Looking Ahead to KubeCon EU 2026: AKS Features to Watch in Amsterdam
I have been keeping a close eye on the AKS roadmap over the past year, and with KubeCon EU 2026 coming up at the end of March in Amsterdam, this feels like a good moment to take stock. The AKS team confirmed they will be running another Pre Day in Amsterdam which I am unable to attend, but if previous KubeCon events are anything to go by, it will be packed with announcements and I will be following my friend Pixel Robots/Richard Hooper for shots of the slides. I expect to see features that have been sitting in preview graduate to GA, alongside new previews landing to coincide with the conference.
To get a sense of where things stand, I went back to the December 2025 AKS Community Call, which was a solid year-in-review from Jorge (AKS Principal PM Lead) covering what shipped across 2025. If you have not watched it, I would recommend it.
The AKS public roadmap on GitHub is also worth bookmarking. It gives a reasonable view of what is in preview and what is being worked on.
What Actually Shipped in 2025
There was a lot. Jorge was explicit in the December call that the deck was not even the full picture for the year, just the most recent and most impactful items. A few stood out to me.
AKS Automatic
AKS Automatic reaching GA was probably the headline story of the year for the service. The distinction between Standard and Automatic is now the primary way to think about the two operational models. Standard is the “bring your own opinions” model; Automatic is “Microsoft own the opinionated model.”
Node Autoprovisioning (NAP)
NAP reached GA over summer 2025. The headline capability is that it removes the need to pre-create node pools. The scheduler and scaler work together to provision the most efficient VM SKU for your pending pods on demand, and it runs a consolidation cycle when workloads are scaled down or removed to pack the remaining things more tightly. Due to my current projects I have yet to run it in anger on a production cluster but I like the theory and what I have seen in my testing.
Virtual Machine Node Pools
These reached GA as well, with autoscaling support added later. Rather than a uniform node pool locked to a single VM SKU, you can now define scale profiles inside a single node pool with different SKUs. Useful if you are trying to manage quota efficiently or if you want heterogeneous capacity without proliferating node pools.
AKS Local DNS
This one went GA and the team were notably enthusiastic about it. The short version: it adds a cache at each node that can serve stale DNS responses if upstream DNS has issues, and it supports falling back from UDP to TCP. The team mentioned they had seen it prevent real outages during upstream DNS degradation events.
GitHub Copilot App Modernization (Containerization Assist)
The containerization assist tooling, which generates Dockerfiles and Kubernetes manifests for existing applications and iterates on them when deployment fails policy checks, reached GA as part of the broader GitHub Copilot App Modernization suite. The interesting bit is the feedback loop: if the generated manifest fails an AKS policy (missing resource requests, for example), it registers the error and corrects the manifest before retrying. Useful for teams getting started with Kubernetes who are not yet familiar with all the guard rails in an Automatic cluster.
Features in Preview Worth Watching
The items below are either explicitly in public preview or were described as “coming in the next few months” in the December call. KubeCon EU is a reasonable target to get meaningful updates on these with the prospect for some to graduate to GA.
Managed Nvidia GPU Device Plugin
In the December call, Jorge announced that AKS now manages the lifecycle of the Nvidia device plugin alongside the GPU container drivers. Previously the device plugin was a DaemonSet you ran yourself. This is currently in public preview and was described as going GA “in the upcoming months.”
Dynamic Resource Allocation (DRA)
DRA is available on AKS 1.34+ and is one of the requirements for the CNCF AI Conformance certification that AKS achieved during 2025. Still relatively new upstream (stabilised in Kubernetes 1.32), this may stay in preview for a while longer yet.
Scheduler Profiles
AKS now allows you to configure profiles on the default scheduler rather than running a separate scheduler and referencing it in every workload. This includes:
- Node bin-packing to maximise resource utilisation and reduce costs
- Pod topology spread for high availability across failure domains
- Support for 18 in-tree scheduling plugins (NodeResourcesFit, PodTopologySpread, VolumeBinding, and more)
Node Customization (Image Caching)
This was mentioned as shipping “very soon in Q1” in the December call. The feature pre-caches images onto the node when provisioned, which addresses some of the cases where people have been using the deallocated scale-down mode to preserve image caches.
AKS Desktop and Agentic CLI
AKS Desktop is a local application (based on the open source Headlamp project) with Azure-specific authentication and cluster management built in. The agentic CLI and MCP server integrations are fully open source and allow you to chat with your cluster, run kubectl commands with permission checks, and inspect upgrade readiness. I have yet to play with the agentic CLI in anger but they are worth evaluating if you manage clusters regularly.
Conclusion
If you are planning to attend KubeCon EU 2026 and use AKS, I would recommend making the Pre Day. I am hoping to catch a recording, but failing that I will be relying on screenshots of slides and wading through the Azure Updates feed afterwards.