CNCF Survey Reveals VMs Remain Critical for Cloud-Native and AI Workloads, Especially for Isolation

The Cloud Native Computing Foundation (CNCF), in collaboration with Broadcom, has released a new report indicating that Virtual Machines (VMs) continue to be a significant component of cloud-native production infrastructure. The survey, conducted by The Linux Foundation Research, found that VMs host an average of 53% of production applications. Notably, 34% of organizations expanding their VM footprint cite AI and agentic workloads as a contributing factor, driven by the need for strict hardware isolation. This finding is significant for practitioners because it challenges the long-held narrative of containers entirely supplanting VMs in cloud-native environments. Instead, it suggests a more nuanced reality where VMs and containers coexist, each serving distinct purposes. The growing demand for AI workloads, with their inherent requirements for robust isolation and security, is providing a new impetus for VM adoption and expansion. This means that cloud architects and DevOps engineers cannot solely focus on containerization but must maintain expertise in managing virtualized environments, especially when dealing with sensitive or performance-critical AI applications. This trend fits within the broader evolution of cloud-native computing, which increasingly emphasizes workload-specific optimization and hybrid deployment models. While containers offer agility and portability, VMs provide a stronger isolation boundary, which is particularly valuable for multi-tenant environments, regulatory compliance, and securing AI models against potential vulnerabilities like prompt injection or data poisoning. The report also highlights the importance of consistent governance and policy enforcement across both VM and containerized environments, a challenge that platform engineering teams are actively addressing. In practice, this means that organizations should invest in tools and strategies that enable unified management of both VMs and containers. Broadcom's VM API project, recently accepted into the CNCF Sandbox, aims to facilitate this by allowing VMs to be provisioned and managed through Kubernetes APIs. Practitioners should monitor developments in Kubernetes-native VM management and confidential computing, as these technologies will be crucial for building secure and scalable AI infrastructure. Furthermore, understanding the trade-offs between VMs and containers for different types of AI workloads—from training massive models to deploying agentic AI systems—will be paramount for making informed architectural decisions. The report also suggests that organizations with larger employee bases tend to have a higher average VM share, indicating that enterprise-scale operations often leverage VMs for their stability and isolation properties.