The next 10 years of KubeVirt
The New Stack Vladik Romanovsky
KubeVirt is pushing deeper into GPUs, confidential VMs, and pluggable hypervisors. The big shift: it wants to stay core-light while letting more pieces snap in around it.
Based on reporting by The New Stack, Vladik Romanovsky — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
KubeVirt started in 2016 at Red Hat with a straightforward pitch: run virtual machines next to containers on Kubernetes. That original job still matters. But the project is no longer just a bridge for legacy workloads. Around it, other projects are already treating KubeVirt as a building block for broader systems, from VM, cluster, and bare-metal services to Kubernetes nodes and edge setups that skip the full control plane.
The scale of the project’s roadmap shows how far it has moved. KubeVirt is now a CNCF incubating project, and the community is working toward graduation. The enhancement trackers for v1.8, v1.9, and v1.10 have swelled from a small queue in v1.6 to more than a hundred items. That’s why the project leans so hard on the KubeVirt Enhancement Proposal process: designs first, code later, and clear ownership through SIGs.
A lot of the near-term work is about hardware. GPUs can already be claimed for VMs through Kubernetes Dynamic Resource Allocation, and that support is beta in v1.9. The same model is being extended to host devices, network devices like SR-IOV, and CPUs, so a VM could ask for all three and have Kubernetes allocate them together. Contributors are also working on a way to generate those claims from the VM spec plus vendor policy, instead of forcing users to hand-write ResourceClaim objects.
Confidential computing is another major thread. KubeVirt is expanding support for confidential VMs and looking at attestation and secret-release services, including Trustee from the Confidential Containers project. The work spans Intel TDX, AMD SEV and SEV-SNP, IBM Secure Execution, and Arm CCA, though they’re at different stages. In v1.9, AMD SEV and SEV-SNP moved to beta, Intel TDX stayed alpha, IBM Secure Execution reached GA, and Arm CCA remains under development.
The project is also trying hard not to bloat its core. One path is structured plugins, which would let users add features outside kubevirt/kubevirt while still keeping KubeVirt’s API guarantees. Another is a pluggable virtualization stack, so alternatives like Cloud Hypervisor, OpenVMM, or Firecracker could live out-of-tree without forcing a fork. KubeVirt’s next decade looks less like a single product release and more like a controlled widening of the platform around it.
My take — AI-written commentary, not fact-checked reporting
This is the right instinct: keep the core tight and let the ecosystem sprawl outside it. Too many infrastructure projects turn into junk drawers because every shiny idea gets welded into the center. KubeVirt seems to have learned that the hard way, which is usually how good platform software gets made.
Read more about this at: The New Stack