Welcome back to another edition of Road to KubeCon, where we’re tracking the latest movements in the cloud-native ecosystem as we prepare for KubeCon + CloudNativeCon North America 2026, happening November 9-12 in Salt Lake City, Utah.In this edition, we consider what happens when compute workloads need to run AI on the edge. Underlying Kubernetes mechanics change, as do observability practices and security decisions. We take a look at the emerging tactics and how a couple of key projects are responding to these changes.Check out the story below on opening up Kubernetes capacity, HPE’s advances in purpose-built compute for AI, a reminder to evolve a deprecated Linux kernel interface, and what to expect at this year’s Kubernetes on Edge Day and Cilium co-located events.HPE’s case for purpose-built edge computeUsing edge compute for AI inference makes sense for many reasons — you reduce unnecessary server calls, cut data egress fees, and keep data away from the cloud for compliance. As a result, many enterprises are turning to edge AI for these reasons and more.Hewlett Packard Enterprise (HPE) is a presenting sponsor of Road to KubeCon. HPE Software helps IT organizations modernize infrastructure, streamline operations, and accelerate AI initiatives across hybrid, multi-vendor environments.However, as HPE’s Aaron Lamond notes in a recent HPE blog post, organizations often try to fit a square peg in a round hole when using unoptimized servers. “As intelligence becomes more distributed, purpose-built compute becomes increasingly critical to operational success,” he writes.For Lamond, “purpose-built compute” should consider how hardware, software, security, and operations fit together. He says ProLiant edge servers are fit for the task because they are edge-optimized for resource-constrained environments that require high-grade security.Edge knowledge returns to Kubernetes on Edge DayEdge deployment on Kubernetes has been gaining steam for years, with CNCF projects like KubeEdge and many vendor platforms emerging to support Kubernetes at the edge. This year’s KubeCon + CloudNativeCon will host a dedicated, co-located Kubernetes on Edge Day to explore this intersection of cloud-native and edge computing.As Mars Toktonaliev and Katerina Arzhayev write on the CNCF blog, “The event was created to address the gap between traditional, data-center-centric cloud native approaches and the operational realities of edge computing.” They mention observability and security as particularly relevant themes.The Kubernetes on Edge Day schedule will showcase a broad spectrum of case studies and operational guidance—helpful for engineers running Kubernetes across distributed, resource-constrained environments.Node swap brings Kubernetes density gains up to 3xBursty agentic AI workloads can require significant upfront memory unexpectedly. After this, memory often sits idle. But idle RAM is expensive, and this allocation caps Kubernetes cluster density.On Monday, Ocean Xie and Yuan Wang dive into this issue on the Kubernetes blog. “Memory is often the first hard limit a Kubernetes cluster hits,” they write. “Nodes run out of RAM long before they run out of CPU, and the new wave of agentic AI workloads makes this worse.”They propose running Kubernetes nodes with node swap, a feature that reached general availability in Kubernetes v1.34 and can act as a “shock absorber” during traffic spikes.In their benchmarking, the authors report density gains of up to three times when node swap is backed by fast NVMe SSDs in certain scenarios. The analysis contributes to emerging methods to mitigate the high computational overhead and costs of agentic AI in practice.Self-hosted platform deployments need prep before installThis week, Fairwinds, a managed cloud-native infrastructure provider, shares helpful guidance for those selling self-hosted AI platforms. According to Munib Ali, director of engineering, the installation is only the beginning.Ali argues a working installer isn’t enough. Too often, platform vendors ship and don’t consider the surrounding mechanics required on the customer end—like establishing team responsibilities upfront, setting up IAM, ongoing maintenance, compatibility with customer private clouds, supported DNS configurations, and more.“Self-hosted Kubernetes deployments for AI platforms often stall when customer prerequisites, environment restrictions, and cross-team handoffs are incomplete,” says Ali. Without this in place, the installation date can slip, or the project may never start.Some cloud-native vendors offer customer-hosted delivery for privacy, sovereignty, or control reasons. But to make this seamless and avoid bottlenecks, teams need to figure out a lot before “go-live,” which should arguably be a shared responsibility.As Kubernetes evolves, so do the demands on the teams running it. Presenting sponsor HPE helps teams address that complexity with software spanning virtualization, cloud management, observability, and automation.Kubernetes shifts toward Linux cgroup v2On Tuesday, Paco Xu, open source team lead at DaoCloud, a cloud-native software company, shares thoughts on cgroup v2, a kernel feature for managing resources like CPU and memory. According to him, cgroup v1 has many limitations.“Compared with cgroup v1, cgroup v2 provides a single unified hierarchy, a more consistent interface, and a stronger foundation for resource isolation and modern resource-management features,” writes Xu. Depending on the Kubernetes version and configuration, cgroup v2 supports memory quality of service updates, container-aware OOM handling, rootless support, and other features.Since v1.35, the kubelet refuses to start on cgroup v1 nodes by default. If you’re on an older release, Xu recommends migrating every Linux node to v2 before upgrading. Operators should consider responding to reap the benefits and avoid a stale architecture.Spotlight on recent Cilium community updatesCilium, the CNCF-graduated project for eBPF-based networking, observability, and security, will host another co-located event in Salt Lake on November 9 this year. Attendees can expect sessions to focus on cloud-native AI and security.As Joe Stringer, Isovalent at Cisco, and Google’s Jordan Rife write on the CNCF blog, “this year’s agenda goes straight at the problems that show up when AI and GPU workloads push Kubernetes networking past what it was built for.”The community continues to respond to new GPU demands and AI-driven observability requirements, as well as new vulnerabilities. Plenty of additional interesting case studies have also emerged recently, showing how Cilium, and eBPF for that matter, are getting more production use.“Cilium’s recent case studies with Splunk, Celonis, and Preferred Networks talk about how eBPF is being used for security and AI use cases, and I think are a good indicator of where things are heading,” Bill Mulligan, Cilium and eBPF community pollinator, tells The New Stack.He also points to BpfJailer, an open-source eBPF security tool from Meta. It’s an experimental open-source rewrite of Meta’s internal closed-source eBPF security tool.Hiveminders into Cilium, Tetragon, or eBPF should know: Isovalent is also hosting its epic Hive Mind Mingle in Salt Lake… but sign up soon; attendance is limited.Follow the Road to KubeConRoad to KubeCon is an eight-part series presented by HPE at KubeCon + CloudNativeCon North America in Salt Lake City. Before you go, explore how HPE Software helps IT teams do more with less complexity.If you haven’t bought your ticket yet, The New Stack readers get a special discount courtesy of the Cloud Native Computing Foundation (CNCF). Use the code KCNA26MED10 when you register at this link for 10% off your ticket.Until the main event, we’ll be here every Friday covering what you need to know on the road to KubeCon.In the meantime, you can catch up on past editions covering Kubernetes v1.37, inference costs, OpenTelemetry updates, and cloud-native AI harnesses. Or read the full archive.The writer for this series, Bill Doerrfeld, is open to pitches. Send PRs, updates, projects, quotes, or hot takes via his personal contact page.The post Kubernetes on cgroup v1 is dead. Here’s what comes next. appeared first on The New Stack.