A Linux Foundation Research study, sponsored by CNCF and Broadcom, explores how organizations are using virtual machines alongside containers, and the growing importance of workload isolation for AI.

Cloud native infrastructure continues to evolve as organizations support an increasingly diverse mix of applications and workloads. While containers and Kubernetes have become important parts of modern infrastructure, virtual machines (VMs) continue to play a significant role in production environments.

New research from Linux Foundation Research, sponsored by the Cloud Native Computing Foundation (CNCF) and Broadcom, provides insight into how organizations are using VMs today, how they expect their infrastructure to evolve, and the role of virtualization in supporting emerging AI and agentic workloads.

The report, Virtual Machines, Agentic AI and the Mix of Workloads: Findings from the 2026 Cloud Native Workloads Survey, draws on responses from 492 cloud native practitioners and leaders surveyed in September 2026.

The findings highlight an important characteristic of today’s infrastructure: organizations are operating a mix of workload types, with VMs and containers frequently forming part of the same broader infrastructure strategy.

Virtual machines remain a significant part of production infrastructure

One of the report’s central findings is the continued prevalence of virtual machines.

Across respondents, VMs accounted for an average of 53% of production applications, with 51% reporting that at least half of their production applications run as VMs.

The research also suggests that this is unlikely to change dramatically in the near term. Among respondents who answered both the current and future workload-share questions, 71% expect the proportion of their production applications running as VMs to remain within the same range over the next 12 months.

Organization size also plays a role. Respondents from organizations with 10,000 or more employees reported an average VM share of 64%, compared with 45% among organizations with fewer than 500 employees.

These findings demonstrate the continued importance of understanding and supporting diverse workload requirements.

Rather than indicating that one deployment model is replacing another, the research describes an infrastructure landscape in which multiple approaches coexist.

VM and container teams are increasingly working together

The research also examines how organizations manage their VM and container environments.

61% of respondents described their VM and container administration teams as collaborative or fully integrated, while 14% reported that the teams operate independently in silos.

This coordination is relevant to the cloud native community as organizations consider how to manage different workload types while maintaining consistent operational practices.

The report also identifies adoption of Kubernetes-based VM management.

Among respondents running VMs, 8% reported using KubeVirt to manage virtual machines on Kubernetes.

KubeVirt, a CNCF project, extends Kubernetes to support VMs as Kubernetes-managed resources, enabling organizations to manage virtualized and containerized workloads through Kubernetes.

The research does not establish how all organizations technically combine their VM and container environments. However, it provides evidence that coordination between the teams responsible for these workloads is already common.

AI and agentic workloads are contributing to demand for virtualization

As organizations explore AI and agentic applications, workload isolation is becoming an important infrastructure consideration.

AI agents may generate and execute code, interact with external tools, and access data or systems on behalf of users. These capabilities introduce security considerations, particularly when organizations need to contain potentially unpredictable code execution.

The survey found that 34% of respondents expanding their VM footprint cited AI and agentic workloads as a contributing factor.

Security and isolation requirements were also cited by 47% of respondents expanding their VM footprint, alongside improved hardware resource utilization (48%) and growth in VM-based applications (53%).

The report explores several approaches organizations use to isolate AI workloads, including hardware-level VM isolation, containers running within VMs, and MicroVM sandboxing.

Among respondents providing valid answers, 62% reported using at least one of these VM-based isolation approaches at the time of the survey.

Looking ahead, respondents indicated growing interest in additional isolation measures:

  • MicroVM sandboxing: 20% reported current use, compared with 30% planning to use it in the next year.
  • Confidential computing: 12% reported current use, compared with 26% planning to use it in the next year.
  • Containers within VMs: 40% reported current use, with 41% planning to use this approach.

These are reported plans rather than confirmed future deployments, but they indicate areas of growing interest as organizations evaluate security requirements for AI workloads.

Supporting different workloads across different environments

The research also highlights the range of environments in which organizations deploy virtual machines.

Among respondents running VMs, 64% reported using on-premises hypervisor platforms, while 51% reported using public cloud infrastructure-as-a-service (IaaS). A further 23% reported using both environments.

Smaller proportions reported running VMs through KubeVirt (8%) or using MicroVM runtimes (2%).

Because respondents could select multiple environments, these results reflect overlapping approaches rather than mutually exclusive deployment choices.

For the cloud native ecosystem, these findings reinforce the importance of accommodating different workload characteristics and infrastructure requirements.

The report also discusses technologies such as Kata Containers, which uses lightweight VMs to provide isolation for containerized workloads.

Together, these approaches illustrate how virtualization and cloud native technologies can be used in different configurations depending on application and operational needs.

What the findings mean for the cloud native community

The findings provide a useful perspective on how production infrastructure is evolving.

Virtual machines continue to account for a substantial proportion of production applications, while organizations are increasingly coordinating the teams responsible for VM and container environments.

At the same time, AI and agentic workloads are introducing additional considerations around workload isolation and infrastructure utilization to drive more VM adoption.

For platform engineers, infrastructure teams, developers, and security practitioners, the research highlights several areas worth exploring: how to coordinate operations across workload types, how to choose appropriate isolation boundaries, and how to manage infrastructure resources effectively.

The results also underscore the value of continued collaboration across the cloud native ecosystem as organizations adopt new technologies while supporting existing applications.

The future of cloud native infrastructure is not necessarily defined by a single workload or deployment model. The findings instead describe an environment in which multiple approaches continue to coexist and evolve.

Explore the full research

The Virtual Machines, Agentic AI and the Mix of Workloads report was authored by Adrienn Lawson of Linux Foundation Research, with a foreword by Dilpreet Bindra, Senior Director of Engineering, VCF Division, Broadcom.

Sponsored by CNCF and Broadcom, the study provides data on production VM adoption, infrastructure environments, organizational coordination, AI workload isolation, and GPU utilization.

We encourage the cloud native community to explore the findings and consider how they relate to their own infrastructure and workload requirements.

Read the full report:

Explore the research data: Linux Foundation Research on Zenodo

Research methodology: The study is based on a web survey of 492 qualified practitioners and leaders involved in building, operating, or securing cloud native infrastructure, conducted in September 2026. Respondents answered on behalf of their organizations. As the report notes, more than one respondent may represent the same organization.

Research source: Adrienn Lawson, Virtual Machines, Agentic AI and the Mix of Workloads, Linux Foundation Research, November 2026.