Source: Tech Times
KubeCon + CloudNativeCon North America 2026 closed on 12 November in Salt Lake City after four days that delivered a clear verdict: Kubernetes is no longer just a container orchestrator, it is becoming the operating system underneath AI workloads. For the first time the conference carved out an entire programming track dedicated to AI inference and autonomous agents, and the shift in conversation was visible everywhere. The community stopped asking whether Kubernetes can support AI and moved on to how to run those workloads reliably, securely and at a cost that survives production pressure.
The numbers behind the shift
The framing is backed by data. According to the 2026 CNCF Annual Cloud Native Survey, 66 percent of organisations that actively host generative AI models now rely on Kubernetes to manage some or all of those inference workloads. Jonathan Bryce, Executive Director of Cloud and Infrastructure at the Linux Foundation and head of the Cloud Native Computing Foundation, described AI as quickly becoming one of the largest compute workloads the industry has ever seen, with the real engineering challenge moving from training models to running them in production. The new track covered GPU scheduling, model serving with tools such as vLLM and KServe, agentic system design built on protocols including MCP and A2A, AI gateways and energy-efficient inference optimisation.
Tooling that fills real gaps
Several launches stood out for platform teams. Cloudsmith introduced an AI Agent Skills registry that lets platform teams host, version and govern the skill bundles AI coding agents use, blocking unapproved skills and versions, because community skill registries were built for sharing rather than control. groundcover expanded its eBPF sensor to tie application-layer request spans directly to the SQL queries they trigger, without SDK changes, and shipped a Crossplane provider so observability configuration can be managed as Kubernetes custom resources inside standard GitOps workflows. DataAgent, which emerged from stealth in September, released its open-source lab under the Apache 2.0 license, giving SRE and DevOps teams a local environment to deploy an autonomous agent, simulate infrastructure failures and evaluate detection and remediation, with automated fixes routed through GitOps pull requests and human review. The conference itself drew 154 speakers and 335 sponsors, with platform engineering and security tracks running at full strength alongside the AI programme.
Why this matters in Europe
The next edition, KubeCon + CloudNativeCon Europe 2027, takes place in Barcelona on 15 to 18 March 2027, and it will measure how far this production story has actually come. For European teams the message is practical: if you run Kubernetes today, your platform is already the natural landing zone for AI inference, and the governance questions around it, GPU scheduling, cost control, supply chain security for agent skills and prompt-driven access to infrastructure, belong in your architecture reviews now rather than later. Running AI workloads on infrastructure you control, in European regions, also fits neatly with the ongoing debate about digital sovereignty and reducing dependency on non-European compute.
If you are planning to run AI inference workloads on Kubernetes and want help with cluster design, GitOps workflows or securing agentic systems, contact Excello Digital for advice from engineers who work with these platforms every day.