Virtualization lets you run several servers on a single physical machine. It is the technique that transformed data centers, and the foundation of almost every modern infrastructure. Here is how it works, what it brings, and where the traps are.
Historically, a physical server ran a single application, wasting much of its power. Virtualization adds a software layer that splits a physical machine into several independent virtual machines (VMs). Each VM behaves like a full computer, with its own operating system.
The hypervisor is the software that creates and runs the VMs by sharing the machine's processor, memory and storage. There are type 1 hypervisors, called bare-metal, installed directly on the hardware and faster, and type 2, hosted, running on top of an existing system.
Several approaches coexist: full virtualization, where the guest system needs no change; paravirtualization, where the guest is adapted for better performance; hardware-assisted virtualization, where the processor helps the hypervisor; and kernel-level virtualization, such as KVM on Linux.
Consolidation reduces the number of servers, and therefore space, power, cooling and maintenance. It simplifies cabling and gives great flexibility: you can run Windows and Linux side by side, and create or move a VM in minutes.
Concentrating several workloads on one machine also concentrates risk: a hardware failure can hit several VMs at once. Watch too for the sprawl of unused VMs, which drains resources, and for hypervisor licensing, whose cost can climb fast. Monitoring and discipline are essential.
Public cloud is virtualization at a provider, billed by usage. But for a stable workload, running your VMs on your own hardware in colocation is often cheaper over time, while keeping control. The choice of hypervisor, and its licensing cost, then weighs heavily in the equation.
Type 1, bare-metal, installs directly on the hardware: the choice for production servers, faster and safer. Type 2 runs on top of an existing system: handy for a workstation or a test, less so for production.
As many as its resources (processor, memory, storage) and the applications' needs allow. Cramming in too many degrades the performance of all of them.
The cloud relies on virtualization, but adds automation, usage-based billing and operation by a third party. You can virtualize without the cloud, on your own hardware.
The build-up of unused or forgotten virtual machines that consume resources without producing anything. Regular cleanup and monitoring prevent it.
Written on 1 September 2026.
Estimate my cost → Compare data centers