CNCF survey: VMs run 53% of production apps, AI isolation drives KubeVirt and MicroVM growth

A Linux Foundation Research survey of 492 cloud native practitioners, sponsored by CNCF and Broadcom, finds that virtual machines account for an average of 53% of production applications and that 71% of respondents expect VM share to remain stable over the next year. The research reveals growing convergence between VM and container operations, with 61% of organizations reporting collaborative or fully integrated administration teams. AI and agentic workloads are emerging as a significant driver for VM expansion, with 34% of those growing their VM footprint citing AI-related needs and 62% of respondents already using VM-based isolation approaches for these workloads.

The survey documents rising interest in advanced isolation technologies: MicroVM sandboxing is planned by 30% of respondents (up from 20% current use), confidential computing by 26% (up from 12%), and containers-within-VMs by 41% (nearly flat from 40%). KubeVirt adoption stands at 8% among VM-running organizations, while deployment patterns span on-premises hypervisors (64%), public cloud IaaS (51%), and hybrid approaches (23%).

VM-container convergence and AI-driven isolation trends

• Workload distribution: VMs average 53% of production applications across surveyed organizations, with 51% of respondents running at least half their workloads as VMs; large enterprises (10,000+ employees) show 64% VM share versus 45% for organizations under 500 employees.
• Team integration: 61% of respondents describe VM and container administration teams as collaborative or fully integrated, while 14% operate in silos, indicating operational convergence between historically separate infrastructure practices.
• Kubernetes VM management: 8% of respondents running VMs use KubeVirt, the CNCF incubating project that extends Kubernetes APIs to manage virtual machines as native resources alongside containers.
• AI isolation adoption: 62% of valid respondents employ VM-based isolation for AI workloads, with current use of containers-within-VMs at 40%, MicroVM sandboxing at 20%, and confidential computing at 12%.
• Planned isolation growth: MicroVM sandboxing shows the strongest growth trajectory (20% to 30% planned), followed by confidential computing (12% to 26%), reflecting heightened security requirements for agentic AI code execution.

Why it matters

These findings validate the continued investment in virtualization infrastructure within Linux and cloud native ecosystems, particularly for projects like KubeVirt, Kata Containers, and emerging MicroVM runtimes. For distribution maintainers and platform engineers, the data supports prioritizing unified VM-container management tooling and kernel features that enable lightweight VM isolation. The AI-driven demand for hardware-level isolation also creates opportunities for open-source confidential computing frameworks and Rust-based MicroVM projects like Firecracker to expand their footprint in production environments.