Power management autoscaling in Azure Virtual Desktop: what it does and when to use it
Partner · Neutral partner: Nerdio
By Mirko Peters ·
Power management autoscaling starts and deallocates existing session hosts on a schedule, based on how much of a host pool's session capacity is in use. It fits host pools with a predictable daily pattern, and it does not create or delete VMs.
Power management autoscaling is the Azure Virtual Desktop feature that powers existing session hosts on and off, following a schedule and the share of session capacity that is in use. Use it when a host pool has a recognisable daily rhythm and you want unused VMs deallocated outside working hours. It does not create or delete VMs. That is the job of dynamic autoscaling, a separate scaling method.
Key points
- Power management autoscaling powers existing hosts on and off. Dynamic autoscaling also creates and deletes hosts, only for pooled pools with a session host configuration (GA July 2026).
- A scaling plan holds the schedules. Pooled schedules have four phases with a capacity threshold and minimum percentage of hosts; personal schedules act on disconnected and signed-out sessions.
- Autoscale turns a host on when used capacity exceeds the threshold, and off when below it and removing a host keeps it below. At exactly the threshold, nothing happens.
- The role Desktop Virtualization Power On Off Contributor must be assigned at subscription level. A lower scope can stop autoscale from working properly.
- Learn says not to combine autoscale with other scaling tools and to disable them for pools with a scaling plan. Azure Automation and Logic Apps scaling cannot run on the same pool.
What power management autoscaling does
Microsoft Learn defines autoscale as the native scaling service that scales session hosts up and down based on the capacity of the host pools and the scaling plan schedule you define. There are two scaling methods:
- Power management autoscaling powers session hosts on and off to adjust to the available capacity in a host pool. Learn names it as the option for host pools with standard management.
- Dynamic autoscaling also creates and deletes session hosts. It can only be used for pooled host pools with a session host configuration. Microsoft announced general availability of enhanced host pool management, including dynamic autoscaling, session host configuration and update, and ephemeral OS disks, in July 2026 (Tech Community). Some Learn pages, for example the scenarios and glossary pages and the module version 5.4.2-preview, still carry preview wording; check the current page before relying on a limit. Learn describes power management autoscaling as generally available.
When autoscale shuts down a VM, it deallocates it.
Power management autoscaling is available in Azure and Azure Government, in regions where you can create host pools. Dynamic autoscaling is Azure only and is not supported in Azure Government.
The building blocks
A scaling plan is an Azure Resource Manager object that defines the schedules for scaling session hosts. A few rules from Learn shape the design:
- One plan can be assigned to several host pools, but only to pools of the type set in the plan, pooled or personal.
- Plan configuration data must be stored in the same Azure region as the host pool configuration. The session host VMs themselves can be in other regions.
- Pooled host pools need a maximum session limit that you set yourself. Learn says not to use the default value.
You can also set an exclusion tag on the plan, for example excludeFromScaling. A VM carrying that tag is not turned on or off and its drain mode is not changed. Excluded hosts still count toward the minimum percentage of hosts.
How a pooled schedule works
A pooled schedule splits the day into four phases. Learn describes them like this:
- Ramp-up: users start signing in and sessions rise quickly.
- Peak: sessions reach the daily maximum and hold steady.
- Ramp-down: users sign out and sessions fall.
- Off-peak: few sessions remain, and a small amount of capacity may stay on for late users.
Each phase has a start time. Ramp-up and ramp-down also expose the capacity threshold and the minimum percentage of hosts. Peak carries the capacity threshold over from ramp-up, and off-peak carries it over from ramp-down. The create-and-assign page recommends the breadth-first load balancing algorithm for ramp-up and depth-first for off-peak, and lets you pick either for peak.
Three terms drive every decision:
- Available host pool capacity is the maximum session limit multiplied by the number of available session hosts.
- Used host pool capacity is the number of active and disconnected sessions divided by the available capacity.
- Capacity threshold is the percentage of used capacity that triggers a scaling action.
The minimum percentage of hosts is the lowest share of all session hosts that must be on in a phase. It is rounded up: 10 percent of 7 hosts keeps 1 host on.
The decision rule, with numbers
Autoscale turns a host on when used capacity exceeds the threshold. It turns a host off when used capacity is below the threshold, turning the host off would not exceed the threshold, and the host has no sessions (force logoff in ramp-down is the exception). Per Learn's FAQ, when capacity equals the threshold, autoscale does nothing.
Learn's ramp-up scenario (scenario 1) starts with two available hosts out of six, a maximum session limit of 5 and a threshold of 30 percent. Capacity is 10 sessions. Three sessions give 30 percent, which is not above the threshold, so nothing happens. A fourth session gives 40 percent and autoscale turns on one host (capacity 15, about 27 percent).
Autoscale does not turn hosts off during ramp-up. Learn's peak scenario (scenario 2) has 5 available hosts, a limit of 5 and 7 sessions, which is 28 percent used. Autoscale turns off no host yet, because it cannot turn off a host without exceeding the threshold. By the capacity formula, removing one host would leave 7 sessions on a capacity of 20, which is 35 percent, above the 30 percent threshold. After two users sign out for lunch, 5 sessions remain. Removing one host would give 25 percent, so autoscale selects a host without user sessions, puts it in drain mode and turns it off.
Ramp-down and off-peak
Ramp-down uses session consolidation. Autoscale counts active and disconnected sessions, assumes disconnected users may come back, and keeps enough hosts for all of them. The formula on Learn's glossary page is required hosts = total user sessions ÷ (maximum session limit × capacity threshold), rounded up. The glossary's example: 8 disconnected sessions, a limit of 10 and a threshold of 80 percent need one host.
You choose how the remaining hosts are emptied:
- Without force logoff: the create-and-assign page tells you to select VMs have no active or disconnected sessions or VMs have no active sessions, and autoscale waits for users to sign out. The FAQ says that with "Stop VMs when VMs have no active sessions" autoscale still treats disconnected sessions as active, because users may reconnect immediately. Excess hosts are deallocated only if they have no sessions.
- Force logoff of users: autoscale picks the host with the fewest sessions, puts it in drain mode, notifies users, signs them out after the wait time you set, and deallocates the VM.
Force logoff applies only in ramp-down. Learn's ramp-down scenario (scenario 3) has 4 available hosts, a limit of 5, 4 sessions (20 percent used), a threshold of 75 percent and a minimum of 10 percent. Applying the glossary formula, 4 ÷ (5 × 0.75) = 1.07, so 2 hosts are required. Two hosts can be turned off without exceeding 75 percent, and because every host has the same number of sessions, autoscale chooses two hosts at random, signs out all their users and turns the hosts off. Later in the same scenario, after the two users who were signed out sign back in (4 sessions, 40 percent) and one user signs out (3 sessions, 30 percent), one more host is turned off, chosen as the host with the fewest sessions. With 3 sessions on the last host (60 percent), autoscale keeps it on, because turning it off would drop capacity to 0, below the 10 percent minimum.
In off-peak, the minimum percentage of hosts keeps capacity on. The create-and-assign page lists that field only for ramp-up and ramp-down, so check which value your plan carries into off-peak. Per the FAQ, autoscale can turn off all VMs only if the minimum percentage of hosts is 0 percent and there are no user sessions on the session hosts, disconnected ones included. A time limit policy, set in Intune or Group Policy, can sign out disconnected sessions.
Personal host pools
A personal schedule has no capacity threshold. Each phase acts on sessions that were disconnected or signed out for 0 to 360 minutes: deallocate, hibernate or do nothing. In ramp-up you choose which VMs to start: assigned only, assigned and unassigned, or none. If autoscale does not start personal desktops, the FAQ says to enable Start VM on Connect or start them yourself.
Setting it up
The step-by-step procedure is in the article "How to create and assign an AVD scaling plan step by step". Three points decide whether power management autoscaling works:
- Role scope. Assign Desktop Virtualization Power On Off Contributor at subscription level. The create-and-assign page says a lower scope prevents autoscale from working properly, while the troubleshooting page also lists host pool and resource group scope; follow the create-and-assign page. For PowerShell, that page requires
Az.DesktopVirtualization4.2.0 or later. - Weekend. Off-peak settings of the last selected day carry over to unselected days, so a Monday-to-Friday plan holds Friday's off-peak settings through the weekend.
- Diagnostics. Autoscale logs can go to a storage account or an event hub. For pooled pools Microsoft recommends the autoscale data in Insights.
When to use it, and when not
Power management autoscaling fits when:
- Working hours are regular and you can name a ramp-up, peak and ramp-down window.
- The host pool uses standard host management, with VMs that already exist.
- A realistic maximum session limit can be set, because used capacity and the threshold derive from it.
- Pooled sessions can end by sign-out, by force logoff in ramp-down, or by a time limit policy for disconnected sessions, so consolidation can free hosts.
- The pool is personal and users disconnect or sign out at predictable times, so the VM can be deallocated or hibernated after a set number of minutes. Hibernate must be enabled on the VM and is not supported with FSLogix or App Attach.
- Compute cost is the driver. The FAQ says using autoscale costs nothing extra, so the saving comes from VMs that are deallocated outside working hours.
Consider another approach when:
- You need hosts created and deleted on demand. That is dynamic autoscaling, and it requires a session host configuration.
- Your demand depends on specific dates. The FAQ says autoscale does not currently support ramping down on specific dates.
- You rely on the Azure CLI. The FAQ says autoscale cannot be configured with it.
- You need disaster recovery. The FAQ says customers must set up their own disaster recovery strategy to manage outages: autoscale only handles scaling existing VMs within the region they were created in.
- Another tool already scales the pool. Learn says autoscale cannot be used with Azure Automation and Azure Logic Apps scaling on the same host pool. More generally, the scenarios page says: "Don't use autoscale in combination with other scaling Microsoft or third-party scaling tools. Ensure that you disable those for the host pools you apply the scaling plans to."
Common traps
- Exclusion tags used too widely. Learn describes a pool of six hosts where five were tagged. At 80 percent used capacity autoscale could not add capacity. After the admin tagged and turned off the last untagged host, the affected user saw "No resources available." Removing the tag from two VMs let autoscale start one host, enough for the 10 percent minimum.
- Drain mode is overwritten. Autoscale changes drain mode on pooled hosts. Use the exclusion tag when you update VMs.
- Load balancing is overridden. In pooled pools autoscale applies the algorithm from the schedule, not the one set on the host pool.
Sources
Last checked 17 November 2026.
- Microsoft Learn: Autoscale scaling plans and example scenarios in Azure Virtual Desktop
- Microsoft Learn: Create and assign an autoscale scaling plan
- Microsoft Learn: Autoscale glossary for Azure Virtual Desktop
- Microsoft Learn: Autoscale FAQ
- Microsoft Learn: Set up diagnostics for Autoscale
- Microsoft Learn: Troubleshoot autoscale issues in Azure Virtual Desktop
- Microsoft Tech Community: Enhanced host pool management for Azure Virtual Desktop is now generally available
FAQ
Does power management autoscaling cost extra?
Microsoft's autoscale FAQ says no: you are not charged extra for using autoscale, and it points to the Azure Virtual Desktop pricing page for rates. For personal host pools, the FAQ adds that autoscale deallocates a session host that was turned on but never used to sign in, after a period of inactivity, to avoid unnecessary compute costs.
Can autoscale turn off every VM in a host pool?
Only if the minimum percentage of hosts is set to 0 percent and there are no user sessions on the session hosts. Microsoft's FAQ says a disconnected session counts as a session and recommends a time limit policy that signs out disconnected users.
Can I configure autoscale with the Azure CLI or skip specific dates such as holidays?
The autoscale FAQ on Microsoft Learn states that autoscale does not offer configuration through Azure CLI and does not support ramping down on specific dates, such as holidays.