What Is Proxmox VE? Choosing a Bare-Metal Hypervisor

What is Proxmox VE?

Proxmox VE (Virtual Environment) is a free, open-source virtualisation platform that installs directly onto a physical server and turns it into a host for both full virtual machines and lightweight containers. It combines a KVM hypervisor for full VMs with LXC for system containers, and wraps both in a single web interface with built-in clustering, software-defined storage, and a REST API. It is built on Debian — the current release, Proxmox VE 9.2, sits on Debian 13 "Trixie" — and is released under the GNU AGPLv3, so there is no per-socket or per-core licence to run it in production. A paid subscription exists, but it buys a stable update channel and vendor support, not features.

Type 1 vs type 2, and what "bare-metal hypervisor" means

A hypervisor is the layer that lets one physical machine run many isolated operating systems. Type 1 (bare-metal) hypervisors install directly on the hardware and manage the CPU, memory, and I/O themselves; type 2 hypervisors — think VirtualBox or VMware Workstation — run as an application on top of an existing desktop OS. Every platform in this guide is type 1: it owns the metal.

Running a type 1 hypervisor implies you have the whole machine to yourself. That is the domain of a dedicated server rather than a shared VPS, and the trade-offs of bare metal versus a virtualised slice — uncontended cores, local NVMe, no noisy neighbours — are the same whether you install a plain Linux distribution or a hypervisor on top. We cover that argument in full in our guide to choosing a Linux distribution for a dedicated server, so this article assumes you have already decided you want your own box and focuses on which hypervisor to put on it.

The contenders at a glance

Four platforms dominate the bare-metal hypervisor conversation in 2026, and they occupy genuinely different positions. Proxmox VE and XCP-ng are open-source and free to run; VMware ESXi is the incumbent enterprise standard, now under Broadcom; and Hyper-V is Microsoft's Windows-native option. The table below compares them on the dimensions that actually drive the decision, and the sections after it dig into each one.

Proxmox VE

Proxmox VE's headline strength is that the full platform is free, with no feature gating. There is no VM cap, no node cap, and no licence to buy before you can cluster or use high availability — everything ships in the AGPLv3 download. The optional subscription is billed per occupied physical CPU socket per year and gates the enterprise (stable) package repository plus support tickets; the free no-subscription repository runs the same code, with the caveat that Proxmox describes it as intended for testing and non-production use because those packages are less heavily validated. In practice, plenty of production deployments run the no-subscription repo and simply stage updates carefully — but if you want a supported update channel and an SLA, the per-socket subscription is how you get it.

The second differentiator is that Proxmox runs both KVM virtual machines and LXC containers from one interface. That distinction is more useful than it first appears. A KVM VM is a full machine with its own kernel — run Windows, BSD, or any Linux, get complete isolation and live migration. An LXC container shares the host kernel, which makes it dramatically lighter: you can pack far more Linux-only workloads onto the same hardware because there is no per-guest kernel or emulated hardware overhead. The honest rule of thumb is to reach for LXC when the workload is Linux and you want density (internal services, app servers, dev environments), and reach for a VM when you need a non-Linux OS, kernel-level isolation, or a hard security boundary. Being able to mix both on one host, rather than committing the whole machine to one model, is something the other three platforms do not offer in the same first-class way.

On storage, Proxmox gives you ZFS, Ceph, and LVM-thin out of the box. ZFS on a single node buys you snapshots, checksumming, and replication, at the cost of RAM — budget roughly 1 GiB of ARC cache per TiB of pool on top of a couple of GiB base. Ceph is the clustered, hyper-converged option, and this is where expectations need calibrating: Ceph wants at least three nodes and a dedicated 10 Gbps network minimum, with 25 Gbps or more recommended once you are running NVMe. Ceph on two nodes or over a shared 1 Gbps link is a common way to end up disappointed.

The trade-off with Proxmox is ecosystem maturity at the top end. It is a widely adopted platform with a large, active community and is frequently the destination for teams leaving VMware, but it does not carry the same breadth of enterprise ISV certifications that VMware accumulated over two decades. If your compliance regime or a third-party software vendor specifically requires a certified hypervisor, that is worth checking before you commit. For most self-hosted workloads it is a non-issue. When you are ready to run it, our Proxmox VE dedicated servers deploy it as a ready image.

VMware ESXi

ESXi is the platform everyone benchmarks against, and for good reason: vCenter, vSphere HA and DRS, and vSAN are mature, deeply integrated, and carry the widest ISV certification story in the industry. If you are already running a vSphere estate, none of that maturity has gone anywhere.

What has changed is the commercial model, and it is the reason so many teams are re-evaluating. Since Broadcom's acquisition, VMware moved its entire catalogue to subscription-only term licensing and retired perpetual licences. Licensing is now counted per physical core across all hosts, with a 16-core-per-CPU minimum billed even on CPUs with fewer cores, and the product is packaged into bundles — VMware vSphere Foundation (VVF) and the larger VMware Cloud Foundation (VCF). For a small cluster on modern high-core-count CPUs, the per-core minimum and subscription floor can push the annual cost well above what the same hardware cost to license before.

Broadcom did reintroduce a free ESXi with version 8.0 Update 3e, which softens the entry point — but read the small print. The free edition is standalone: it cannot join vCenter, and it cannot be used with third-party backup tools like Veeam because the required APIs are gated. That makes it fine for a lab or a single isolated host, and unsuitable as the foundation of a managed, backed-up production cluster. The realistic read in 2026 is that ESXi remains technically excellent and the right choice if you are committed to the VMware ecosystem — but the licensing is precisely why "what do we move to?" has become such a common question.

XCP-ng

XCP-ng is the open-source Xen-based platform, incubated in the Xen Project and developed by Vates. The current supported line is XCP-ng 8.3 LTS, and its management plane is Xen Orchestra (XO) — currently at the 6.1 release — which handles administration and backup from a single web interface. The model mirrors Proxmox's in spirit: the platform is open-source and free, and Vates sells pro support and subscriptions on top. XO itself you can build and run yourself from source at no cost, or consume as a supported appliance.

Where XCP-ng earns its place is for teams who want an open-source platform but prefer a VMware-like operational model — a clean separation between the hypervisor hosts (organised into pools) and a central orchestrator that manages them. It has a credible storage and backup story: XOSTOR provides LINSTOR-based replicated, hyper-converged block storage, and Xen Orchestra delivers agentless, integrated full and incremental backups and replication. If your team's muscle memory is XenServer or a pooled-host, central-console way of working, XCP-ng will feel more natural than Proxmox's node-centric model. Its community is smaller than Proxmox's or VMware's, which mostly shows up as fewer third-party tutorials and integrations to lean on.

Hyper-V

Hyper-V is Microsoft's type 1 hypervisor, and its case is almost entirely about the Windows ecosystem. If your estate is Active Directory, System Center, and Windows Server, Hyper-V slots in with tooling — Windows Admin Center, Failover Cluster Manager, live migration, Storage Spaces Direct — that your team already knows, with no new vendor relationship to manage.

The licensing model is different from the others because the hypervisor is a role inside a licensed Windows Server rather than a standalone product. That has two consequences worth understanding. First, your Windows Server edition determines your VM rights: Standard grants the right to run two Windows Server VMs per licensed host, while Datacenter grants unlimited — so a dense Windows virtualisation host generally pushes you to Datacenter. Second, the free standalone Hyper-V Server product is discontinued; the last version was Hyper-V Server 2019, and there is no free 2022 or 2025 equivalent. Hyper-V makes the most sense when you are a Windows-centric shop already paying for Windows Server licences and integrating with Microsoft's management stack; it makes the least sense as a way to run predominantly Linux workloads. If Windows is your target, our Windows dedicated servers cover the licensing and configuration.

Choose the right hypervisor

The features overlap enough that the decision usually comes down to a few conditions. Here is the honest version:

• Choose Proxmox VE if you want a genuinely free, open-source platform with no per-socket or per-core licence, you value being able to run full VMs and dense LXC containers on the same host, and you want ZFS or Ceph storage built in. It is the strongest default for self-hosted infrastructure and the most common landing spot for teams leaving VMware. The one thing to verify first is whether any vendor or compliance requirement demands a certified hypervisor.
• Choose VMware ESXi if you are already invested in the vSphere ecosystem, depend on specific ISV certifications or vSAN/DRS features, and can absorb the post-Broadcom per-core subscription. If you are starting fresh with no VMware commitment, the licensing math rarely favours it any more.
• Choose XCP-ng if you want open-source and free but prefer a VMware-style pooled-host operations model with a central orchestrator, or you are migrating from XenServer. Xen Orchestra's integrated backup and XOSTOR make it a coherent package for that workflow.
• Choose Hyper-V if you are a Windows-centric organisation, you are already licensing Windows Server (especially Datacenter for unlimited VM rights), and tight Active Directory and System Center integration matters more than platform cost.

One cross-cutting note on clustering: three of these platforms — Proxmox, ESXi, and XCP-ng — do high availability properly only once you have at least three nodes. Proxmox and Ceph both want three nodes for reliable quorum, and a two-node cluster cannot maintain quorum on its own when a node fails without an external tie-breaker (a QDevice). If you are deploying a single host today, that is fine — just size the network and plan the node count before you commit to HA, rather than discovering the three-node floor after the second server is in production.

Deploying Proxmox VE on Serverside

If Proxmox VE is your pick, Serverside deploys it as a ready image from the OS catalog, provisioned onto bare metal in under a minute — no manual ISO mounting or installer babysitting. You get full root on the whole machine, so repository choice, updates, and cluster configuration are entirely yours.

The hardware suits virtualisation directly: high-core-count AMD EPYC platforms with ECC DDR5 give you the cores and memory headroom that dense VM and container hosts need, and the fast networking matters specifically for the workloads Proxmox is good at — Ceph and ZFS replication traffic are bandwidth-hungry, and this is where a proper uplink between nodes pays off. Always-on DDoS mitigation on our ASN 55285 network sits in front of the box, protecting the management plane and your guests alike.

Ready to build one yourself? Browse our Proxmox VE dedicated servers to deploy the ready image, or follow our step-by-step Proxmox VE setup guide to install and configure it from the ground up.