As KubeCon + CloudNativeCon North America 2026 approaches, the conversation around Kubernetes is moving well beyond containers and web-scale applications. Platform teams are now being asked to support VMs, AI inference, and emerging agentic workloads, all while GPUs, memory, power, and data center capacity remain constrained. In this VMblog pre-show Q&A, Natalie Fisher, Principal Cloud Native Strategist for VMware Cloud Foundation at Broadcom, makes the case that these demands can be met without building a separate infrastructure stack for each new workload type. She also takes on a common misconception head-on: that VMware cares only about virtualization.
Fisher discusses how VMware Cloud Foundation (VCF) and VMware vSphere Kubernetes Service (VKS) help organizations manage GPU capacity, memory, performance, and cost, and why infrastructure efficiency may become a defining challenge as inference grows. She also explores the governance, identity, and auditability questions raised by autonomous agents, and how platform engineering can make the secure path the easiest one for developers. Fisher highlights Broadcom’s open source contributions, including the donation of Velero to the CNCF and the acceptance of VM API into the CNCF Sandbox, and shares details on Broadcom’s KubeCon plans, including Booth #129, keynote and Expo Theater sessions, and customer events in Salt Lake City.
++
VMblog: What’s your elevator pitch for KubeCon? If attendees only remember one thing about your company after visiting your booth, what should it be?
Natalie Fisher: We want attendees to walk away knowing that we are deeply invested in the future of the cloud-native ecosystem.
Kubernetes is becoming the foundation for an increasingly diverse set of workloads: traditional cloud-native applications, VMs, AI workloads, and, increasingly, agentic applications. We’re bringing Kubernetes together with the compute, networking, and storage capabilities enterprises already depend on so they can support new workload types without building a separate infrastructure stack each time.
For platform engineers and infrastructure teams, the goal is to give them a consistent foundation they can build on as application and infrastructure requirements continue to evolve.
If I had to correct one misconception, it’s that VMware only cares about virtualization and VMs. Virtualization is part of our DNA, but the reality is that our customers aren’t running only one type of workload anymore. That’s why we’ve invested so heavily in Kubernetes and containers alongside virtualization. We want customers to have a common private cloud infrastructure foundation across their workloads rather than creating another infrastructure silo every time the application architecture changes.
VMblog: Beyond the booth presence, what’s your broader community engagement strategy for KubeCon? How are you contributing to the cloud-native ecosystem’s growth?
Natalie Fisher: We have been contributing to Kubernetes for more than a decade, and Broadcom remains one of the leading contributors to Kubernetes and projects across the CNCF ecosystem. To us, community engagement means more than consuming open source or contributing code. It also means helping create projects and communities that can grow beyond any one company.
Velero is a great example. Earlier this year, we donated Velero to the CNCF, where it was accepted as a Sandbox project under vendor-neutral, community-driven governance. We remain actively involved in its development and roadmap, with several of our engineers serving as maintainers. For Velero 1.19, we’re contributing work around block-based data movement, Changed Block Tracking, and more efficient incremental backup and restore. Velero is widely used for Kubernetes backup and recovery, and we’re excited to continue contributing while helping the community move the project toward its next stage of CNCF maturity.
VM API was also recently accepted into the CNCF Sandbox. VM API provides a hypervisor-agnostic Kubernetes API for managing virtual machines, giving the broader community an open foundation it can build on and extend.
Beyond those projects, we continue to contribute to etcd, Cluster API, Harbor, Kubernetes SIGs, and other areas across the ecosystem. And contribution isn’t only code. Our employees participate as maintainers, reviewers, SIG members, program committee members, community organizers, and speakers.
One of the things I value most about working in open source is getting to work alongside companies and individuals with very different experiences and ideas about where the ecosystem should go next. Those different viewpoints help drive innovation and standardization across the ecosystem, and ultimately give users more freedom to choose the technologies and approaches that best fit their needs.
VMblog: The industry narrative has shifted from Kubernetes for web scale to Kubernetes for AI scale. How has your technology evolved to meet that shift, particularly as inference is projected to overtake training as the dominant AI compute workload?
Natalie Fisher: AI infrastructure can feel like building on sand because architecture decisions, models, and requirements keep shifting. The move from web scale to AI scale changes the infrastructure conversation. For AI, particularly inference, running the workload is only the first hurdle. The harder question is whether you can run it efficiently at scale.
GPUs are expensive, and power and data center capacity are constrained. Inference also introduces workloads whose demand can change dramatically over time. That makes utilization, scheduling, and infrastructure efficiency increasingly important.
This is where Kubernetes and virtualization come together. VKS provides a CNCF-conformant Kubernetes foundation for AI workloads, while VCF provides the infrastructure capabilities underneath it. Organizations can leverage Kubernetes and a rapidly evolving AI ecosystem without sacrificing the resource management, isolation, and operational capabilities they rely on from virtualization.
As inference grows, I think infrastructure efficiency will become one of the defining challenges for AI platforms. Organizations that can get more useful work from the infrastructure and accelerators they already own are going to have a meaningful advantage.
VMblog: How is VCF and VKS helping organizations manage GPU capacity, memory, performance, and cost?
Natalie Fisher: We are actively working with DRA, the Dynamic Resource Allocation effort, to scale Kubernetes workload deployments and connect that work to our scheduling features in VCF. That work touches four areas: GPU capacity, memory, performance, and cost.
For capacity, vSphere can allow multiple containers or VMs to share a single physical GPU through time-slicing or NVIDIA Multi-Instance GPU (MIG), while DRS dynamically places workloads on available hardware. VKS exposes GPUs as native Kubernetes resources, so developers can request fractional or dedicated instances using standard YAML and the nvidia.com/gpu resource type.
For memory, MIG slicing divides a physical GPU such as an H100 or A100 into isolated instances with dedicated high-bandwidth memory, helping prevent contention in multi-tenant inference. VCF also supports NVMe memory tiering, which can offload colder host memory to NVMe and free system DRAM for data preprocessing and model embeddings.
On performance, VMs and pods can achieve near-bare-metal GPU throughput using Dynamic DirectPath I/O and vGPU drivers. We support NVLink and NVSwitch for multi-GPU scaling, as well as PVRDMA with NVMe-oF networking to help bypass CPU bottlenecks during distributed training.
For cost, fractional GPU utilization and memory tiering can increase workload density per host, reducing the hardware footprint. Running containers via VKS inside VCF consolidates management and can lower licensing costs compared with fragmented bare-metal clusters. Running AI workloads on-premises can also reduce cloud GPU markups, data egress fees, and API token costs.
One example is ThinkOn’s work with the Canadian federal government. ThinkOn used VCF as the foundation for moving CanChat, an internally developed AI chat application, from a hyperscale cloud to a sovereign cloud with controls for different classes of government information. The project achieved an authority to operate in months rather than years, and approximately 20,000 federal employees were using the service at the time of the presentation. It illustrates how organizations can extend existing infrastructure expertise into private AI while maintaining control over sensitive workloads.
VMblog: Agentic AI workflows are emerging as a new class of Kubernetes workload with autonomy and orchestration requirements that differ from traditional services. How is your platform adapting to support these agentic systems?
Natalie Fisher: Agentic AI is interesting because it changes the infrastructure conversation. An agent isn’t simply another application. Agents can operate autonomously, invoke tools and APIs, maintain context across multiple steps, and act on behalf of a user. That creates new requirements around identity, isolation, governance, observability, and resource consumption.
If an agent is calling tools through an MCP server, interacting with other agents, or creating workloads within the infrastructure itself, how do you audit those actions? Where does that audit trail lead? How do you establish what that agent is actually authorized to do?
VCF Private AI Services provides a private cloud foundation for AI alongside traditional VM and Kubernetes workloads. But I think the bigger shift is what happens when the infrastructure consumer is no longer always a human developer.
As agents begin requesting resources and interacting with other systems, platforms are going to need clear boundaries around what those agents can access and what actions they can take. Emerging standards such as Model Context Protocol will make those interactions easier, but they also make governance increasingly important.
We’re still early in this transition, and the industry is still defining many of these patterns. Most organizations aren’t operating millions of autonomous agents today, but I think platform teams should start thinking about the infrastructure implications now. Identity, policy, observability, auditability, and the ability to manage rapidly growing numbers of workloads need to become part of the platform architecture before scale makes them much harder problems to solve.
VMblog: Security, compliance, and governance remain top concerns, especially as AI systems introduce new supply chain and runtime risks. What’s your approach to helping organizations build secure, compliant cloud native systems without sacrificing velocity?
Natalie Fisher: Security works best when the secure path is also the easiest path for developers. I think that’s one of the most important roles of a platform team: build the security controls into the platform rather than asking every application team to independently solve the same problems.
Within VCF, we approach security at multiple layers. At the infrastructure layer, capabilities around workload isolation, micro-segmentation, and encryption help establish strong boundaries around workloads and data. At the Kubernetes layer, platform teams can incorporate policy and governance into standardized environments so developers don’t have to become experts in the underlying controls just to deploy an application.
AI is going to expand that challenge. Organizations now have to think not only about application and container supply chains, but also models, enterprise data, tools, and the systems agents are permitted to access. Private AI capabilities within VCF give organizations greater control over where those workloads and sensitive data operate.
This is where I think platform engineering becomes especially important. Security can become part of the golden path. Developers and data scientists can move quickly while platform and security teams establish the guardrails underneath them. You aren’t sacrificing velocity for security; you’re designing the platform so the two work together.
VMblog: Give us your executive pitch. How do you articulate ROI and business value to C-suite leaders who need to justify cloud native and AI infrastructure investments?
Natalie Fisher: Executives are asking a harder question now: how do we get more value from increasingly expensive infrastructure?
The conversation used to focus largely on adopting Kubernetes or modernizing applications. Now it also has to account for real constraints around GPUs, memory, power, and data center capacity. Companies can’t control the price of every component or predict every future workload, but they can control how efficiently they use the infrastructure they already own.
That’s where cloud-native infrastructure and virtualization create significant value together. Kubernetes provides a common application and orchestration layer, while virtualization drives utilization, isolation, and operational efficiency underneath.
I frame the ROI around flexibility and efficiency. Increase utilization of existing investments. Reduce unnecessary infrastructure silos. Give teams a consistent operating model. And preserve the ability to adopt new technologies without rebuilding the infrastructure foundation every time the industry changes.
That last piece is particularly important. Re-platforming is expensive. We don’t know exactly what the next workload or architecture will look like, especially with how quickly AI is evolving. A flexible infrastructure foundation gives organizations room to evolve without continually starting over.
VMblog: Are you organizing any exclusive customer events, partner meetings, or community gatherings at KubeCon?
Natalie Fisher: Absolutely. We’ll be at Booth #129 on the Expo floor with demos and plenty of opportunities to connect with our engineers, maintainers, and cloud-native team. Broadcom is a Platinum Sponsor of KubeCon + CloudNativeCon, and we’re also a Diamond Sponsor of Cloud Native AI + Inference Day on November 9.
On Monday, November 9 at 10:10 a.m. MST, we’re participating in the Cloud Native AI + Inference Day keynote, “Path to Resilient App Delivery: AI is Hard, Production is Harder,” at the Salt Palace.
On Tuesday, November 10 at 2:35 p.m. MST, we’ll have an Expo Theater session, “One Kubernetes API for VMs, without rebuilding the infrastructure beneath it.”
And on Wednesday, November 11, we’re hosting two events at Matteo Ristorante Italiano. From 4 to 6 p.m., we’ll have a closed-door executive roundtable on “Scaling Secure and Governed AI Platforms.” Then from 6 to 9 p.m., we’re hosting a reception and panel discussion, “Building The Agentic Future,” with food, drinks, and conversation. Registration is required for both events.
We’ll also have several community-led sessions and demos throughout the week. Full details, including registration information, are available in our KubeCon event blog: “AI, Kubernetes, and Cloud Native Innovations: Broadcom at KubeCon North America 2026.” For details on all of these, including how to register, you can check out our blog at https://blogs.vmware.com/cloud-foundation/2026/10/12/ai-kubernetes-and-cloud-native-innovations-broadcom-at-kubecon-north-america-2026/
