's basics.

Deploying on Serverside

Whichever side of this comparison you land on, the substrate is the same: our dedicated servers with NVMe storage and private networking come with Proxmox VE pre-installed in under a minute — and XCP-ng installs cleanly via our ISO + KVM-over-IP console if the Xen path is yours. ASN 55285, always-on DDoS mitigation, and hardware that either hypervisor will happily consume.

The Proxmox side of the bookshelf: what Proxmox VE is, the ESXi comparison, setup, networking, and clustering.