New data from a Linux Foundation Research survey of 492 cloud‑native practitioners reveals that virtual machines (VMs) continue to run the majority of production applications—averaging 53%—and that this proportion is unlikely to shift dramatically in the next year. At the same time, AI and agentic workloads are prompting organizations to expand VM footprints for isolation, prompting engineers to reassess architecture, operations, and security practices.
VM share remains stable across organization sizes
More than half of respondents (51%) reported that at least half of their production apps run as VMs. When asked about future expectations, 71% said the VM share would stay within the same range over the next 12 months. Larger enterprises (10,000+ employees) reported a higher average VM share (64%) compared with smaller firms (
VM and container teams are converging
Collaboration between VM and container administration groups is now the norm: 61% described their teams as collaborative or fully integrated, while only 14% operate in silos. This convergence is reflected in the modest adoption of Kubernetes‑based VM management—8% of VM users reported using KubeVirt to treat VMs as Kubernetes resources.
AI workloads are accelerating VM‑based isolation
AI agents that generate and execute code introduce new security concerns, leading 34% of respondents expanding their VM usage to accommodate AI and agentic workloads. Security and isolation requirements were cited by 47% of those expanding VM footprints, alongside hardware utilization (48%) and growth in VM‑based applications (53%).
Practitioners are already employing VM‑centric isolation methods: 62% use at least one of hardware‑level VM isolation, containers inside VMs, or MicroVM sandboxing. Interest is growing in specific techniques:
- MicroVM sandboxing: 20% current use, 30% plan to adopt within a year.
- Confidential computing: 12% current use, 26% plan to adopt.
- Containers inside VMs: 40% current use, 41% plan to adopt.
Deployment environments remain diverse
VMs are spread across on‑prem hypervisors (64% of respondents) and public‑cloud IaaS (51%). A notable 23% run VMs in both environments, indicating hybrid strategies are common. Smaller fractions use KubeVirt (8%) or MicroVM runtimes (2%).
Related CloudNinjas coverage: hands-on guides.
What This Means For Practitioners
AI engineers should evaluate VM‑based isolation options—especially containers‑in‑VM and MicroVMs—to contain unpredictable code execution. Cloud and platform engineers need to support mixed workloads by ensuring tooling, monitoring, and CI/CD pipelines work across both VMs and containers, and consider KubeVirt where a unified Kubernetes control plane is desired. DevOps/SRE teams must align processes between VM and container groups, sharing observability data and incident response playbooks. Security engineers should prioritize assessments of isolation mechanisms, plan for microVM or confidential computing adoption, and incorporate VM‑level hardening into threat models for AI workloads.
Going forward, watch for increased adoption rates of KubeVirt, MicroVM runtimes, and confidential computing solutions as organizations seek tighter isolation without sacrificing the flexibility of container orchestration. Evaluating these technologies now can smooth future transitions and reduce the operational friction of managing a heterogeneous workload landscape.
