Remote Support Start download

Proxmox LXC vs. VM 2026: The Decision Guide After a Year with PVE 9

ProxmoxLXCVM
Proxmox LXC vs. VM 2026: The Decision Guide After a Year with PVE 9

A little over a year ago, Proxmox VE 9 was released, and PVE 9.2 has been running in production for most of our customers for several months now. Two things have changed noticeably in that year: container live migration has grown out of experimental status, and native handling of OCI images finally turns LXC in Proxmox into a serious alternative to classic Docker hosts. So it is time to revisit the old question container or VM on Proxmox – not in theory, but with the eye of someone running production clusters.

This article is the update to our fundamental comparison Docker vs. LXC vs. VM. That piece covers the three virtualization approaches at a general level. Here we focus on the single decision that gets made on a Proxmox host in practice: LXC container or KVM VM. And what is different about that decision in 2026 compared to 2024.

What Really Changed Between Proxmox LXC and VM in 2026

The core architecture is unchanged: a VM boots its own kernel via KVM/QEMU, an LXC container shares the host kernel and isolates only namespaces and cgroups. What has changed are the details that shape day-to-day operations.

Container live migration between nodes. What has been standard for VMs for years, Proxmox has now caught up on for LXC with PVE 9. Combined with shared storage – usually Ceph or a TrueNAS iSCSI/NFS share from our TrueNAS configurator – a running container can now be moved between nodes without downtime. It is not flawless in every edge case, but for planned maintenance windows it is stable.

OCI image support in LXC. With the current 9.x point releases, Proxmox can use OCI images directly as the basis for LXC containers. In practice: instead of building a custom LXC template, an image from an OCI registry can be launched as a container – without the detour through Docker-in-LXC. For many smaller services – reverse proxy, monitoring agent, mail relay – that is a real time saver.

Density per host. With modern mid-market hosts we currently build – two CPU sockets, 256 to 512 GB of RAM, NVMe storage – the point where RAM is sufficient but licensing cost becomes the limiting factor has moved much closer. The more workloads run as LXC instead of VM, the better the hosts amortize.

When the LXC Container Wins

For pure Linux workloads, LXC on Proxmox 2026 is the better choice in most cases. Not without exception, but as the default position.

Infrastructure services. DNS, DHCP, reverse proxy, monitoring agents, log collectors, internal APIs – these are services that boot in seconds as LXC, consume very little RAM, and can be backed up cleanly through Proxmox. A typical nginx reverse proxy with Let’s Encrypt runs in an LXC with 256 MB RAM without any trouble.

Databases without multi-tenant requirements. PostgreSQL, MariaDB, Redis – classic services dedicated to a single application are a good fit for LXC. Near-native performance is measurably noticeable for I/O-heavy workloads, since there is no virtio layer in between. Multi-tenant DBaaS is a different story – that calls for the stronger isolation of a VM.

Dev and test environments. Wherever developers need a quick Linux context, a build agent, or a staging instance, LXC is the honest choice. Snapshots are cheap, container start takes seconds, and templates can be centrally maintained.

Pure container hosts. Where a Docker host with twenty containers used to run, an LXC with OCI support can do the same job in 2026 – without nesting workarounds, without a double management layer. For teams coming from the Docker world, the transition is smaller than it sounds.

When the KVM VM Is Still Required

Despite all the LXC improvements, there are scenarios where a VM is not up for debate.

Windows and anything non-Linux. Windows Server, Windows clients for terminal server environments, BSD-based appliances, pfSense/OPNsense boxes as VMs: LXC cannot run any of these. The KVM VM is the only option here.

Compliance, isolation, tenant separation. Where multiple customers must stay separated on a host – in multi-tenant hosting, in certain GDPR constellations, with separate clients at a tax firm – a dedicated kernel per VM is non-negotiable. A kernel vulnerability that could affect every container on a host simultaneously can never attack the host from inside a VM.

Workloads with their own kernel modules. Anyone running custom kernel drivers, iSCSI target servers with special modules, or older appliance software needs a kernel that belongs to the VM alone. LXC shares the host kernel – touching it potentially breaks everything.

GPU passthrough for dedicated use. For AI inference, CAD workstations running as VMs, or transcoding with an NVIDIA card, VFIO passthrough to a VM is the established path. GPUs can be shared in LXC, but exclusive assignment of a GPU to one workload is simply cleaner as a VM.

Live migration with maximum maturity. Even though container live migration works in production in 2026: KVM live migration has been in use for over a decade, tested in every imaginable scenario, and remains the most robust tool when an entire cluster has to be maintained without downtime. For business-critical workloads with SLA requirements, that still tips the scale.

Container or VM on Proxmox: The 2026 Decision Matrix

CriterionLXC ContainerKVM VM
Linux workload with standard kernelYes, defaultOnly when required
Windows or BSDNot possibleRequired
Compliance with tenant separationOnly with hardeningDefault
Database for a single applicationYes, preferredAlternative
Multi-tenant DBaaSAvoidRequired
Reverse proxy, DNS, DHCPDefaultOverkill
Dev and test environmentYes, defaultOnly when OS variety needed
Exclusive GPU passthroughSuboptimalRequired
Custom kernel modulesNot possibleRequired
Live migration criticalMature but youngMature, decades

OCI Support in LXC: What It Actually Delivers

OCI support is the feature most discussed in 2026 – and it deserves a differentiated view. An OCI image in LXC is not a Docker replacement in the sense that docker-compose files can be adopted 1:1. What it is: a clean way to use standardized images from a registry as the foundation for long-lived, server-style managed containers.

For short-lived microservices with heavy deployment automation, a real Kubernetes cluster or Docker Swarm is still the better answer. For the twenty internal services of a mid-market IT operation that run stably and change rarely, OCI in LXC is the leaner, more manageable path. We build these setups regularly in 2026: a Proxmox cluster, LXC with OCI images for standard services, VMs for the line-of-business applications – all under one UI, under one backup policy.

Container Live Migration: How Mature Is “Mature”?

The second major innovation from PVE 9 is container live migration. After a year of practice, our honest assessment: for planned maintenance windows it is reliable, for unplanned failover scenarios it is not yet as stress-free as VM live migration. Network setups with many open connections, active SMB shares, databases with long transactions – all of that can cause occasional hangs that a clean VM migration would not know.

For HA scenarios where a node fails outright, the classic path still applies: Pacemaker-style policies in Proxmox, containers start on the backup node, short downtime included. For maintenance – kernel update, hardware swap, Ceph rebalance – container live migration is mature enough for most environments. If you are unsure which pattern fits your cluster, our article on VM live migration without shared storage offers the related fundamentals.

Practical Example: A Mid-Market Cluster in 2026

A typical setup we have built several times this year: three Proxmox nodes with local Ceph, 256 GB RAM per node, 25 GbE backbone, and a central TrueNAS M30 for large file shares and backups. Between 60 and 90 workloads run on each cluster. Roughly 70 percent of those are LXC containers – reverse proxies, monitoring, DNS, internal APIs, databases, backup agents. The remaining 30 percent are VMs: Windows Server for domain controllers and terminal services, one OPNsense per site, line-of-business applications with specific kernel requirements, and two or three Linux VMs for services that need extra isolation.

The energy and hardware bill of this split is noticeably better than two years ago, when the same workload count would have meant roughly 50 percent more VMs. For customers that means: fewer CPU sockets, fewer RAM modules, lower surrounding licensing cost – without compromising isolation where it actually matters.

Advice and Implementation

The LXC-or-VM decision is usually not a question of principle but of per-workload fit. If you have an existing Proxmox cluster and want to know which VMs could cleanly become containers – or which new workloads are better off as a VM – that is exactly our everyday work. As an IT provider for Proxmox around Neuburg, we accompany these consolidations regularly, from architecture through migration to day-two operations.

If at the same time you are rethinking the storage layer – because Ceph is too expensive, the old SAN contract is expiring, or performance simply no longer fits – our TrueNAS support covers the storage side.

FAQ

Is it still worth moving from VMs to LXC containers in 2026?

For pure Linux services, yes in most cases. Container live migration and OCI support have defused the last two big arguments against LXC in production. For Windows, GPU passthrough or hard tenant separation, VMs remain the answer.

Can I simply convert an existing Linux VM into an LXC container?

A P2V-style migration via pct restore from an rsync snapshot technically works. In practice there are catches: kernel modules in the old VM, systemd services with kernel access, hard-coded network configuration. In most projects a clean rebuild from an LXC template pays off more than a 1:1 conversion.

How secure are LXC containers compared to VMs?

LXC containers on Proxmox run unprivileged with an AppArmor profile by default, which significantly reduces the attack surface. Still: a compromised kernel potentially affects every container on a host. VMs isolate the kernel as well – for multi-tenant scenarios, that remains the gold standard.

Does container live migration work without shared storage?

Not reliably. It really needs shared storage such as Ceph or a shared NFS/iSCSI share. For setups without shared storage, offline migration with a short downtime is the honest path.

What does OCI support in LXC concretely mean?

Proxmox can use images from an OCI registry as the basis for LXC containers – similar to how Docker pulls images. The container then runs as a classic LXC, not as a short-lived Docker container. A clean alternative for long-lived services, but Docker and Kubernetes remain better suited for dynamic microservice deployments.

What role does TrueNAS storage play in this?

For a Proxmox cluster with many LXC and VM workloads, a reliable external storage layer is half the battle. ZFS-based TrueNAS appliances deliver the snapshots, replication, and performance a mid-market cluster needs – natively attached to Proxmox via NFS and iSCSI. Which M-series fits is quickest to clarify with the TrueNAS configurator.


Planning a new Proxmox cluster or thinking about rebuilding an existing one as LXC-first? Get in touch – we have been building these environments for mid-market customers for years and will tell you honestly where LXC wins and where the VM should stay.

More on these topics:

Need IT consulting?

Contact us for a no-obligation consultation on Proxmox, OPNsense, TrueNAS and more.

Get in touch