Why Disk Swapping Matters: The AVD Cost You’re Still Paying After Shutdown

October 8, 2026

Autoscaling is usually the first place organizations look when optimizing Azure Virtual Desktop costs, and for good reason.

If nobody is using a session host, there is little reason to keep paying for its compute resources. A well-designed autoscaling strategy can shut down unnecessary capacity during evenings, weekends, and periods of lower demand, significantly reducing the number of compute hours an organization consumes.

However, the problem is that shutting down a VM only solves part of the cost equation. When an Azure VM is stopped and deallocated, compute charges stop, but its managed disks continue to exist and incur storage charges.

An organization can build an aggressive autoscaling strategy, deallocate hundreds or thousands of desktops every night, and still pay for high-performance storage attached to machines that spend most of those hours doing absolutely nothing.

That is where automated disk swapping becomes valuable. Rather than treating the storage configuration of a virtual desktop as something that remains static throughout its lifetime, Hydra can adjust the disk tier based on whether that desktop is actually being used.

Premium Performance When You Need It, Budget Storage When You Don’t

There are good reasons to use SSD storage for virtual desktops. Storage performance can affect logon times, application launches, Windows services, and plenty of other operations users experience throughout the workday.

Saving money by intentionally putting an actively used desktop on slow storage is usually a lousy tradeoff, particularly when the goal is to provide an experience comparable to a physical PC, where SSD has become the norm.

A powered-off desktop has very different requirements. At 2:00 AM, a deallocated personal desktop does not care about IOPS or latency because nobody is logging on, launching Outlook, opening applications, or installing updates. Yet, if its OS disk remains on a higher-cost SSD tier, the organization continues paying for that storage configuration throughout those idle hours.

Hydra’s automated disk swapping addresses that mismatch by changing the storage tier along with the VM’s lifecycle. When a configured session host is deallocated, Hydra can automatically move its disk from the configured SSD tier to Standard HDD. Before the machine starts again, Hydra restores the disk to its configured SSD tier.

The result is straightforward: use performant storage while the desktop is running, and use inexpensive storage while it is sitting idle.

How Small Disk Savings Become Big AVD Savings at Scale

The difference between SSD and HDD pricing on a single virtual desktop might not immediately get anyone’s attention. The economics change substantially when that difference is multiplied across hundreds or thousands of machines and thousands of idle hours throughout the year.

The Math on a 1,000-Desktop Estate

Consider running 1,000 persistent AVD desktops where the average user needs their desktop for approximately 45 hours per week.

There are 168 hours in a week, meaning those desktops could potentially spend around 123 hours, roughly 73% of the week, powered off.

Traditional autoscaling attacks the compute cost during those hours, while disk swapping extends the same optimization principle to storage.

The basic calculation is simple: storage cost difference × percentage of time idle × number of desktops. If moving between the organization’s chosen SSD and HDD tiers creates even a $10 monthly difference per desktop, and the desktops are eligible to remain on the lower-cost tier for roughly 70% of the month, 1,000 desktops represent approximately $7,000 per month in potential savings.

At 2,000 desktops, that same small per-machine optimization becomes approximately $14,000 per month, or $168,000 annually.

Those figures are illustrative rather than universal because actual savings depend on many factors like Azure region, disk size, disk SKU, operating schedules, and customer-specific pricing.

The more important point is the multiplier. Repeating that optimization automatically across a large desktop estate every evening, weekend, and holiday is where it starts becoming impactful.

Why Persistent Desktops Benefit Most from Disk Swapping

Pooled AVD environments already have another powerful optimization option: eliminate unnecessary infrastructure entirely. Hydra can dynamically provision temporary session hosts when additional capacity is required and remove them when demand falls.

If a temporary VM and its associated resources no longer exist, there is obviously little reason to worry about optimizing their idle storage costs. Many organizations opt to use ephemeral disks for these use cases, eliminating storage costs altogether while getting local disk NVME-like performance.

Persistent desktops are different because the machine itself needs to remain. A user’s assigned desktop may be powered off for 16 hours every weekday and nearly the entire weekend, but IT cannot necessarily delete it every night and simply create a different desktop the following morning. The persistence of the machine is part of the service being delivered.

That makes disk swapping particularly useful for personal AVD environments. Instead of choosing between persistence and cost optimization, organizations can optimize the resources surrounding the desktop according to its actual usage.

The machine remains assigned to the user, its OS disk remains intact, and its configuration remains persistent, but that does not mean the disk needs to remain on its more expensive performance tier 24 hours a day.

This becomes especially important at scale. A thousand persistent desktops sitting powered off overnight may have virtually no compute cost during that period, but they still represent a thousand managed disks. Disk swapping allows the storage layer to participate in the same cost optimization strategy that organizations already apply to compute.

Pair Disk Swapping with Self-Service Startup Schedules

Aggressive cost optimization becomes considerably less attractive if it makes the user experience worse. A persistent desktop that is cheap overnight but takes forever to become available in the morning is not exactly going to earn IT a parade. The operational challenge is making sure resources are ready again when users actually need them.

Hydra’s self-service capabilities help close that loop.

Users with personal desktops can control the power state of their assigned machines and configure startup schedules through the My Sessions & Hosts portal. A user who normally begins working at 8:00 AM, for example, can schedule the desktop to start ahead of the workday rather than leaving it running overnight or waiting for it to start only after attempting a connection.

Combined with disk swapping, that creates a much more complete lifecycle. After the user’s workday ends, the desktop can shut down and deallocate, allowing Hydra to move its disk to the lower-cost HDD tier.

Before the user’s scheduled startup, Hydra restores the configured SSD tier and starts the VM so that the desktop is ready for normal use again. The organization captures savings during the idle window without requiring the user to understand anything about disk SKUs, autoscaling rules, or Azure infrastructure.

Cost optimization strategies that depend on every employee remembering to shut down a machine, submit a request, or follow some special process tend to deteriorate over time. Automating the infrastructure lifecycle while still giving users appropriate self-service controls makes the optimization part of normal operations rather than another behavior IT has to enforce.

Think in Hours, Not Just Machines

Stop treating infrastructure as having a single configuration. A persistent virtual desktop might exist for three years, but that does not mean every component of that desktop needs to operate at its highest-cost configuration for every hour of those three years.

If a typical user needs a desktop for 45 hours during a 168-hour week, that means the desktop is actively required for only about 27% of the available time. The remaining 73% represents an optimization window.

Autoscaling asks whether the organization needs to pay for compute during that window; disk swapping asks whether it needs to pay for high-performance storage during that same period.

The answer does not have to involve compromising the user’s experience. While the desktop is running, give it the storage performance required for the workload. When the desktop is deallocated, move it to a storage tier appropriate for a machine that is doing nothing. Before the user needs it again, move it back. The infrastructure changes according to demand while the experience presented to the user remains consistent.

AVD Cost Optimization Shouldn’t Stop at the Power Button

As environments grow and organizations become more disciplined about cloud costs, optimization needs to extend beyond whether a VM happens to be running.

Every persistent resource creates an opportunity to ask a simple question: are we paying for capability right now that nobody is consuming?

A premium-performance disk attached to a deallocated desktop is a good example. There may be an excellent reason for that disk to deliver higher performance during the workday, but there is very little benefit to paying for the same capability throughout nights and weekends.

Hydra’s automated disk swapping applies the same elasticity organizations already expect from compute to the storage underneath their virtual desktops. When the desktop is needed, its disk is restored to the appropriate performance tier. When the desktop is deallocated, the disk can move to inexpensive HDD storage until that performance is required again.

On one virtual desktop, the difference may seem insignificant, but across hundreds or thousands of persistent desktops, multiplied by every idle night, weekend, and holiday throughout the year, it becomes another meaningful layer of AVD cost optimization.

Want to learn more about how Hydra can help drive AVD cost optimization? Get a demo today!