KubeCon EU 2026 recap

I recently returned from attending KubeCon EU in beautiful Amsterdam, Netherlands and these are some announcements that caught my eye.

To nobody’s surprise, the AI theme continued its upward trend, dominating announcements surrounding Kubernetes. Here are some notable ones that deserve some extra attention:

**llm-d:** An open-source, Kubernetes-native framework for distributed LLM inference was accepted as a CNCF Sandbox project. Why is llm-d a big deal? Its goal is to support any model, any accelerator, any cloud. It was a collaborative effort between some of the biggest and most recognizable names in the tech industry, led by Red Hat, Google Cloud, IBM Research, CoreWeave, and NVIDIA. Some of the most popular, high-adoption projects are collaborations from big tech companies precisely because those companies are often consumers of the technology themselves. Kubernetes, PyTorch, and Cilium are just a few examples, and we can now add llm-d to that list.

**NVIDIA’s Dynamic Resource Allocation (DRA):** NVIDIA donated its DRA Driver for GPUs to the CNCF, moving governance from a single vendor to full community ownership under the Kubernetes project. This standardizes and opens up how GPU resources are managed in Kubernetes, including scheduling, fractional sharing, MIG partitioning, and multi-node NVLink support. While this is arguably self-serving in that it helps NVIDIA strengthen its ecosystem position, it will undoubtedly benefit the many developers already relying on NVIDIA accelerators.

**Gateway API Inference Extension:** An extension to Gateway API, long a standard for Kubernetes networking, that enables model-aware routing for inference workloads. Rather than routing purely on path or headers, it can inspect request bodies to identify the target model and route intelligently, accounting for things like KV-cache state, queue depth, and serving priority.

Honorable mention:

Though not a KubeCon announcement, I attended a Microsoft Azure Kubernetes event while in Amsterdam and a couple of cool projects stood out. KAITO and AI Runway are both worth keeping an eye on. They’re open-source projects from Microsoft, with KAITO already a CNCF Sandbox project since October 2024, and together they remind me a lot of “enterprise Ollama on Kubernetes.” I use Ollama regularly for running local LLMs in development and testing, so I can see the appeal of these tools, especially with the added enterprise features like HuggingFace model discovery, GPU memory fit indicators, and real-time cost estimates that AI Runway brings.

By far the biggest announcement/update was regarding the **Kubernetes AI Conformance Program**. The program was formally launched at KubeCon NA 2025 in Atlanta just a few months prior, though it had been previewed in alpha at KubeCon China and KubeCon Japan earlier that summer, with the aim of standardizing AI workloads on Kubernetes to ensure portability of training, inference, and agentic workloads across platforms. At KubeCon EU 2026, the program expanded with new requirements tied to the latest Kubernetes release, the addition of agentic workload validation, and four more newly certified platforms bringing the total to 31. This helps enable and ensure the viability and adoption of aforementioned projects like llm-d, NVIDIA DRA, and Gateway API Inference Extension.

Some thank you’s

Google

I’d like to thank Google for once again inviting me to give a talk at their booth on KubeRay for GKE. KubeRay is a Kubernetes operator for Ray, which is a framework for running distributed Python applications and ML model training:

Giving my talk on distributed ML model training with KubeRay on GKE

Isovalent

Finally, I’d like to congratulate the Cilium community on the project’s 10th anniversary. The *Cilium: Up and Running* O’Reilly book, authored by Nico Vibert, Filip Nikolic, and James Laverack from Isovalent, was celebrated at the event and signed copies were distributed. I had the wonderful pleasure of being a technical reviewer for the book. Congratulations to Nico, Filip, and James!

Left to right: James Laverack, Nico Vibert, Filip Nikolic, (and me)

BONUS: A few photos of my time in Amsterdam and KubeCon

Save the date for upcoming KubeCons!