Hypervisors are the new data centre OS
So why are we virtualizing full OS’s on them?
Disclaimer - I’m the founder of Return Infinity and the lead architect behind the BareMetal exokernel which, very recently, added the functionality to be run as a unikernel. There is some bias ahead.
Introduction
Over the last two decades, data centers have gone through quite the transformation. What began as racks of physical servers, each running a full operating system, has evolved into virtualized infrastructure where workloads are abstracted from physical hardware. Virtualization delivered enormous gains in efficiency and flexibility, but it also carried forward an assumption: every virtual machine should look like a physical server, complete with a general-purpose OS.
That assumption no longer makes sense in today’s environment. The hypervisor, not Linux or Windows, has effectively become the operating system of the data center. Yet we’re still duplicating entire stacks of functionality by running full operating systems inside every VM. It’s time to reconsider that model and look toward a better solution: Unikernels.
The legacy of virtualization
Virtualization became mainstream in the early 2000s - physical servers were underutilized, often running at minimal capacity. Hypervisors like VMware, Xen, and later KVM, enabled multiple workloads to run on the same hardware, securely isolating them from each other while maximizing utilization.
To make this work, each virtual machine was designed to mimic a physical server. It booted its own OS on virtual hardware, ran its own drivers, and provided a familiar environment for applications. At the time, this was the easiest way to "lift and shift" workloads from bare metal into a virtualized world.
But two decades later, workloads and infrastructure patterns have changed dramatically. We’ve shifted from monolithic applications to microservices, from "machine-centric" to "service-centric" models. Yet the way we package workloads inside VMs still reflects the outdated physical server mindset.
Hypervisors as the new OS
At scale, what does the “operating system” of the data centre actually do? It schedules CPU time, manages memory allocation, controls networking, handles storage, and enforces security isolation. These are the classical responsibilities of an operating system kernel.
But in a modern data centre, these duties don’t just live inside guest operating systems - they also live inside the hypervisor. When a VM requests CPU or network access, it’s the hypervisor that decides how and when that happens. In this sense, the hypervisor has become the real OS, not just for a single machine but for the entire data center.
Linux or Windows inside a VM may look like the operating system, but they’re really just software layers sitting on top of the true OS: the hypervisor.
The problem of redundancy
So why keep running full-blown operating systems inside each VM? The truth is, much of what those OSes provide is redundant.
For instance:
- Each guest OS comes with its own scheduler, memory manager, and driver stack - even though the hypervisor also manages these functions.
- Large parts of the OS kernel go unused in application-centric VMs, yet still consume CPU cycles and memory for various background tasks and services.
- Every OS instance introduces operational overhead - patching, upgrading, and securing dozens or hundreds of kernels.
This duplication isn’t just inefficient; it’s also costly and risky. The more OS layers you run, the larger your attack surface and the greater your management burden!
Why unikernels make more sense
Unikernels offer a fundamentally different approach. Instead of running an application on top of a general-purpose OS, a unikernel compiles the application together with only the minimal operating system components it actually needs. The result is a single-purpose VM image that can run directly on a hypervisor.
The benefits are clear:
- Speed: Unikernels boot in milliseconds, enabling near-instant, on-demand, scaling.
- Efficiency: With no unnecessary kernel features or background processes, they consume far fewer resources.
- Security: A smaller codebase means fewer vulnerabilities and less surface area for attacks.
- Simplicity: There’s no need to manage drivers, shells, or packages that your application never uses.
In short, unikernels align with the reality that the hypervisor is the true operating system. They eliminate redundancy by removing the general-purpose OS layer that no longer adds value.
Operational advantages for data centres
The practical implications for operators are significant:
- Higher density: Without OS bloat, more workloads can fit on the same hardware.
- Lower costs: Less overhead translates into better utilization, reduced licensing, and support expenses.
- Faster scaling: Millisecond boot times enable workloads to respond instantly to demand.
- Improved security: With fewer moving parts, the likelihood of zero-day exploits or unpatched vulnerabilities drops dramatically.
- Simplified management: Administrators manage applications, not redundant OS instances.
This isn’t just about theoretical gains. It’s about operational simplicity and real cost savings at scale.
Looking Forward
The trajectory is clear. Data centers are no longer about managing individual machines; they’re about orchestrating services. Software like containers and Kubernetes already reflect this shift in the developer ecosystem, but at the infrastructure level, hypervisors remain central.
The next logical step is for workloads to run as unikernels directly on hypervisors, bypassing the baggage of full general-purpose OS layers. This model treats the hypervisor as the "data centre OS", while workloads become lean, efficient, and secure processes within it.
Over time, orchestration frameworks will likely adapt to schedule unikernels as seamlessly as they do containers today. When that happens, the legacy of treating VMs as "virtual servers" will finally give way to a more modern, service-native approach.
Conclusion
The data centre has already undergone one major transition: from physical servers to virtual machines. But we’re still carrying forward an outdated assumption - that every VM needs a full general-purpose operating system inside it.
The reality is that the hypervisor has become the true operating system of the data center, handling the core responsibilities of scheduling, resource management, and isolation. Running full OSes on top is not only redundant but also wasteful.
Unikernels point the way forward. They strip away the unnecessary layers, leaving only what’s required to run applications efficiently and securely on the hypervisor. Just as we once moved beyond racks of physical servers, we now have the opportunity to move beyond heavyweight guest operating systems.
The hypervisor is the new data centre OS. It’s time our workloads reflected that truth.
-Ian