VMware to KubeVirt: what carries over and what does not
A feature-by-feature look at moving vSphere workloads to Kubernetes with KubeVirt, including the four areas where parity is only partial.
Most teams looking at a VMware exit ask the same first question: what do we lose? The short answer is less than you might expect. KubeVirt runs virtual machines on Kubernetes, next to containers. Each VM runs in a pod as a QEMU process on KVM, so VMs and containers share one scheduler, one storage layer and one network.
Your VMs keep their disks, operating system and network settings, so the applications inside them run unchanged.
What carries over one to one
| Capability | On VMware | On Kubernetes with KubeVirt |
|---|---|---|
| Hypervisor | ESXi | KVM |
| Live migration | vMotion | Between nodes on shared storage |
| Storage live migration | Storage vMotion | Volume migration while the VM runs |
| High availability | vSphere HA | VMs restart on a healthy node |
| Guest tools | VMware Tools | QEMU guest agent and VirtIO drivers |
| Distributed storage | vSAN | Replicated block storage |
| External SAN and NAS | VMFS, NFS datastores | CSI drivers: NetApp, Pure, Dell, HPE |
| Virtual networking | Distributed switch | Cilium, existing VLANs kept |
| Backup | Veeam, VADP tools | Velero, Veeam Kasten or Veeam plug-in |
| Monitoring | Aria Operations | Prometheus and Grafana |
| Automation | PowerCLI and Terraform | Kubernetes API, GitOps and Terraform |
CPU and memory hot-add, templates and clones, snapshots, resource pools and quotas, roles with single sign-on, and GPU passthrough and vGPU are all supported as well.
Where parity is partial
Four VMware features have no exact equivalent today. It is better to know them before you start than to discover them during cutover.
- Workload balancing (DRS). The Kubernetes descheduler rebalances VMs periodically rather than continuously.
- Fault tolerance (vSphere FT). QEMU COLO exists but is experimental and not orchestrated. Most teams rely on high availability with fast restart instead.
- Micro-segmentation (NSX). Cilium network policies cover most segmentation needs, but NSX's full feature set is broader.
- Orchestrated disaster recovery (Site Recovery Manager). Second-site replication works; failover is manual rather than a single orchestrated runbook.
How the move itself works
The migration runs in four steps:
- Assess. A vCenter inventory export, plus network and storage mapping.
- Copy. Forklift copies VM disks to the KubeVirt cluster while the VMs keep running, and keeps them in sync until switchover.
- Switch over. In a window agreed with your team, in batches.
- Verify. The original VM stays on VMware, switched off, until you sign off.
Existing VLANs, IP addresses and SAN or NAS storage carry over, so the network and storage teams do not have to redesign anything on day one.
Where the cluster can run
KubeVirt does not care where the servers are. The same setup runs on in-house server clusters (including the hosts you run VMware on today, after migration), in air-gapped environments with an internal registry and mirrored updates, in colocation, or on rented bare metal such as Hetzner.
A good first step
Send an RVTools export or a list of hosts and VMs, plus any renewal quote, to contact@cloud-exit.com. That is enough to map your estate against the table above and quote a fixed price.
This guide is backed by a service we run every day.
How our VMware migration works