AKS will start automatically converting any remaining VM Availability Set (VMAS) node pools to VM Scale Sets (VMSS) on Sept 30, 2026 — that's the operational soil-shift. If your infra still relies on VMAS semantics (fixed NIC attachments, per-VM extensions, predictable instance names, or scripts that look for /dev/sdc), this update will surface those assumptions in painful ways.
AKS v20260904 bundles a few other things: it includes support for Kubernetes 1.36 and continues to roll patches for currently supported Kubernetes streams; previews for newer minor releases and broader GA for earlier supported versions are also included. The migration deadline is the actionable emergency — Microsoft is moving everyone to a VMSS-first node pool model and will auto-convert clusters that still expose VMAS node pools.
Why this matters
VMSS and VMAS are functionally similar to Kubernetes teams on the surface, but their operational contracts diverge in ways that matter for cluster behavior:
- Instance identity and naming conventions change; anything that parses instance names or expects deterministic hostnames will break.
- Node lifecycle semantics differ for in-place replacements vs scale-set orchestration; boot-time extension ordering and custom cloud-init hooks may behave differently.
- Networking: static public IP patterns and per-VM NIC scripts can fail, and Azure-assigned private IP persistence is different under scale-set reimage/upgrade.
- Autoscaler and tooling: older autoscaler integrations or bespoke scaling scripts that assume VMAS will need VMSS-aware configuration.
This is the right call from Microsoft — VMSS is the scalable, cloud-native primitive that makes managed autoscaling, rolling upgrades, and platform-driven features reliable at scale. But it's overdue notice: teams that treat node pools as immutable hardware will be surprised.
What to check right now
Inventory and test. Specifically:
- Identify clusters with VMAS node pools and any node-global scripts that reference instance names, device paths, or expect extension-managed state.
- Validate your use of the Azure Disk CSI driver and StorageClasses: ensure drivers and storage class parameters are up-to-date and compatible with VMSS node pools.
- Confirm cluster-autoscaler or any horizontal scaling tooling supports VMSS node pools and the AKS provider version you run.
- Look for daemonsets or privileged init containers that mount /dev or probe cloud instance metadata. Those can fail under scale-set reimages.
If you want a dry run, migrate a staging cluster to VMSS and run an identical upgrade/scale cycle; this is the test that will catch subtle breakage faster than reading release notes.
Other notable changes in v20260904
Azure OpenAI in Microsoft Foundry introduced more explicit data-zone routing for deployments, which can improve availability and regional routing while keeping deployments scoped to a single resource; expect some routing decisions to be handled by the platform and to plan fallbacks for variable latency and model locality.
Azure DevOps picked up integrations for GitHub Copilot Code Review (public preview) and preview integrations with GitHub Advanced Security, which will shift how teams run AI-assisted reviews and how they budget CI/CD for large orgs.
The Azure SDK for Rust continued maturing, with core components (identity, core libraries) and several storage/secret clients progressing toward 1.0 stability—a sign Azure wants a first-class, strongly typed Rust surface for cloud APIs.
The architecture center also landed new patterns for node-pool management, Dapr microservices, large-scale IoT, OPC UA integrations, and Foundry chat architectures — useful if youre reworking designs to fit VMSS node pools or the new Foundry routing model.
Parting take
Microsoft forcing VMAS→VMSS is both the right operational move and a blunt instrument. It simplifies platform capabilities for Azure at scale, but it will break assumptions that have quietly accreted in infra repositories and init scripts for years. If you haven't inventoried node-topology assumptions and exercised a VMSS migration in staging, youre running out of runway.
Read the migration notes, run the migration in a sandbox cluster, and treat this as an opportunity: VMSS unlocks more reliable autoscaling and rolling upgrades — but only if you stop treating your nodes like pets.
Sources
- Azure/AKS GitHub releases – v20260904
- Microsoft Developer Changelog (Azure platform updates, including AKS and Azure Storage Mover)
- Kubernetes v1.37 enhances resilience, autoscaling; AKS preview rollout
- What's new in Azure OpenAI in Microsoft Foundry Models
- Azure DevOps Released Features (Features timeline – September 4, 2026)
- Copilot Code Review Reaches Azure Repos, Billed Per Review
- Azure SDK Blog RSS feed – From beta to stable: Announcing the Azure SDK for Rust
- What's New in Azure Architecture Center – September 2026 changelog