KubeVirt 1.8 Introduces Hypervisor Abstraction Layer, Breaking KVM Lock-In for Kubernetes Virtual Machines
The CNCF incubating project's latest release adds a multi-hypervisor backend, Intel TDX attestation for confidential computing, and PCIe NUMA topology awareness for AI workloads, while scaling its test framework to 8,000 virtual machines.
Overview
The KubeVirt project released version 1.8 on March 25 at KubeCon + CloudNativeCon Europe 2026 in Amsterdam, delivering what the community describes as its most architecturally significant update since the project joined the Cloud Native Computing Foundation in 2022. The headline feature is a Hypervisor Abstraction Layer that decouples KubeVirt from its longstanding dependency on KVM, opening the door to alternative hypervisor backends while preserving KVM as the default.
KubeVirt extends Kubernetes to run traditional virtual machines alongside containers using native Kubernetes APIs and the same control plane, targeting organizations with workloads that cannot be readily containerized. The release is aligned with Kubernetes v1.35.
Hypervisor Abstraction Layer
Until version 1.8, KubeVirt was tightly coupled to QEMU and KVM as its sole virtualization backend. The new Hypervisor Abstraction Layer allows integrators to plug in alternative backends while the existing KVM-first behavior remains unchanged. Red Hat’s Vladik Romanovsky first outlined this direction at KubeCon North America 2025, noting the project’s exploration of “support for multiple VMMs beyond QEMU/KVM.”
The architectural shift is significant because it positions KubeVirt as a vendor-neutral virtualization layer within Kubernetes rather than a KVM-specific tool, potentially broadening adoption among organizations using different hypervisor stacks.
Confidential Computing and Performance
The release introduces Intel TDX attestation support, enabling virtual machines to cryptographically certify that they are running on confidential hardware. This capability, developed by the Confidential Computing Working Group, addresses demand from regulated industries handling sensitive workloads.
For AI and high-performance computing users, KubeVirt 1.8 adds PCIe NUMA topology awareness alongside other resource improvements that the community describes as achieving “near-native performance”. The project’s scale testing framework has been expanded to 8,000 virtual machine instances using KWOK, with documented linear memory scaling: virt-api consumes approximately 3.89 KB per VMI while virt-controller requires roughly 173 KB per VMI.
Networking and Storage Improvements
The passt user-space network binding has been promoted from a plugin to a core binding, reflecting its maturity as a networking option. Operators can now live-update Network Attachment Definition references without requiring a VM restart, enabling network changes on running workloads without guest disruption.
KubeVirt has also decoupled from NAD definitions at the virt-controller level, reducing API calls and removing permissions previously required for VM activation at scale. This architectural change improves both performance and security posture for large deployments.
On the storage side, a new ContainerPath volume type maps container paths directly for VM storage, described as an “escape hatch” for cloud provider credential patterns. Changed Block Tracking enables storage-agnostic incremental backups leveraging QEMU and libvirt capabilities, reducing backup windows and storage footprint.
Production Adoption and Outlook
Pure Storage’s Portworx division has launched a dedicated KubeVirt offering and reports running over 5,000 VMs in production with a claimed 50 percent cost reduction versus traditional virtualization platforms. Earlier adopters include Arm, CoreWeave, and Kubermatic.
The release arrives as enterprises continue evaluating alternatives to VMware following Broadcom’s acquisition, with Nutanix separately announcing plans to add KubeVirt support to its bare-metal Kubernetes distribution for edge deployments. KubeVirt currently holds CNCF incubating status, with project maintainers noting that a graduation application is under consideration. The community has stated that the next release “is looking to be larger still.”