Industry executives and experts share their predictions for 2026. Read them in this 18th annual VMblog.com series exclusive.
By Nico Vibert, Technical Marketing Engineer Director, Isovalent
Kubernetes networking is entering a new phase as organizations prepare their infrastructure for life beyond VMware and for the rapid rise of AI driven workloads. The network is no longer only a transport layer. It has to serve as the bridge between the legacy world that teams are migrating away from and the modern platforms they are building toward. eBPF and Cilium continue to shape how these teams think about connectivity, security, and observability.
In last year’s predictions I highlighted how Kubernetes networking was beginning to adjust to AI workloads, with new patterns emerging around topology awareness, bandwidth demands, and smarter datapaths. The pace of that shift has only increased. Many of the early signals from 2025 are now becoming day-to-day operational realities, and they influence every prediction for 2026.
Here are the trends that will matter most in 2026.
Running VMs on Kubernetes
Interest in running VMs on Kubernetes keeps accelerating, especially as many organizations rethink their long-term dependence on VMware. The idea of putting containers and VMs onto one platform with one network and security model is compelling. The momentum will keep growing in 2026, but so will the realization that this path is still harder than it looks.+
The operational model is very different from VMware. Day-to-day lifecycle management takes more skill and patience. Networking issues are still where the pain shows up first, especially around performance, predictability, and identity preservation during migrations. The ecosystem is working quickly to close these gaps, with kernel improvements, more capable datapaths, and better identity continuity. But the learning curve will remain steep, and teams will have more realistic expectations about what it takes to move legacy workloads into Kubernetes.
Improving the KubeVirt Experience
KubeVirt is becoming the go-to way to run VMs inside Kubernetes, but the user experience is still more complex than most teams expect. Operators who are used to vCenter find the world of YAML, CRDs, and CLI-driven workflows fragmented and unfamiliar.
Networking is still where most people feel the pain. IPs move around more than operators expect, multi tenancy takes real planning, and keeping a VM’s identity stable during a migration is still hit or miss. There will be solid progress in 2026, especially around setup, upgrades, and making network behavior less hacky. But even with those improvements, running KubeVirt is not a push-button experience. It takes people who know what they are doing and who are willing to work through the rough edges.
The Rise of the Kubernetworker
Network engineers are stepping back into Kubernetes environments, and that trend will accelerate in 2026. Large clusters now require routing, segmentation, and traffic engineering skills that go far beyond what most platform teams are used to. As the recent State of Kubernetes Networking report indicated, it’s platform teams that are in charge of Kubernetes networking. But that model will change. Modern networking also requires fluency in Linux, containers, DevOps practices, and service mesh patterns.
This blend of skills is creating a new kind of operator. The Kubernetworker understands both the physical and logical worlds. They manage network policy on Monday and debug BGP or load balancing behavior on Tuesday. They use Cilium as a common surface for connectivity, security, and observability, which helps bridge the gap between networking and platform teams. In 2026 this hybrid role will become a fixture inside large organizations.
Kubernetes Networking Adapts to AI Workloads
AI is rewriting the expectations placed on Kubernetes. GPU-heavy workloads create intense east-west traffic and require topology awareness, predictable throughput, and smarter scheduling. Traditional CNIs were never designed for this type of pressure.
Kubernetes networking is adjusting. New tools and projects are helping the scheduler place workloads near accelerators, avoid congestion, and react to real-time signals from the datapath. eBPF makes this possible without adding layers of software or slowing anything down. In 2026, AI will continue to shape both the control plane and the dataplane, and Kubernetes networking will become far more aware of the hardware beneath it.
Microsegmentation Becomes Urgent Again
Lateral movement is still the core problem in most breaches, and teams are revisiting how they segment Kubernetes environments. Segmentation today stretches across multiple layers, from broad environment boundaries to fine-grained workload policies to runtime-level controls inside the datapath.
eBPF and Cilium make this easier by enabling identity-aware and protocol-aware enforcement at high speed and without agents. In 2026 segmentation will rise even higher on the priority list as teams secure both microservices and AI-driven applications. Organizations will demand simpler designs that offer real protection without slowing down operations.
Where Things Are Headed
Networking is becoming one of the most important parts of Kubernetes. Organizations moving off VMware or pushing large-scale AI workloads all need the same thing: simple, reliable, and programmable networking. eBPF and Cilium will continue to shape that foundation.
2026 will be defined by the convergence of traditional and cloud native networking, the rise of the Kubernetworker, and the rapid evolution of Kubernetes to support AI. The fundamentals are improving quickly, and the demands on the network are growing even faster. The ecosystem is building the tools needed to keep pace.
##
