KubeCon Europe 2026: What 12,000 Developers Are Watching in Amsterdam

KubeCon + CloudNativeCon Europe 2026 opens tomorrow in Amsterdam — 12,000 developers, architects, and platform engineers descending on RAI Amsterdam for four days of cloud native infrastructure, AI workload scheduling, and the debut of a brand-new track that did not exist twelve months ago: Agentics Day.

This is the largest cloud native event in Europe this year. If you build, deploy, or operate software on Kubernetes — which at this point means the majority of production backend systems — here is what is actually worth paying attention to.

The AI on Kubernetes Track Is Now the Biggest Track

Forty-one percent of AI developers are now cloud-native. That number comes from CNCF's own survey data, and it explains why the AI/ML track at KubeCon EU 2026 has more sessions than any other track.

The conversations have moved past "can you run AI workloads on Kubernetes" and into specifics: GPU scheduling, LLM inference pipelines, multi-tenant GPU clusters, and what happens when your inference latency requirements collide with Kubernetes scheduler defaults.

The session drawing most attention from the pre-conference buzz is Mukund Muralikrishnan from Wayve — an autonomous vehicle company running real-time AI inference in production — talking about shared GPU scheduling rules. Wayve operates AI inference at latency budgets that leave no room for scheduling jitter, and their session covers what it took to make Kubernetes behave correctly under those constraints.

Amazon EKS is keynoting on the same theme. Jesse Butler's session is titled "From Complexity to Clarity: Engineering an Invisible Kubernetes" — which is EKS-speak for their pitch that managed Kubernetes should abstract away the parts that have nothing to do with your application. Expect demos of Karpenter (their open-source autoscaler), kro (Kubernetes Resource Orchestrator), and their Cedar policy engine. AWS booth is at #700 if you're on the floor.

Google's Jago Macleod closes the final keynote day with a session literally called "From Cloud Native to Accelerator Native" — the framing that GPU clusters are the new compute primitive and Kubernetes is adapting to them, not the other way around.

Agentics Day: The Newest Co-Located Event and Why It Matters

March 23 is co-located event day, and the one that did not exist at KubeCon EU 2025 is Agentics Day. This is the CNCF community's formal acknowledgement that AI agents running inside Kubernetes clusters — not just AI workloads running on Kubernetes — are production reality in 2026.

The Model Context Protocol (MCP) is the glue. MCP standardises how AI agents connect to tools, data sources, and APIs. Several sessions are covering MCP in production Kubernetes environments: agents that query cluster state, agents that respond to alerts, agents that perform remediation without human approval.

Microsoft's sponsored keynote is on exactly this: "Scaling Platform Ops with AI Agents: Troubleshooting to Remediation." Their demo shows Azure AKS agents identifying degraded pods, tracing root cause through OpenTelemetry spans, and triggering remediation — the kind of on-call workflow that currently wakes someone up at 3am.

Lin Sun from Solo.io is opening with a keynote titled "The Future of Cloud Native Is… Agentic" — a direct statement about where the CNCF community sees the next two years going. This is not a research preview. Agentic workloads on Kubernetes are shipping in production at companies in the room.

If you are building AI agents or evaluating MCP tooling, Agentics Day is the one co-located event worth watching even if you are not in Amsterdam. CNCF typically posts session recordings within 48 hours.

Kyverno Graduates to CNCF: What It Means for Policy-as-Code

The cleanest project news coming into the conference is Kyverno graduating to CNCF Graduated status on March 16 — one week before the conference. Timing was not accidental. KyvernoCon, the first-ever Kyverno-dedicated co-located event, runs on March 23.

Kyverno is a Kubernetes-native policy engine. It validates, mutates, and generates Kubernetes resources based on policies written as Kubernetes resources themselves — no separate policy language required, no OPA/Rego to learn. CNCF Graduated status means the project has passed the technical due diligence process, has multiple independent maintainers, and has demonstrated production adoption at scale.

For platform engineering teams, graduation means: the risk of backing a project that stalls or gets abandoned has dropped significantly. Kyverno is now in the same graduation tier as Kubernetes itself, Prometheus, Envoy, and Argo.

The practical impact: expect Kyverno policy-as-code to appear in more enterprise compliance tooling, more cloud provider managed offerings, and more security scanning pipelines over the next twelve months.

CiliumCon and the eBPF Layer That Powers Modern Kubernetes Networking

Cilium is celebrating ten years since its first commit at CiliumCon on March 23. For anyone who has not tracked eBPF-based networking closely: Cilium is the de facto standard for Kubernetes networking at scale. It replaced iptables-based kube-proxy at Google, Meta, Microsoft, and most large Kubernetes deployments because eBPF programs run in the kernel at line rate rather than copying packets to userspace.

The SIG Network keynote on AI networking features engineers from Red Hat, Google, NVIDIA, and IBM — which tells you something about where the complexity is concentrating. Running LLM inference workloads at scale requires different network guarantees than running web services: lower latency variance, higher bandwidth between GPU nodes, and traffic shaping that understands tensor parallelism patterns.

CiliumCon sessions will cover Cilium 1.17 features, multi-cluster networking, and the intersection of eBPF observability with AI workload debugging. If you operate large Kubernetes clusters or are moving AI inference to Kubernetes, this is technically dense but directly applicable.

Platform Engineering: The 56% Talent Gap Problem

The second major non-AI thread running through the conference is platform engineering and developer experience. CNCF survey data puts it plainly: 56% of organisations report a shortage of engineers with platform engineering skills.

Abby Bangser from Syntasso is keynoting on "From Cloud-Native Apps to Cloud-Native Platforms" — the argument that Kubernetes expertise has to be packaged into self-service internal developer platforms so that application teams can deploy and operate without needing to understand the full stack underneath. This is the Internal Developer Platform (IDP) pattern, and it is now standard practice at companies above a certain engineering headcount.

The co-located Platform Engineering Day on March 23 covers concrete tooling: Backstage for developer portals, Crossplane for infrastructure-as-code, and Radius (Microsoft's open-source application platform) alongside other CNCF projects in the IDPification stack.

DigitalOcean is presenting a case study that deserves attention: they reduced their observability team from 12 engineers to 4 by standardising on OpenTelemetry and building better tooling on top of it. That ratio — 12 to 4 — is the kind of number that ends up in CFO presentations.

What OpenTofu, WASM, and Argo Are Doing on March 23

OpenTofu Day covers the open-source Terraform fork that has absorbed most of the community that left after HashiCorp's licence change in 2023. OpenTofu 1.9 shipped earlier this year with provider-defined functions and improved state management. If you are still on Terraform, the sessions will be relevant — most OpenTofu content is directly portable.

WasmCon is running the same day. WebAssembly on Kubernetes has moved from experimental to production at a small number of companies. The primary use case is running untrusted code in sandboxed environments without the overhead of a full container — useful for plugin systems, edge compute, and multi-tenant SaaS platforms.

ArgoCon covers Argo CD, Argo Workflows, and Argo Rollouts. Argo CD is the most widely deployed GitOps tool in Kubernetes environments. AWS is demoing GitOps patterns at their booth, and several sessions cover progressive delivery patterns that combine Argo Rollouts with traffic shaping.

How to Follow Along If You Are Not in Amsterdam

CNCF posts session recordings publicly, usually within 24–48 hours of each day ending. The schedule is at kccnceu2026.sched.com and sessions are tagged by track.

For real-time coverage: the CNCF Twitter/X account posts keynote highlights live. The Kubernetes Slack (#kubecon channel) is active during the conference with attendee commentary.

If you are evaluating whether Kubernetes is the right platform for AI inference workloads, the AWS and Google keynotes on March 24 and 26 respectively are the most directly actionable for a decision you might need to make in the next quarter. Both will be on YouTube within 48 hours.

The LLM API Pricing Tracker is useful context — the cost gap between running inference yourself on Kubernetes versus paying managed API pricing is what is driving the GPU-on-Kubernetes adoption being discussed at the conference.

Key Takeaways

• KubeCon EU 2026 runs March 23–26 in Amsterdam (RAI Amsterdam) — 12,000+ attendees, the largest cloud native event in Europe this year
• AI on Kubernetes is the largest track — GPU scheduling, LLM inference pipelines, and multi-tenant GPU clusters are the primary topics, not AI as a concept
• Agentics Day is new for 2026 — MCP-connected AI agents managing Kubernetes clusters are production reality; Microsoft and Solo.io keynoting on it
• Kyverno graduated to CNCF Graduated status on March 16 — same tier as Kubernetes, Prometheus, Envoy; safe to standardise on for enterprise policy-as-code
• Cilium hits 10 years — eBPF-based networking is now the production standard at Google, Meta, Microsoft; CiliumCon covers Cilium 1.17 and AI networking
• Platform engineering talent gap is 56% — self-service IDPs (Backstage, Crossplane, Radius) are the industry response; Platform Engineering Day on March 23
• No Kubernetes release at this event — Kubernetes 1.36 is scheduled for April 22; conference covers roadmap sessions and SIG previews
• Session recordings go public within 24–48 hours — follow CNCF YouTube and the #kubecon Kubernetes Slack channel