KVM, OpenStack and Proxmox are not three rival products. KVM is the part of the Linux kernel that runs virtual machines. Proxmox VE and OpenStack are management systems that sit on top of KVM and decide where those machines run, how they get disks and networks, and who may start them. So "OpenStack vs KVM" compares a whole office building with the lift inside it: OpenStack usually uses KVM, and so does Proxmox.
What matters for someone renting a server is the bottom layer. A KVM VPS is a full virtual machine with its own kernel, which is why Docker, WireGuard, custom firewalls and Windows all work on it. A container VPS (OpenVZ or LXC) shares the host's kernel and quietly loses some of that. This article walks up the stack once, in plain words, then shows how to check what you are really renting.
The stack, from the metal up
Every KVM virtual server, whether it comes from a small host or from a hyperscale cloud, is built from roughly the same five layers:
- The hardware. A server CPU with virtualization extensions, Intel VT-x or AMD-V. Without them KVM cannot work.
- KVM. A set of Linux kernel modules (
kvm.kopluskvm-intel.koorkvm-amd.ko) that let the CPU run guest code directly and keep each guest's memory apart. The KVM project calls it a full virtualization solution for Linux on x86 hardware with virtualization extensions. - QEMU. The program that gives each virtual machine its devices: a disk, a network card, a BIOS or UEFI firmware, a graphics card for the console. KVM does the CPU work and QEMU does the pretend hardware. The fast devices are the "virtio" ones, a paravirtual standard published by OASIS that guests recognise as real devices.
- libvirt. A common control layer that starts, stops and describes virtual machines in XML. libvirt speaks to KVM, Xen, LXC and others, and tools such as
virsh, virt-manager and OpenStack's compute service use it. - The manager. Proxmox VE, OpenStack, a hosting control panel, or a shell script. This is the layer with the web interface, the API, the billing hooks and the cluster logic.
When a hosting company says "KVM VPS", it is describing layers 2 and 3. The manager on top changes the provider's daily life far more than yours.
KVM in one paragraph, and why everybody uses it
Avi Kivity announced KVM at the Israeli start-up Qumranet in October 2006, and it entered the mainline Linux kernel with version 2.6.20 in February 2007, as LWN recounts. Red Hat bought Qumranet in 2008 for about 107 million dollars. Because it is part of Linux itself, KVM improves with every kernel release and costs nothing to use.
The biggest clouds run on it too. AWS describes its Nitro System as using "a custom-developed, minimized hypervisor based on KVM" in its security whitepaper, and Google says Compute Engine is built on a hardened KVM. Neither uses stock QEMU, though: both replaced the device layer with their own. Same engine, different body.
Is OpenStack a hypervisor?
No. OpenStack is a collection of services for running a cloud: Nova starts virtual machines, Neutron builds networks, Cinder provides block storage, Glance stores images and Keystone handles logins. Nova itself does not run guests; it asks a hypervisor driver to. The Nova documentation states plainly that "KVM is configured as the default hypervisor for Compute", reached through libvirt. Every OpenStack user survey since the early years has found KVM to be the most used hypervisor, typically in 80 to 90 percent of deployments.
The current release is 2026.2 "Hibiscus", published on 30 September 2026, six months after 2026.1 "Gazpacho". OpenStack earns its complexity when many teams or customers need self-service through an API across dozens of hosts. Our longer piece on OpenStack's advantages and disadvantages in 2026 goes deeper.
And Proxmox VE?
Proxmox VE is a Debian-based system that bundles KVM virtual machines and LXC containers behind one web interface, with clustering, high availability, Ceph or ZFS storage and built-in backups. Its own description is "an open source server virtualization management solution based on QEMU/KVM and LXC", released under the AGPLv3. Version 9.2 arrived on 21 May 2026, built on Debian 13.5 with Linux 7.0 and QEMU 11.0. Subscriptions buy support and the enterprise update channel, not features.
Where OpenStack is built for many tenants and an API, Proxmox is built for one organisation and a person at a browser. That is the honest summary; the full comparison, with versions and sizing, is in Proxmox vs OpenStack.
| KVM | Proxmox VE | OpenStack | |
|---|---|---|---|
| What it is | Kernel hypervisor | Virtualization platform with web UI | Cloud software made of many services |
| Runs VMs itself? | Yes, with QEMU | Through KVM/QEMU, plus LXC containers | Through a driver, usually libvirt with KVM |
| Typical size | One machine | One server to a mid-size cluster | Dozens to thousands of hosts |
| Who uses it directly | Linux admins, every cloud | Companies and hosting providers | Clouds, telcos, research centres |
| Licence | GPL, part of Linux | AGPLv3, optional subscription | Apache 2.0 |
KVM vs OpenVZ vs LXC: the difference you will actually feel
Containers are the other way to sell a "VPS". OpenVZ and LXC do not create a virtual machine at all. They fence off part of one Linux system, so every customer shares the host's kernel. That makes them light and dense, and it brings three practical limits:
- No kernel of your own. You cannot choose your kernel version or load kernel modules. WireGuard is a kernel feature, so on a container VPS it works only if the host has loaded the module for you.
- Docker is awkward. Docker needs kernel features such as overlay storage and network namespaces. The OpenVZ wiki documents Docker inside OpenVZ 7 only as experimental and "not supported in production", and LXC needs nesting switched on by the host.
- Linux only. No Windows, no BSD, no custom appliance images.
OpenVZ also carries its age. OpenVZ 7 runs on a Linux 3.10 era kernel from the Red Hat 7 family, and its commercial twin, Virtuozzo Hybrid Server 7, left maintenance in July 2024 according to the Virtuozzo lifecycle policy. LXC and its newer manager Incus are healthy, modern projects, but the shared-kernel rules apply to them too.
We checked this on our own test machine, a Debian 13 KVM server, with a Debian 13 container running inside it. In the container, modprobe wireguard answered "Module wireguard not found", because a container cannot load modules; on the KVM machine itself, creating a WireGuard interface simply worked. That is the whole difference in two commands.
How to check what your VPS really runs
Three commands, run on the server itself, tell you more than any sales page:
# as any user, on the VPS
systemd-detect-virt
uname -r
lscpu | grep -i 'hypervisor vendor'systemd-detect-virt is the one to trust. On a KVM virtual machine it prints kvm; in a container it prints lxc or openvz. uname -r shows the running kernel; on a container it is the host's kernel, often older than your distribution would normally ship.
Do not rely on lscpu alone. In our container test it still reported "Hypervisor vendor: KVM", because the container sat inside a KVM machine and saw the hardware underneath. Only systemd-detect-virt answered lxc correctly.
What KVM does not promise
KVM separates guests cleanly, but it does not decide how generous the provider is. Two things are worth understanding before you compare plans:
- vCPUs are usually shared. On most VPS plans, including ours, a vCPU is a share of a physical core, scheduled alongside other guests. Short bursts run at full speed; constant 100% load for hours is what bigger plans or dedicated hardware are for. If you suspect a crowded host, run
topand look at thest(steal) value: it is the share of time your vCPU wanted to run but had to wait. - The manager decides the extras. Backups, console access, IPv6, reinstall images and how fast support answers all come from the layer above KVM and from the people running it, not from KVM itself.
Nested virtualization, running your own hypervisor inside a VPS, is a KVM feature the provider has to switch on. If you need it for Android emulators, a test Proxmox or Hyper-V labs, ask before you buy rather than after.
Questions worth asking any VPS host
- Is it a KVM virtual machine or a container? What does
systemd-detect-virtprint? - Can I run Docker and WireGuard without asking for anything to be switched on?
- Can I install Windows or my own ISO, and in which plans?
- Are the disks NVMe, and are backups included or extra?
- Do I get IPv6, and a whole IPv4 address of my own?
- Where exactly is the server, and what latency should my users expect?
Where RS Computers fits
Every RS Computers VPS and VDS is a full KVM virtual machine with its own kernel, NVMe storage, a dedicated IPv4 address, IPv6 and free weekly backups, in Amsterdam, Prishtina and Dublin. Docker and WireGuard work out of the box; our guides to Docker on a VPS and a WireGuard server on Debian 13 start from a fresh server. Windows Server runs on VPS Mini and every VDS plan. If you are moving virtual machines off VMware, our step-by-step guide to converting a VMware VM for a KVM VPS covers the disk formats and drivers.
Frequently asked questions
What is the difference between OpenStack and KVM?
KVM is the hypervisor in the Linux kernel that runs virtual machines. OpenStack is cloud software that manages many servers and tells a hypervisor, by default KVM through libvirt, which virtual machines to start. They work together rather than compete.
Is Proxmox the same as KVM?
No. Proxmox VE uses KVM and QEMU for its virtual machines, and adds a web interface, clustering, storage, backups and LXC containers on top. A Proxmox VM is a KVM VM.
Is a KVM VPS better than OpenVZ?
For most uses, yes: a KVM VPS has its own kernel, so Docker, WireGuard, kernel tuning and Windows work, while an OpenVZ container shares the host's kernel and OpenVZ 7 is built on a Linux 3.10 era kernel. Containers can be cheaper and very light, which suits simple Linux workloads.
How do I know if my VPS is KVM?
Run systemd-detect-virt on the server. It prints kvm for a KVM virtual machine and lxc or openvz for a container. Do not rely on lscpu alone, which can report KVM from inside a container.
Does a KVM VPS reserve CPU cores and RAM just for me?
KVM keeps each guest's memory separate, but whether CPU cores are shared or reserved depends on the provider's plans. On most VPS plans, ours included, vCPUs are shared cores that can burst; the st value in top shows whether you are waiting for the host.
The short version
KVM is the engine, QEMU builds the virtual hardware, libvirt drives it, and Proxmox or OpenStack run the fleet. When you rent a server, ask for KVM, check it with systemd-detect-virt, and judge the rest on disks, network, backups, location and the people answering your messages. If you want to see the difference for yourself, start a KVM VPS and run the three commands above in the first minute.