KubeVirt Graduates from CNCF as Virtual Machines Find New Purpose in AI Security
The open-source project that merges VMs and containers on Kubernetes has reached a major milestone just as enterprises seek safer ways to run sensitive AI workloads. Confidential computing is reshaping how virtual machines fit into cloud infrastructure.
As organizations evaluate where to deploy their most confidential AI systems, virtual machines are experiencing a resurgence rather than obsolescence. They are now positioned as complementary technology to containers within Kubernetes environments, not as legacy infrastructure destined for retirement.
This moment arrives as KubeVirt, the open-source initiative enabling virtual machines to run inside Kubernetes clusters, achieved graduation status from the Cloud Native Computing Foundation this week. The project's leadership is now charting a course toward AI isolation, agent sandboxes and confidential computing—a mature technology gaining renewed traction. According to Itamar Holder, principal software engineer and KubeVirt maintainer at Red Hat Inc., this shift addresses a critical barrier for many organizations.
I think it's one of the blockers for a lot of companies to join public cloud in general, like companies that really care about super sensitive data. I think it unblocks them to join.
Itamar Holder, Red Hat
Holder and Vladik Romanovsky, technical lead and KubeVirt maintainer at Red Hat, discussed the project's evolution during an exclusive conversation at KubeCon + CloudNativeCon NA, speaking with theCUBE Research's John Furrier and Rebecca Knight.
Confidential computing and the new case for VMs on Kubernetes
The fundamental premise of KubeVirt is allowing virtual machines and containers to coexist on a single Kubernetes cluster without requiring separate infrastructure stacks. This capability has gained importance in the artificial intelligence landscape, with Nvidia Corp. developers contributing to the project's implementation of dynamic resource allocation for GPU hardware, Holder noted. Romanovsky views these two computing models as collaborative rather than competitive.
VMs will do kind of the heavy lifting and larger isolation, and the pods that are running around them would provide a better scaling for cache applications. In general, it belongs together. VMs and containers can run as one application as well.
Vladik Romanovsky, Red Hat
KubeVirt is also exploring applications in AI agent sandboxing, an area where Red Hat is developing supporting infrastructure. The project's maintainers aim to position KubeVirt as an additional isolation backend option for the Kubernetes agent sandbox repository, Holder explained.
https://www.youtube.com/embed/MZJPxSHA5vQ?feature=oembed
I think that we can offer a lot of benefits that other isolation backends don't have. We can use live migration and complex hardware configurations with DRA and a lot of other sweet stuff.
Itamar Holder, Red Hat