AVD Image Sprawl Accelerates Over Time…but Containerizing Your Applications Can Eliminate It
Microsoft Azure Virtual Desktop (AVD) has grown significantly in the last 5 years. The infrastructure supporting the many AVD customers is now well battle hardened and proven. When AVD environments struggle, the root cause is rarely systemic networking, compute, or service related. More often, the friction builds slowly through application management decisions that feel reasonable at the time but compound over months and years.
One of the clearest manifestations of this is image sprawl, which we briefly highlighted in our blog earlier this year (2025).
Most AVD environments do not begin complex. They typically start with a single, well-tested desktop image deployed in a host pool. Early results are positive. Users are productive, performance is stable, and operational effort feels contained.
Then applications begin to accumulate.
Installing Applications into Images Is a Legacy Practice
Installing applications directly into a desktop image is a pattern inherited from traditional on-prem VDI. It is even a method used for physical endpoints several years ago. It was seen as the path of least resistance and in some cases the only viable option for some organizations.
Each application added to an image modifies the operating system in ways that are not always visible or predictable. Registry changes, shared libraries, services, scheduled tasks, and environment variables all become part of the image’s state. Even when applications are removed, remnants often remain behind.
Over time, images drift from their original baseline. Performance issues appear without a clear trigger. Behaviour becomes inconsistent between desktops that were supposedly derived from the same base image. Troubleshooting grows more difficult because the image no longer reflects a known-good state.
This gradual degradation is what many teams recognize as image rot and once it sets in, images become risky to change and expensive to replace.
Why Image Sprawl Is Inevitable in AVD Without Containerization
As AVD environments mature, real-world requirements begin to challenge the assumption that a single shared image is viable.
Licensed software is one of the most common pressure points. Many organisations rely on applications that should only be available to a limited group of users due to cost or licensing constraints. When these applications are baked into a shared image, they become visible to everyone on that host pool, even if access is restricted through other means. In heavily regulated environments, that visibility alone can create compliance concerns and can lead to uncomfortable conversations with vendors when it comes time for license renewals. This leads some companies to create a separate host pool and custom image for those who require the licensed software – possibly hosting the application in RemoteApp.
Security-sensitive or legacy applications introduce similar complexity. Not every user should be aware of, or able to interact with older applications that are still required for edge use cases. The typical workaround is to isolate those applications by creating a separate image and often a separate host pool.
Application conflicts further exasperate the problem of image sprawl. When two applications cannot reliably coexist, teams often resolve the issue by separating them operationally rather than addressing the underlying install model.
Individually, these decisions are understandable. Collectively, they make image sprawl unavoidable.
One image becomes two. Two become four. Before long, AVD environments contain multiple images, specialised host pools, and an expanding set of exceptions all driven by application lifecycle limitations rather than user or business needs.
The Compounding Cost of Image Sprawl
The immediate cost of image sprawl is operational overhead. Customers have estimated the additional overhead introduced by this sprawl to cost in excess of $200k per year. Another cost of image sprawl is disk storage due to a reliance on traditional OS disks required due to operational workflows for application management – by standardizing on ephemeral disks across an estate of 1k virtual machines it is possible to save $300k a year but moving to these free disks is made difficult when installing application in images. Every additional image requires its own build process, testing cycle, patching routine, and validation effort. Changes that were once straightforward now ripple across multiple images and host pools.
More damaging, however, is the impact on agility. Application updates slow down because they are tied to image rebuilds. Testing windows expand. Rollbacks become increasingly complex. As a result, updates are deferred, increasing the risk of security vulnerabilities and compatibility issues.
In recent years, we dealt with a customer who had over 100 images in their on-premises VDI platform. Updating and maintaining these images was literally a full-time job for someone on the EUC team. When this firm weighed up moving from their on-prem VDI to AVD they realized the mountain of work they had ahead of them to try to reverse engineer the existing images and decided ultimately to course correct away from installing all applications into images. Not only saving considerable money but reducing operational overheads significantly.
There is also an organisational cost. Environments with significant image sprawl become fragile knowledge systems, understood in full by only a small number of people. When those individuals are unavailable, progress stalls and risk increases.
At that point, AVD begins to resemble the legacy environments it was intended to modernise.
Decoupling Applications from Images Changes Your AVD Cost Model
The underlying issue is not Azure Virtual Desktop itself. It is the assumption that applications must live inside desktop images.
When applications are decoupled from images and delivered dynamically, the AVD operating model fundamentally changes. Images can remain lean, stable, and long-lived. Application lifecycle management becomes independent of host pool lifecycle management.
Cloudpaging application containers are designed to support this approach for virtually any Windows application, including complex and legacy software that traditionally resists modern application delivery methods. Applications are executed in a container sandbox delivered with their own virtual file system, eliminating conflicts and preventing residual changes when applications are updated or removed.
Because applications are delivered dynamically, they can be assigned on a per-user basis rather than per image. In multi-session AVD environments, this ensures that applications are only visible to the users entitled to them so no need to sweat license renewal or exposing legacy applications widely anymore.
Updates, removals, and rollbacks can occur without rebuilding images or reprovisioning session hosts. As a result, many of the scenarios that previously forced new images or dedicated host pools simply disappear.
Customers have achieved over 80% concurrency as a result of delivering application dynamically as Cloudpaging application containers saving 500k compute hours equating to a $200k a year saving.
The dynamic delivery of applications as Cloudpaging application containers also makes the adoption of Infrastructure as Code, in the context of AVD, much simpler. The base image can be kept lightweight and repeatable since virtually all additional Windows applications can be delivered dynamically as users access the desktops. This cuts out a lot back and forth between automated image build and regression testing by not requiring every iteration of an application update to be tested as part of the automated build.
Eliminate Image Sprawl by Tackling the Root Cause
Image sprawl is not caused by poor planning or weak governance. It is the natural outcome of coupling application delivery to image management in an environment designed for elasticity and scale.
By separating, AVD environments become simpler to operate and easier to evolve. Fewer images lead to fewer host pools, shorter testing cycles, and a significantly reduced operational burden over time. Moving to future platforms such as Windows 365 or something else will also be less burdensome since Cloudpaging application containers work on any Windows desktop.
Cost Savings Realized by Leveraging Cloudpager for 2,000 Azure Virtual Desktop
Here’s a comprehensive overview of the estimated cost savings realized by a company leveraging Cloudpager to help establish its transition from a persistent to non-persistent Azure Virtual Desktop environment.
Persistent VDI environments maintain a dedicated virtual machine (VM) for each user, incurring ongoing costs for storage, compute, and maintenance, even when users are inactive.
Non-persistent environments use ephemeral disks and dynamic provisioning, which align resource usage with actual demand. This is particularly impactful for organizations using Azure Virtual Desktop for daily operations, as well as disaster recovery (including Windows 365 licensing, as well).
| Category | Persistent Model | Non-Persistent Model Savings | Description |
| Disk Storage (1,000 VMs) | $300,000 per year | $300,000 per year saved | Eliminating persistent disks and associated snapshot/backups via ephemeral disks. |
| Compute Efficiency (1,000 Users @ 80% Concurrency) | N/A | ~$200,000 per year saved | 20% of VMs (~200) do not need to run continuously. Approximately 500,000 compute hours saved annually at a rate of $0.40. |
| Disaster Recovery (2,000 Users) | ~$520,000 per year | $520,000 per year saved | Persistent DR VMs incur disk and patching costs even when idle; non-persistent setup costs near zero until a DR event. |
| Operational & Maintenance Savings | $100,000 – $200,000 per year | $100,000 – $200,000 per year saved | Reduced time maintaining individual AVDs, patching, and managing images. |
The estimated cost savings for this organization (storage, compute, disaster recovery, and operational efficiencies) is approximately $1.25M to $1.5M USD per year.
Addressing the Real Cost of AVD Image Sprawl
Azure Virtual Desktop works best when desktop images are treated as stable releases, not application delivery mechanisms that get updated frequently.
When applications move out of images and into a dynamic, containerized delivery model, image sprawl stops accelerating and, in many environments, begins to reverse. Application updates become less disruptive to employees and overall costs for running AVD falls.
Join the Movement to Containerize Everything
For all things applications, stick with Numecent. Want to learn more? Request a demo with our technical solutions team and sign up for our newsletter below: