Proxmox VE vs XCP-ng vs Incus: Self-Hosted Hypervisor Showdown for VPS & Bare Metal (2026)

Three years ago the open-source virtualization conversation was basically "Proxmox or VMware." VMware's 2024 licensing overhaul under Broadcom blew that up, and by 2026 the realistic shortlist for anyone building their own server infrastructure looks very different: Proxmox VE, XCP-ng, and the fast-rising newcomer Incus.

I run seven aggregator sites plus 50+ client projects across Hostinger shared hosting, managed VPS boxes, and a couple of bare-metal nodes I use for staging and heavier batch jobs. Over the last two years I have installed all three of these hypervisors at least once on real hardware - not a nested lab toy, but boxes that actually carry traffic and run nightly data imports. This is the comparison I wish existed when I was deciding what to put on a new dedicated server instead of paying for another managed panel.

I will be blunt where I have an opinion, give you the version numbers and prices that actually matter in 2026, and flag the gotchas that only show up after you have lived with each one. Let's get into it.

The 30-Second Verdict

If you only read one paragraph: Proxmox VE is the safe default - the most complete turnkey platform with the best documentation and community. XCP-ng is the choice when you are escaping VMware and want a VM-centric, enterprise-feeling stack with rock-solid backup tooling via Xen Orchestra. Incus is the lightweight, no-subscription, Linux-native option that punches far above its weight if you are container-first and comfortable on the command line.

What Each One Actually Is

Proxmox VE

Proxmox VE is a complete operating system. You install it on bare metal and it replaces whatever was there - it ships as a Debian-based distribution with a KVM hypervisor for full VMs and LXC for system containers, all wrapped in a polished web UI. The big news for 2026 is Proxmox VE 9.0, released on August 5, 2025, rebased on Debian 13 "Trixie", followed by 9.1 on November 19, 2025. The 9.x line ships QEMU 10.0.2, LXC 6.0.4, ZFS 2.3.3, and Ceph Squid 19.2.3 - all recent enough that you are not fighting old kernel quirks.

XCP-ng

XCP-ng is also a full install-on-bare-metal OS, but it is built on the Xen hypervisor rather than KVM. It started in 2018 as a community continuation of Citrix XenServer after Citrix locked features behind licensing. The current release is XCP-ng 8.3 LTS (shipped October 2024, still the LTS baseline through 2026). Unlike Proxmox, XCP-ng deliberately keeps the host thin: management lives in a separate browser app called Xen Orchestra, which you deploy as its own VM or appliance.

Incus

Incus is the odd one out, and the most interesting. It is the community fork of Canonical's LXD, created in late 2023 after Canonical moved LXD under a restrictive contributor agreement. It is now maintained under the Linux Containers project - the same people behind LXC itself. Crucially, Incus is a daemon, not an OS. You install it on top of whatever Linux distribution you already run (Debian, Ubuntu, Alpine, etc.), and it manages system containers and KVM-backed VMs through one unified CLI and API. It does not replace your OS, manage your firewall, or hand you a backup scheduler. It is a building block, not a turnkey appliance - although the new IncusOS minimal hypervisor image (introduced in 2025) is closing that gap for people who want an appliance-style install.

Pricing: The Part Everyone Gets Wrong

All three are free to download and run in production with zero license keys. The difference is what you pay for support and repository access.

Proxmox's pricing list valid from January 13, 2025 through January 11, 2026 puts the entry "Community" tier at €115 per CPU socket per year and "Standard" at €530. Here is the nuance people miss: you do not have to pay anything. The free Proxmox install is fully functional. The subscription buys you the stable enterprise apt repository, signed updates, and support tickets. For my own staging box I run the no-subscription repo and click away the nag dialog. For a client's production cluster, I quote the Community tier because signed, tested updates are worth €115 a socket when downtime has a dollar cost.

XCP-ng is genuinely free with no socket math, which is its single most underrated selling point if you are running a fat dual-socket box. The thing to budget for is Xen Orchestra: the easy "XOA" appliance is rate-limited on some features, so most self-hosters build XO from source (free, a bit of work) or pay Vates for the supported appliance.

My take: if budget is the deciding factor and you have multiple sockets, XCP-ng or Incus win outright. If you want signed enterprise updates and will pay for peace of mind, Proxmox's €115 tier is cheap insurance.

Management Experience

This is where the daily-driver feeling diverges hard.

Proxmox gives you a built-in web UI the moment installation finishes. VMs, containers, storage, clustering, backups, firewall - all in one panel at port 8006. When I onboarded a junior dev to manage a client's two-node Proxmox cluster, they were creating and snapshotting VMs within an hour without me writing a runbook. That is the bar.

XCP-ng separates concerns: the host is minimal and you administer everything through Xen Orchestra in the browser. XO is genuinely excellent for VM lifecycle, scheduled backups, and multi-host management - arguably the best backup UX of the three. The cost is one extra moving part: you have to stand up and maintain XO itself, and if XO is down you are back to the CLI.

Incus was CLI-only for most of its life, and the CLI is fantastic - incus launch images:debian/13 web01 and you have a container in seconds. The web UI (incus-ui) exists now and is improving, but backup scheduling and detailed network config still want the command line. If ssh and YAML profiles make you happy, Incus is a joy. If you want to point and click for everything, it will frustrate you today.

Performance and Architecture

Architecture drives the performance profile more than raw benchmarks do, so let me explain the shape of each.

Proxmox (KVM) leans on the Linux kernel for CPU scheduling and memory management. KVM VMs behave like well-integrated Linux processes, which means tuning is familiar and the ecosystem of KVM knowledge applies directly. LXC containers on Proxmox share the host kernel and are extremely light.

XCP-ng (Xen) uses a type-1 microkernel design: a thin Xen hypervisor boots first, then a privileged control domain (dom0) sits beside guest VMs. This gives stronger isolation between the control plane and guests and finer control over exactly which CPU features get exposed to each VM - which matters for live migration across slightly mismatched hardware. The trade-off is that everything is a full VM; there is no lightweight system-container equivalent to LXC.

Incus was built from the ground up for LXC system containers. Containers start in seconds, share the host kernel, and carry roughly 20-50 MB of RAM overhead each. That density is the headline. Incus can also run full KVM VMs when you need a different kernel or true isolation, so you are not locked into containers.

To put the density claim in concrete terms from my own usage: on a modest VPS where I stage builds for several of my aggregator sites, I keep around a dozen Incus system containers - one per site's PHP/MySQL stack - on a box with 8 GB RAM, and the container overhead is small enough that almost all the RAM goes to the actual MySQL buffer pools and PHP-FPM workers rather than to virtualization tax. Trying the same as a dozen full KVM VMs on Proxmox would have meant carving out a full guest OS footprint each time, and I would have needed a bigger box. For workloads that are "Linux process that needs isolation" rather than "needs its own kernel," containers win on density every time, and Incus is the cleanest container-first tool of the three.

Storage

Storage is where Proxmox quietly dominates.

Proxmox ships first-class support for ZFS, Ceph, LVM/LVM-thin, iSCSI, and NVMe-oF out of the box. Proxmox VE 9.0 added snapshot support on thick-provisioned LVM shared storage, which is a real win for anyone on Fibre Channel or iSCSI SANs. If you want hyperconverged storage, Proxmox + Ceph is the most documented path in the open-source world.

XCP-ng supports a range of storage repositories (local LVM, NFS, iSCSI, and more) but still lacks official Ceph and ZFS support, which is a meaningful gap if distributed or ZFS-based storage is central to your design. There are community routes, but "community route" is exactly the phrase you do not want in a storage layer.

Incus supports ZFS, Btrfs, LVM, Ceph RBD, and directory backends, with ZFS being the recommended and best-integrated option. For a single node, an Incus + ZFS pool gives you instant snapshots and copy-on-write clones that make spinning up identical test environments trivial.

Containers in 2026: The OCI Convergence

Here is a genuinely new development worth its own section. In 2025-2026 all three platforms moved toward running standard OCI (Docker-format) container images directly, which historically was a Docker/Podman job, not a hypervisor one.

• Proxmox VE 9.1 (November 2025) added the ability to pull OCI images from registries or upload them and use them as LXC container templates.
• Incus got there first: since Incus 6.3 you can launch application containers straight from OCI images, with all the normal Incus config options applying, run inside the same safe container environment as system containers.
• XCP-ng stays VM-centric here - you would run Docker inside a guest VM rather than have the hypervisor pull OCI images natively.

This convergence matters because it blurs the old line between "hypervisor" and "container runtime." For my use case - where most workloads are containerizable web stacks - being able to pull an OCI image at the hypervisor layer means fewer nested abstractions and one less Docker daemon to babysit.

Backups and Disaster Recovery

Nobody enjoys this topic until the night they need it. From running nightly data imports of 100-200 records across seven sites, I have learned that the backup you do not test is a backup you do not have.

Proxmox pairs with Proxmox Backup Server (PBS), a separate free product doing incremental, deduplicated, encrypted backups with verification. The PBS + PVE combination is excellent and is what I recommend for any Proxmox deployment that matters.

XCP-ng arguably has the best built-in backup UX through Xen Orchestra: scheduled full and delta backups, continuous replication to a second host, and easy restore - all from the same UI you already use. If backup simplicity is your top priority, XO is hard to beat.

Incus gives you the primitives - snapshots, incus export to a tarball, and remote copy between Incus servers - but no scheduler. You wire it into cron or systemd timers yourself. Powerful, but it is on you.

Clustering and Scale

If you are running one box, skip this; if you are planning to grow, read it.

Proxmox has mature, built-in clustering with Corosync and supports live migration, HA failover, and Ceph hyperconvergence across nodes - all from the same web UI. XCP-ng organizes hosts into pools with shared storage and live migration, and Xen Orchestra manages many pools centrally; it is a comfortable fit for traditional enterprise VM estates. Incus also clusters (it has a built-in distributed database and can pool many nodes), and it is surprisingly capable, but the tooling assumes more operator skill than Proxmox's clicky cluster wizard.

So Which One Should You Run?

After living with all three, here is how I actually decide:

• Choose Proxmox VE if you want one polished platform that does VMs and containers, you value documentation and a huge community, and you might want Ceph or HA clustering later. It is the lowest-regret default for most people. This is what I deploy for clients who need someone else on their team to manage it.
• Choose XCP-ng if you are migrating off VMware, your world is full VMs (not containers), and you want best-in-class backup and replication through Xen Orchestra with zero per-socket licensing. Stronger guest isolation via Xen is a bonus for multi-tenant scenarios.
• Choose Incus if you are container-first, comfortable on the CLI, want maximum density on modest hardware, and refuse any subscription nag whatsoever. It is my pick for personal infrastructure and lean staging boxes where I want to pack many isolated stacks onto one VPS.

One honest caveat I would not have appreciated two years ago: do not over-index on the hypervisor if you are renting a single managed VPS. If your provider already gives you a KVM VPS, layering a full hypervisor inside it (nested virtualization) is usually the wrong move - run Incus or plain Docker on it instead, and save the bare-metal hypervisor decision for an actual dedicated server where you control the hardware. I have watched people burn a weekend nesting Proxmox inside a budget VPS only to fight CPU-flag and performance issues that a container would have sidestepped entirely.

Frequently Asked Questions

Is Proxmox VE really free for commercial use?

Yes. The full Proxmox VE install is free to use in production, including commercially. The paid subscription (from €115/socket/year) buys access to the tested enterprise repository, signed updates, and support - it is not a license to run the software. You can run the no-subscription repository indefinitely; you just get a nag dialog and the "test" update channel.

Does XCP-ng cost anything?

No license fee and no per-socket charge. XCP-ng 8.3 LTS is free. The only money on the table is optional: paid support from Vates, or the supported Xen Orchestra appliance. Many self-hosters build Xen Orchestra from source for free instead.

Can Incus run virtual machines or only containers?

Both. Incus runs LXC system containers (its specialty, very low overhead) and full KVM virtual machines through the same CLI and API. Since Incus 6.3 it can also launch application containers directly from OCI (Docker-format) images.

Which has the best storage support?

Proxmox VE, by a clear margin - native ZFS, Ceph, LVM-thin, iSCSI, and NVMe-oF. Incus has solid ZFS, Btrfs, LVM, and Ceph RBD support. XCP-ng still lacks official Ceph and ZFS support, which is its biggest storage limitation in 2026.

What replaced VMware for most people in 2026?

There is no single answer, but Proxmox VE captured the largest share of departing VMware users thanks to its completeness, while XCP-ng won the crowd that wanted a VM-centric, XenServer-like experience. Incus is the dark horse for container-heavy and homelab users who never needed full VMware-style VMs in the first place.

Final Word

There has never been a better time to self-host your own virtualization. All three of these are production-grade, genuinely free to start, and actively developed. If you are choosing today and want a single recommendation to stop the analysis paralysis: install Proxmox VE 9.1 on your first dedicated server, learn it well, and only reach for XCP-ng or Incus once you have a specific reason - VMware migration or extreme container density, respectively. Whichever you pick, set up tested backups before you put anything real on it. That is the lesson that has saved me more than any benchmark ever did.