New research: Virtual machines remain central to cloud native

The Linux Foundation Research division, in partnership with the Cloud Native Computing Foundation and Broadcom, published a comprehensive market study detailing how enterprise infrastructure teams deploy virtual machines alongside containerized environments. According to CNCF Blog's report, the study titled Virtual Machines, Agentic AI and the Mix of Workloads examines operational data from 492 cloud native practitioners surveyed in September 2026. The findings show that virtualization remains an active foundation for enterprise operations, driven by complex application demands and emerging workload isolation requirements rather than disappearing in favor of pure container stacks.

This analysis breaks down the survey findings for infrastructure architects, platform engineering leads, site reliability specialists, and enterprise security teams who evaluate cloud native roadmaps. It details current workload distributions between virtual machines and containers, reviews team collaboration patterns across IT operations, and outlines why autonomous artificial intelligence systems are renewing enterprise demand for hardware-level process isolation. The data provides concrete operational benchmarks across enterprise company sizes, tracking how practitioners balance on-premises hypervisors, public cloud platforms, and Kubernetes-native virtualization technologies across production systems.

New research: what actually changed

The September 2026 survey authored by Adrienn Lawson reveals that virtual machines still host the majority of enterprise cloud workloads. Across all 492 vetted practitioners and engineering leaders surveyed, virtual machines accounted for an average of 53 percent of production applications. Furthermore, 51 percent of participants reported that at least half of their entire application fleet runs within virtualized environments. Rather than representing legacy technical debt awaiting migration, these workloads reflect deliberate deployment choices where organizations match specific runtime guarantees, storage persistence, and networking characteristics to the underlying virtualization stack.

Operational planning data shows that this distribution will remain steady across enterprise production environments over the coming year. Among respondents answering comparative workload-share questions, 71 percent expect their current proportion of virtual machine applications to remain within the same range across the next 12 months. Enterprise scale sharply amplifies this operational footprint. Organizations employing 10,000 or more people reported an average virtual machine share of 64 percent, whereas organizations with fewer than 500 employees reported an average virtual machine allocation of 45 percent.

New research: how it works

Platform engineering groups manage these dual runtimes by breaking down administrative divisions between operational units. The report found that 61 percent of respondents describe their virtual machine and container administration teams as either collaborative or fully integrated into unified engineering teams. In contrast, only 14 percent of surveyed organizations reported that their infrastructure groups operate as isolated silos. This collaborative operational structure allows engineering teams to deploy centralized tooling, standardize configuration automation, and maintain uniform security posture across disparate compute substrates without introducing duplicate administrative overhead.

Infrastructure environments span both traditional data centers and public cloud infrastructure platforms. Among organizations operating virtual machines, 64 percent rely on on-premises hypervisor software, 51 percent run on public cloud infrastructure-as-a-service platforms, and 23 percent run workloads simultaneously across both environments. For cloud native convergence, 8 percent of respondents operating virtual machines deploy KubeVirt, a CNCF project that allows engineers to manage virtual machines natively as Kubernetes API resources. Additionally, 2 percent use specialized MicroVM runtimes, while projects like Kata Containers provide lightweight hypervisor isolation directly beneath standard container interfaces.

New research: why it matters

The emergence of artificial intelligence agents that write, evaluate, and execute software code is actively accelerating enterprise demand for virtualization. The survey revealed that 34 percent of respondents expanding their virtual machine footprint cited artificial intelligence and agentic workloads as a primary driver. Autonomous agents require access to local tools, corporate network interfaces, and arbitrary code interpreters. Because these dynamic operations introduce unpredictable code execution patterns, platform security teams are deploying hypervisor-level boundaries to prevent compromised agent tasks from breaching adjacent multi-tenant processes.

Security and workload isolation requirements were cited by 47 percent of organizations expanding their virtual machine footprint, while 48 percent pointed to improved hardware resource utilization and 53 percent cited general growth in virtual machine applications. Overall, 62 percent of respondents with valid answers confirmed they currently deploy at least one virtual machine isolation architecture to protect runtime environments. By encapsulating non-deterministic model processes within hardened hypervisor contexts, platform operators prevent unauthorized privilege escalation and lateral network traversal across production data clusters.

New research: who is affected

Large enterprises with complex regulatory obligations and thousands of internal applications experience the most direct operational impact from these architecture requirements. For companies with more than 10,000 staff members, managing an average 64 percent virtual machine distribution requires continuous alignment between legacy virtualization software, bare-metal hardware pools, and modern container platforms. Infrastructure leaders at these organizations cannot simply mandate containerization; instead, they must invest in unified control planes that can reliably orchestrate both container pods and traditional hypervisor guests without sacrificing operational compliance.

Platform engineers and site reliability teams supporting artificial intelligence initiatives must also adjust runtime infrastructure designs. Because 40 percent of respondents currently operate containers inside virtual machines, operations teams must balance virtualization overhead against enterprise isolation mandates. Rather than treating Kubernetes and hypervisors as competing deployment options, infrastructure engineers are required to master hybrid management patterns that combine declarative container APIs with strong hypervisor isolation boundaries across both private data centers and hyperscaler cloud regions.

New research: what to watch

The study outlines deliberate multi-year shifts toward specialized sandboxing technologies designed to secure high-risk workloads. Over the next 12 months, planned adoption of MicroVM sandboxing is projected to rise from 20 percent of respondents to 30 percent. Interest in hardware-rooted confidential computing is experiencing an even steeper growth trajectory, with planned adoption climbing from 12 percent current implementation to 26 percent over the coming year. Meanwhile, traditional architectures running containers within virtual machines remain steady, moving from 40 percent current adoption to 41 percent planned usage.

As organizations explore agentic capabilities, the cloud native ecosystem will increasingly rely on technologies that bridge hypervisor isolation and declarative container orchestration. Platforms such as KubeVirt and Kata Containers demonstrate how virtualization layers adapt to cloud native operational workflows without demanding total infrastructure rewrites. Enterprise leaders like Broadcom senior engineering director Dilpreet Bindra and CNCF maintainers are tracking how these emerging isolation runtimes manage high-density GPU allocation, secure sensitive model weights, and maintain low execution latency across increasingly hybrid infrastructure deployments.