Picture this: you’re minding your own business on a Monday morning, coffee in hand, and you realize you haven’t checked the Azure Local release notes since spring. You open the What’s New page and — oh boy — there’s a doooooozy of accumulated changes staring back at you. Six months of monthly drops, each with their own OS version, .NET bump, and a growing list of features that have quietly gone GA or landed in preview.
—
First, the Baseline: What Version Are You On?
The August 2026 release of hyperconverged deployments of Azure Local is version 12.2608.1003.8. That’s the current tip of the spear. Each Azure Local release is supported for six months, whether you’re on 23H2 (11.x.x.x) or 24H2 (12.x.x.x). If you’re still on anything in the 11.x range, you’re already out of support — October 2025 (version 11.2510) was the final 23H2 release. Get off it.
Quick check from any cluster node:
# Check your current Azure Local solution version
Get-SolutionUpdate | Select-Object Version, State, InstalledDate | Sort-Object InstalledDate -Descending | Select-Object -First 5—
Step 1: OS and Driver Hygiene First
Every monthly release ships a new OS build, and you need compatible drivers to go with it. This is the part people skip and then wonder why things break.
In the 2608 release, all new and existing deployments of Azure Local run OS version 26100.33296, and you also need a driver that’s compatible with OS version 26100.33296 or Windows Server 2025. For Integrated System or Premier solution hardware from the Azure Local Catalog, the OS is preinstalled — but for everything else, work with your OEM to get a compatible OS image and driver.
Don’t assume your existing drivers carry over cleanly. Check with your OEM before applying the feature update. This has bitten more than a few people.
—
Step 2: Update Your stack-hci-vm Extension (Seriously, Do It Now)
A handful of the VM management improvements that shipped in 2604 require an updated CLI extension. Before you do anything else with Azure Local VMs from the CLI:
# Update the stack-hci-vm extension — required for graceful shutdown and SDN bypass features
az extension update --name "stack-hci-vm"
# Verify the version
az extension show --name "stack-hci-vm" --query version -o tsvWhy does this matter? GPU acceleration for Azure Local VMs is now generally available — Azure Local supports attaching or detaching full GPUs (DDA) or GPU partitions (GPU-P) to Azure Local VMs during creation or as a Day-2 operation using the Azure CLI or Azure portal. That GA status means nothing if your CLI extension is six months old.
Also from 2604: graceful VM stop and restart now happens by default. If you need to hard-stop a VM (say, it’s hung), use the --skip-shutdown flag:
# Graceful restart (default behavior post-2604)
az stack-hci-vm restart --name myVM --resource-group myRG
# Hard stop — skips guest OS shutdown
az stack-hci-vm stop --name myVM --resource-group myRG --skip-shutdown—
Step 3: GPU Story — From Metrics to Full Cloud Management
The GPU narrative has evolved significantly across these releases and it’s worth understanding the progression.
2605 brought GPU metrics for GPU-P configurations — you can now monitor GPU metrics in Azure Local for GPUs configured using GPU Partitioning (GPU-P).
2603 added hardware support — starting with 2603, the NVIDIA RTX PRO 6000 Blackwell Server Edition GPU is supported on Azure Local VMs and AKS enabled by Azure Arc, allowing you to run GPU-accelerated workloads using these GPUs.
2608 takes it further with centralized management. GPU Management from Azure (preview) lets you manage and monitor GPU resources in Azure Local from Azure, including centralized GPU inventory, partition configuration, GPU assignment to workloads, and lifecycle management of GPU-accelerated workloads.
There are two attachment modes to understand: Discrete Device Assignment (DDA) allows you to dedicate a physical GPU to your workload, while GPU-P allows you to share a GPU with multiple workloads by splitting it into dedicated fractional partitions. DDA gives you full GPU access and best app compatibility; GPU-P is better for density. You can’t assign a physical GPU as both DDA and GPU-P — it’s one or the other.
One gotcha worth flagging: GPU partitioning is unsupported if your configuration isn’t homogeneous — mixing GPUs from different vendors in the same system, or using different GPU models from different product families from the same vendor, are both unsupported. Plan your hardware accordingly before you go all-in on GPU-P.
—
Step 4: VM Lifecycle Improvements Worth Knowing
Graceful shutdown from the portal (2608): The Azure portal now gives you two options when stopping an Azure Local VM — “Shut down (recommended)” which powers down the guest OS first, and “Turn off” which is the equivalent of pulling the power cord. This requires 2604 or later, so make sure you’re current.
VMConnect for Azure Local VMs (2601): If you have VMs with no network connectivity or boot failures, VMConnect is your friend. It’s been in preview since January. Think of it as the Azure Local equivalent of console access — you can connect to Windows and Linux VMs even when the network is gone.
Add Node via Azure Portal (2608, preview): You can now add nodes to an existing Azure Local cluster directly from the Azure portal with a new guided scale-out experience. Previously this was a PowerShell-only operation, which was fine but added friction. One caveat: each new physical node that you add to your system must closely match the rest of the nodes in terms of CPU type, memory, number of drives, and the type and size of the drives. The portal won’t save you from hardware mismatches.
—
Step 5: Storage Expansion — SAN + iSCSI Goes GA
This is a big one for anyone running larger or more complex environments. Starting in 2604, the disaggregated deployment model arrived, which means you can now run compute and storage independently. Azure Local now supports a disaggregated infrastructure architecture, allowing customers to deploy compute and storage resources independently while continuing to benefit from an Azure-consistent management and operational experience — enabling organizations to scale infrastructure more flexibly by separating compute and storage to align with workload demands.
iSCSI SAN integration went through preview in 2605 and hit GA in 2607. Azure Local now supports iSCSI-connected external SAN storage in disaggregated deployments, enabling customers to connect and manage supported SAN arrays over iSCSI and expanding storage connectivity options beyond Fibre Channel.
One thing to watch: your Azure Local instance supports a maximum of 16 nodes in hyperconverged mode, but with disaggregated deployments and SAN storage, that ceiling goes away. If you’re planning a large-scale deployment, this is the architecture to evaluate.
—
Step 6: Security Baseline Changes — Check Your Passwords
Two security changes from 2607 that may catch you off guard:
The security baseline now enforces a 14-character minimum password length for local account password requirements, helping align with security industry standards and compliance with the Azure Security Baseline. If you have automation or service accounts with shorter passwords, update them before applying 2607 or later.
Also from 2604: a user logon text and banner now help comply with DISA STIG and CIS requirements, and these new settings are protected by drift control — to customize the values, first disable drift control.
Speaking of drift control — it’s been a recurring theme since 2602. The drift detection framework continuously validates component-level state against approved baselines. If you’re doing any manual configuration of nodes, be aware that drift detection will flag it.
—
Step 7: Confidential VMs — Worth a Look If You’re in Regulated Industries
Confidential VM functionality is now in public preview on Azure Local, powered by AMD SEV-SNP technology — this capability extends workload attestation and memory-encryption protections to your VMs.
For context on what SEV-SNP actually does: it’s a Trusted Execution Environment technology from AMD that offers memory encryption, unique CPU keys, encryption for the processor register state, integrity protection, and firmware rollback prevention — collectively hardening guest protections to deny hypervisor and other host management code access to VM memory and state.
This is a preview, so I wouldn’t run production workloads on it yet. But if you’re in healthcare, finance, or government and have been waiting for hardware-backed VM isolation on-prem, this is the feature to start evaluating now.
—
Step 8: Update Workflow Changes (2608)
This release introduces the “prepare updates” workflow, allowing you to prepare updates now and install them at a later time. Feature updates now include a new update state, “Pending OEM Validation,” indicating that the update is awaiting OEM validation before it can proceed.
This is a quality-of-life improvement for shops that need to coordinate update windows. The “Pending OEM Validation” state is new — don’t panic if you see it. It just means your hardware vendor hasn’t signed off yet. Wait for it to clear before proceeding.
You can also now download update packages directly from the Azure portal for limited-connectivity or offline scenarios — handy for air-gapped or edge environments where pulling packages from the internet isn’t an option.
—
What to Watch Next
- Confidential VMs on Azure Local — still preview, but the AMD SEV-SNP story is maturing fast. If you’re in a regulated industry, get a test cluster running this now so you’re not scrambling when it hits GA.
- GPU Management from Azure (preview, 2608) — centralized GPU inventory and lifecycle management from the portal is exactly what multi-site GPU deployments need. Keep an eye on this one.
- KMS v2 migration for AKS — if you’re running AKS enabled by Azure Arc on Azure Local, KMS v1 deprecation has been called out since 2603. Plan your cluster redeployments using KMS v2 before it becomes urgent.
- Disaggregated deployments — if you’re hitting the 16-node ceiling or need independent compute/storage scaling, this is the architecture to start planning around now.
