Table of Contents What is new in Windows 11 25H2? Setup and configuration Hypothesis and results Microsoft officially launched Windows 11 25H2 back in September 2025, and while its new features are making headlines, the critical question for the EUC community remains: what is the performance impact of 25H2 for VDI scenarios? For every new OS release, rolling it out to production without understanding its impact on user density, scalability, and application responsiveness can be a significant gamble that could negatively affect the user experience and server scalability. This research will provide data to answer these questions by analysing the VDI performance of Windows 11 25H2 Enterprise, specifically addressing whether Microsoft optimised 25H2 and whether it could affect the scalability of your VDI environment compared with previous releases. What is new in Windows 11 25H2? This release will be delivered as an enablement package (eKB) on top of 24H2, focusing on improvements rather than major architectural or kernel-level changes. For VDI stability and image management, this is a welcome approach. From an enterprise perspective, the most impactful change for VDI optimization is the introduction of policy-based removal of Microsoft Store apps, allowing for cleaner, lighter base images. This is complemented by features such as Quick Machine Recovery to enhance lifecycle management. However, the most impactful changes for VDI administrators may be what’s removed: this update officially deprecates and removes legacy components, most notably the Windows Management Instrumentation command-line (WMIC), which could break older optimization and monitoring scripts. On the user-facing side, 25H2 includes the redesigned Start Menu and continues to expand AI capabilities through Copilot. The AI capabilities of 25H2 might not be directly relevant to VDI environments without access to GPUs or NPUs; however, expanded AI integration could introduce a new, potentially significant resource load that must be quantified to understand its impact on VDI scalability and user density. For a complete overview of all new functionalities, see the Microsoft website . Setup and configuration This research has been executed with LoadGen in an on-premises lab environment using our default testing methodology . The default delivery technology is Citrix Virtual Apps & Desktops version 2507 LTSR. To have a clear understanding of how the performance differs, the following scenarios are included: Windows 11 23H2 Windows 11 24H2 Windows 11 25H2 Citrix Virtual Apps and Desktops 2507 LTSR has full support for all Windows 11 releases. Citrix maintains an updated list of supported OS’s . Please note that Windows 11 is also supported on CVAD 2402 (CU3 and later) and 2203 (CU7 and later). Any known issues with Windows 11 releases are maintained on the Citrix website. In the GO-EUC lab, for each Windows version, a vSphere template has been created from the original ISO using Hashicorp Packer. Based on each template, a build machine has been created, where the default application sets are installed and configured. In this deployment, the latest available updates are installed, and the image is optimized using the default selected template in the Citrix Optimizer, as is best practice. To mitigate unexpected Windows Defender behaviour that could affect test results, all Microsoft Defender-related services were disabled. Because the tests for this research were conducted over multiple days, leaving Defender enabled could trigger background scans during test runs, skewing results due to altered load patterns. This configuration is strictly for testing purposes and is not recommended for production workloads. All virtual machines are configured with 4 vCPUs and 8GB of memory each. A total of 32 machines are deployed. The number of VMS did not fully utilize the host capacity, which has 512GB of memory and an AMD EPYC 7542 32-Core Processor. Fully utilizing the host could introduce unwanted or unexpected bottlenecks, skewing the results. Hypothesis and results Based on previous research, it is hypothesised that the new Windows 11 25H2 release will have a slightly higher overall performance footprint than earlier versions. The main driver is expected to be expanded AI capabilities, along with potentially added services and background processes. Even though this update primarily focuses on refinements and optimizations rather than major functional or architectural changes, some increase in resource utilization is still expected. To validate this hypothesis, the analysis examines performance from two perspectives: the underlying hypervisor with the host metrics and the metrics that are collected at the virtual machine level. Let’s start with the hypervisor perspective.

Host CPU utilization shows a slight increase in average usage across the Windows versions, with the largest difference observed between 24H2 and 25H2. On average, this difference peaks at just 0.5% utilization, which is not significant and, in the context of GO-EUC, should have no direct impact on overall capacity or scalability. From a host-level perspective, active memory usage follows a pattern similar to CPU utilization, showing a slight increase with Windows 11 25H2. This difference is not significant enough to introduce any direct impact or scalability concerns. The disk command metrics, which represent total disk operations at the host level, show no significant differences across Windows versions, with all versions falling within the same range. Since storage performance is primarily driven by workload activity, this outcome aligns with expectations for the previous metrics. The next charts will zoom in on the performance of the virtual machine perspective. At the VM level, the same pattern is observed, as expected. Based on the average results, these differences would not be noticeable to end users. However, in CPU-constrained environments, they are still worth monitoring to prevent any performance issues that might arise when CPU usage is high. Overall, the memory footprint at the VM level is very similar across all versions, with an average difference of only 1%. To put this into perspective, a 1% difference equates to approximately 81 MB of memory, which is not noticeable during normal day-to-day usage. Interestingly, based on this data, Windows 11 23H2 shows slightly higher memory consumption than 25H2, which may indicate improved optimization or more efficient memory usage by processes in the newer release. Both read and write results are nearly identical, which is expected since disk activity is primarily driven by the workload, and the workload remains consistent across all tests. The charts show the running processes during the test. Even though new functionality is introduced in newer Windows versions, the difference is limited to a maximum of two additional processes. This largely explains why there is little to no measurable performance difference between the Windows releases. Based on the results of this research, Windows 11 25H2 does not introduce any significant performance or scalability regression for VDI environments when compared to 23H2 and 24H2 in our labtests. While the hypothesis anticipated a slightly higher footprint due to expanded AI capabilities and additional background services, the measured impact remains marginal and well within acceptable boundaries for enterprise EUC deployments. From both a hypervisor and virtual machine perspective, CPU and memory utilization show only minimal increases, with differences so small that they are unlikely to affect user density or end-user experience in real-world scenarios. Disk activity remains effectively unchanged, confirming that workload behavior, not the OS version, continues to be the dominant factor for storage performance. The observed increase in the number of running processes in 25H2 provides a plausible explanation for the slight rise in resource usage, but this does not translate into a practical scalability concern. In short, Windows 11 25H2 appears to be a predictable evolution rather than a disruptive release for VDI. For EUC architects and VDI administrators, the takeaway is clear: from a performance and scalability standpoint, there is no reason to delay adoption of Windows 11 25H2. As always, validating the OS in your own environment and workload is best practice, but based on this data, 25H2 can be introduced with confidence without sacrificing scalability or user experience. Photo by Brianna Parks on Unsplash Retrospective of 2025 and prospective of 2026 Evaluating the visual quality and colour differences of fullscreen H.265 Video compression with YUV 4:2:0 and YUV 4:4:4 in Citrix HDX You may also like Beyond Balanced: Benchmarking Windows 11 Power Plans Windows 11 25H2 vs 26H2 preview: an early performance comparison Azure VDI Storage Benchmark: Premium SSD vs Standard SSD Performance and Cost Breakdown Upcoming events Dutch Citrix User Group Echteld, Netherlands GO-EUC is a full community effort that is based on a non-profit foundation. All contributions are based on a voluntary basis to share insights with the community. Please consider helping this community by buying a beer .