Windows 11 25H2: WSAIFabricSvc launches 8 WorkloadsSessionHost processes and consumes about 5 GB of RAM

Question

Windows 11 25H2: WSAIFabricSvc launches 8 WorkloadsSessionHost processes and consumes about 5 GB of RAM

Leonid Frolov 20

Hello,

I would like to report a serious memory consumption issue in Windows 11 25H2 related to WSAIFabricSvc and WorkloadsSessionHost.exe.

System information:

Device: Microsoft Surface Laptop for Business
RAM: 32 GB
Windows version: 10.0.26200.8894
Windows 11 version: 25H2
WindowsWorkload.Manager package: WindowsWorkload.Manager.1_1.2511.1196.0_x64__8wekyb3d8bbwe

Problem description:

After starting Windows, I noticed unusually high memory consumption. Approximately 16 GB of 32 GB RAM was already in use, even though the computer was not running an especially heavy workload.

Task Manager initially showed several WorkloadsSessionHost processes. I therefore collected memory statistics using PowerShell.

There were eight instances of:

C:\Program Files\WindowsApps\

WindowsWorkload.Manager.1_1.2511.1196.0_x64__8wekyb3d8bbwe\

WorkloadsSessionHost.exe

All processes were started with the following command line:

-ServerName:Microsoft.Windows.Private.Workloads.SessionHost

All eight processes had the same parent process.

Before stopping the Windows AI Fabric service, the memory statistics were:

Physical memory used: 15.94 GB
Available memory: 15.68 GB
Committed memory: 17.91 GB
Commit usage: 53.1%
WorkloadsSessionHost instances: 8
WorkloadsSessionHost private memory: 2449.1 MB
WorkloadsSessionHost combined working set: 4454 MB

One WorkloadsSessionHost instance alone showed:

Private memory: 575.9 MB
Working set: 2969.1 MB
CPU time: approximately 1 second

The processes were almost completely idle, but continued to retain several gigabytes of memory.

I then performed the following diagnostic test in an elevated PowerShell session:

Stop-Service WSAIFabricSvc -Force

Get-Process WorkloadsSessionHost -ErrorAction SilentlyContinue |

Stop-Process -Force

After waiting 20 seconds, all WorkloadsSessionHost processes were gone.

The measured amount of physical memory released was:

5.19 GB

A second memory report after stopping the service showed:

Physical memory used: 11.30 GB
Available memory: 20.37 GB
Committed memory: 12.74 GB
Commit usage: 37.8%
WorkloadsSessionHost instances: 0

Therefore, stopping WSAIFabricSvc and terminating WorkloadsSessionHost released:

Approximately 4.6–5.2 GB of physical memory
Approximately 5.17 GB of committed memory

This is a reproducible and significant amount of memory on a machine with 32 GB RAM.

Additional observations:

Removing or disabling Recall did not resolve the problem.
After rebooting, eight WorkloadsSessionHost processes returned.
Their memory consumption remained approximately the same.
The processes originate from an official Microsoft WindowsApps package.
The problem is not caused by Chrome, Claude, Visual Studio Code, or another third-party process.
Stopping WSAIFabricSvc immediately resolves the excessive memory consumption.
Setting WSAIFabricSvc to Manual and stopping it keeps WorkloadsSessionHost from consuming the memory, at least until another Windows component starts the service again.

This behaviour appears to be either:

Excessive preloading of Windows AI workloads or models;
Incorrect retention of shared or memory-mapped resources;
A memory leak in Windows AI Fabric or Windows Workload Manager.

Five gigabytes of background memory consumption for inactive AI functionality is excessive and noticeably reduces system performance.

Please investigate the following components:

WSAIFabricSvc
WindowsWorkload.Manager
WorkloadsSessionHost.exe
Microsoft.Windows.Private.Workloads.SessionHost

I can provide additional PowerShell reports, process IDs, screenshots, and diagnostic logs if required.

Thank you.

Deleted

This comment has been deleted due to a violation of our Code of Conduct. The comment was manually reported or identified through automated detection before action was taken. Please refer to our Code of Conduct for more information.
Santiago Carvalho 0 Reputation points

2026-08-31T20:26:49.26+00:00

Windows 11 25H2 – WorkloadsSessionHost consuming 4–5 GB of RAM on Ryzen AI 7 350 – WSAIFabricSvc workaround

I have a Lenovo IdeaPad Slim 5 14AKP10 (83NJ), with Ryzen AI 7 350, 16 GB DDR5-5600 RAM and Windows 11 25H2 (build 26200.9278).

I experienced severe and reproducible RAM consumption caused by multiple WorkloadsSessionHost.exe processes.

The symptoms were:

Windows 11 idle initially used approximately 33–40% of RAM.

After some time, multiple WorkloadsSessionHost.exe instances appeared.

The number of instances increased from 1 to 3, 4, 7 and eventually 9.

CPU, GPU and NPU usage of the processes remained essentially at 0%.

Total RAM usage increased to approximately 60–70%.

With 16 GB RAM, this represented several gigabytes of memory being consumed while the computer was otherwise idle.

The processes came from:

C:\Program Files\WindowsApps\WindowsWorkload.Manager.1_1.2511.1196.0_x64__8wekyb3d8bbwe

The installed package was:

WindowsWorkload.Manager.1_1.2511.1196.0_x64__8wekyb3d8bbwe

I also found that WorkloadManager appeared together with the WorkloadsSessionHost processes.

I tested several configurations and found a reproducible relationship with the Windows AI Fabric infrastructure.

Initially, setting WSAIFabricSvc to Manual was not sufficient. The service could remain stopped while the workloads were subsequently created.

I then disabled the service completely:

WSAIFabricSvc → Startup type: Disabled

and restarted Windows.

After reboot:

RAM stabilized around 35–36%.

CPU was approximately 2%.

GPU: 0%.

NPU: 0%.

No WorkloadsSessionHost processes appeared.

I then opened Windows Settings and performed several searches in the Settings search box.

The workloads still did not appear, and RAM remained around 38%.

This was a significant improvement compared with the previous behavior, where using the Windows Settings search was followed by the creation of multiple WorkloadsSessionHost processes and RAM usage rising to approximately 60–70%.

I am therefore currently leaving WSAIFabricSvc disabled.

This appears to be a workaround rather than a definitive fix. It also disables Windows AI Fabric functionality, so users should understand that some Windows AI features may no longer function.

I am posting these observations because several other Windows 11 25H2 users have reported very similar behavior involving WorkloadsSessionHost.exe, WindowsWorkload.Manager and WSAIFabricSvc.

If anyone has identified a way to disable only the offending workload or the specific Windows Settings search integration, rather than disabling the entire AI Fabric service, I would be interested in testing it.

1 answer

Your answer

Deleted

This comment has been deleted due to a violation of our Code of Conduct. The comment was manually reported or identified through automated detection before action was taken. Please refer to our Code of Conduct for more information.

Answer 1

Daphne Huynh (WICLOUD CORPORATION) 1,305 Microsoft External Staff Moderator

Welcome to Microsoft Q&A Forum!

Thank you for the detailed investigation and for providing before-and-after memory statistics.

Based on your findings, WSAIFabricSvc (Windows AI Fabric Service) and the associated WorkloadsSessionHost.exe processes appear to be closely associated with the unusually high memory usage observed on the system. Across multiple tests, stopping the service and terminating the workload host processes consistently reduced both physical and committed memory consumption by approximately 5 GB. The fact that this reduction is both significant and reproducible suggests that these components are reserving a substantial amount of memory even when there is little or no visible workload activity.

While the available data does not conclusively determine the root cause, the results indicate that the behavior warrants further investigation. At this stage, it is difficult to determine whether the memory usage is the result of AI model preloading and caching, retention of memory-mapped resources, aggressive resource management, or a potential software defect. However, several observations strengthen the correlation:

The behavior is consistently reproducible across system reboots, making a temporary or one-time condition unlikely.
Eight WorkloadsSessionHost.exe instances are continuously launched by the WindowsWorkload.Manager package and remain active over time.
Removing Recall does not change the behavior, suggesting that the memory consumption is not solely tied to Recall-related features.
Stopping WSAIFabricSvc consistently frees several gigabytes of memory, showing a clear and direct relationship between the service and the elevated resource usage.

Taken together, these findings suggest that the memory consumption is persistent, systemic, and closely linked to the Windows AI Fabric infrastructure rather than to individual applications or normal user activity. CPU utilization remains minimal despite the sustained high memory consumption.

As a temporary mitigation, I would like to provide you some useful information that you may consider:

1. Temporarily disable the Windows AI Fabric service

Stop-Service WSAIFabricSvc -Force Set-Service WSAIFabricSvc -StartupType Manual

This prevents the service from starting automatically after reboot and may reduce the immediate memory impact.

2. Verify whether Windows re-enables the service

After installing Windows updates or feature updates, confirm that the service startup type remains unchanged
If the service is restarted automatically, determine which Windows feature or component depends on it before disabling it again.

3. Keep Windows and Windows AI components updated

Install the latest cumulative updates for Windows 11 25H2.
Check for updates to the WindowsWorkload.Manager package, as changes to AI-related components may be delivered independently through Store or system package updates.

4. Review whether AI features are required

If you do not currently rely on Recall, Copilot+, or other AI-powered Windows experiences, leaving WSAIFabricSvc set to Manual may be a reasonable temporary workaround.

Reference: What are Windows AI APIs? | Microsoft Learn

5. Avoid repeatedly terminating WorkloadsSessionHost.exe

While forcefully ending individual WorkloadsSessionHost.exe processes may temporarily free memory, they are managed by the WindowsWorkload.Manager package and may be restarted automatically. Stopping the underlying service is generally the more effective and controlled approach.

At present, your testing demonstrates a clear correlation between WSAIFabricSvc and the memory consumption. For the time being, I recommend setting the service to Manual, as this appears to be the most practical workaround. I appreciate your patience while we continue to investigate and determine whether this behavior is expected or can be addressed in a future update.

I hope this information is helpful and thank you for choosing Microsoft Q&A to raise your concern.

Leonid Frolov 20 Reputation points

2026-07-24T05:57:22.5533333+00:00
Thank you for your response.

I would like to clarify one important point: I do not use Microsoft AI services on this computer.

I do not use Microsoft Copilot, Recall, Click to Do, or any other Microsoft AI feature intentionally. My AI tools are third-party applications and web services, primarily Anthropic Claude. They do not depend on Windows AI Fabric or WorkloadsSessionHost.

This is why the observed behaviour is especially concerning.

Windows automatically starts eight WorkloadsSessionHost processes and allows them to consume approximately 4.6–5.2 GB of memory, even when I am not actively using any Microsoft AI functionality.

On a computer with 32 GB of RAM, this represents roughly 15–16% of all physical memory. In some startup sessions, total system memory usage reached approximately 60–66%, while after stopping WSAIFabricSvc and removing the WorkloadsSessionHost processes, it dropped to approximately 38%.

I understand that Windows may preload models, runtimes, or shared resources to make AI features available more quickly. However, consuming several gigabytes of RAM for features that the user has not enabled and does not use is not reasonable behaviour.

The operating system should not silently reserve such a significant portion of the computer’s resources for optional AI functionality.

At minimum, I would expect:

Microsoft AI components to remain unloaded until an AI feature is actually used;

A clear Windows setting allowing users to disable the entire AI Fabric infrastructure;

Memory to be released when Recall, Click to Do, Copilot, and related features are disabled;

Documentation explaining why eight WorkloadsSessionHost instances are required and what each one is doing;

Confirmation whether approximately 5 GB of idle memory consumption is considered expected behaviour or a defect.

Currently, setting WSAIFabricSvc to Manual significantly improves memory usage and overall system responsiveness. However, manually changing a system service should not be necessary just to prevent unused optional AI components from consuming a large amount of RAM.
Daphne Huynh (WICLOUD CORPORATION) 1,305 Reputation points Microsoft External Staff Moderator

2026-07-24T09:16:03.1733333+00:00
Thank you for your detailed follow-up and for clarifying your usage scenario.

I understand your concern, particularly given that the observed memory consumption appears immediately after startup, is reproducible across reboots, and occurs regardless of whether Recall or other visible AI-related Windows features are actively being used.

Your feedback also raises several valid expectations from a user perspective:

AI-related components should ideally be loaded on demand when the corresponding feature is used.

Users should have clear visibility into which Windows components are consuming resources and why.

Disabling AI features should release resources that are no longer required.

Documentation should explain the purpose and resource requirements of the underlying AI infrastructure.

Microsoft should clarify whether the observed memory footprint is expected or indicative of an issue.

At this time, setting WSAIFabricSvc to Manual appears to be an effective workaround on your system and aligns with the memory reductions you have measured. However, I agree that requiring manual service configuration is not an ideal long-term solution if the associated functionality is not being used.

Thank you again for providing detailed measurements, reproducible test results, and a clear explanation of the impact on your system. The data you have collected helps establish a strong correlation between WSAIFabricSvc, WorkloadsSessionHost.exe, and the elevated memory usage, and it provides important information for further review by the Windows engineering team.
Leonid Frolov 20 Reputation points

2026-07-24T09:47:26.09+00:00

Thank you for the clarification.

Could you please confirm whether this issue has actually been escalated to the Windows engineering team?

If so, is there an internal bug ID, tracking ID, or Feedback Hub reference associated with it?

At the moment, the workaround is effective, but I would like to understand whether Microsoft is formally investigating this as a product defect rather than only documenting the workaround in this forum thread.

Also, please confirm whether approximately 5 GB of memory usage by eight idle WorkloadsSessionHost processes is considered expected behavior on a 32 GB Copilot+ PC.
Daphne Huynh (WICLOUD CORPORATION) 1,305 Reputation points Microsoft External Staff Moderator

2026-07-27T03:32:17.0033333+00:00
Thank you for your follow-up questions.

I understand your request for clarification regarding the status of the investigation.

Regarding your question about escalation, I am unfortunately unable to confirm whether this issue has been formally escalated to the Windows engineering team, nor can I provide any internal bug IDs, tracking numbers, or engineering work item references. Please kindly note that, due to Microsoft policy, internal tracking IDs and engineering records are not publicly disclosed. Any information related to an internal investigation would only become publicly available once the review process has been completed and are decided to publish the findings through official channels, such as guidance articles, release notes, known issue documentation, or other public communications. Until then, I apologize I am unable to verify, discuss, or share the status, outcome, or existence of any potential internal investigation.

If you would like direct case tracking, status updates, or a formal review of your findings, I recommend opening a support case with Microsoft Support. This allows a support engineer to review your collected diagnostics, determine whether the issue aligns with any known investigations, and, if appropriate, escalate it through the proper support channels.

Additionally, if your concern is specifically related to Windows AI components, you can submit feedback using Feedback Hub and open a support request through Microsoft's official support channels:

Windows help and learning | Microsoft Support

Explore the Feedback Hub - Windows Insider Program | Microsoft Learn

Contact Us - Microsoft Support

Regarding your question about memory consumption, I am also unable to confirm whether approximately 5 GB of memory usage across eight WorkloadsSessionHost.exe processes is considered expected behavior on a Windows 11 25H2 Copilot+ PC.

The information you have gathered provides valuable evidence for further investigation. If obtaining a definitive engineering assessment is important, opening a support case would be the best path, as it allows the issue to be formally reviewed, tracked, and potentially escalated through Microsoft's support process.

Once again, thank you for the thorough diagnostics, detailed testing, and clear documentation of your findings. Your investigation provides valuable information that can assist with further review of the behavior.
Leonid Frolov 20 Reputation points

2026-07-27T06:07:16.9666667+00:00

Thank you for confirming the clear and reproducible correlation between WSAIFabricSvc, WorkloadsSessionHost.exe, and the approximately 5 GB memory impact. However, I still do not have an answer to the key question: Has this issue actually been escalated to the Windows engineering team as a product defect? Please confirm whether an internal bug ID, tracking ID, or Feedback Hub reference has been created. Also, please clarify whether Microsoft considers the following behaviour expected on a 32 GB Copilot+ PC: - 8 idle WorkloadsSessionHost.exe processes - approximately 2.45 GB private memory - approximately 4.45 GB combined working set - approximately 5 GB of physical and committed memory released after stopping WSAIFabricSvc At the moment, the proposed workaround only confirms that disabling the service reduces the memory usage. It does not explain why the service consumes this amount of memory when no Microsoft AI feature is being used. I would also like to point out that the PowerShell command in the response is syntactically incorrect. The two commands must be separated by a newline or semicolon: Stop-Service WSAIFabricSvc -Force Set-Service WSAIFabricSvc -StartupType Manual
Deleted

This comment has been deleted due to a violation of our Code of Conduct. The comment was manually reported or identified through automated detection before action was taken. Please refer to our Code of Conduct for more information.
Leonid Frolov 20 Reputation points

2026-07-27T06:11:07.3733333+00:00
Thank you for the workaround.

I would like to ask a more fundamental question.

Is there any officially supported way to completely remove Windows AI Fabric, WindowsWorkload.Manager, WorkloadsSessionHost, Recall, Click to Do, and other related Microsoft AI components from the system?

I do not use Microsoft AI services on this computer. I use third-party AI applications and web services, and I do not need Windows to preload or maintain local Microsoft AI infrastructure in the background.

I purchased this computer and a licensed copy of Windows to run my own applications. The system resources belong to the user and should remain available for the software the user has chosen to run.

In my case, these unused components consumed approximately 5 GB of memory. That is a significant amount of RAM on a 32 GB system.

Setting WSAIFabricSvc to Manual or Disabled is only a workaround. The components remain installed, and Windows updates may enable them again.

I would therefore like to know:

Is there an official Microsoft script or supported procedure to completely remove these AI components?

Can WindowsWorkload.Manager and Windows AI Fabric be uninstalled without damaging Windows Update or system stability?

Is there a supported Group Policy, registry setting, or optional feature that disables the entire Microsoft AI stack system-wide?

Does Microsoft plan to provide users with a single setting such as “Remove all Windows AI features”?

If complete removal is not supported, why are optional AI components allowed to consume several gigabytes of memory when the user has not enabled or used them?

Users should be able to decide whether optional AI functionality is installed and allowed to consume their computer’s resources.

Please provide an official and supported removal or permanent disablement method.
Leonid Frolov 20 Reputation points

2026-07-27T13:52:22.4733333+00:00
Thank you for the clarification.

Since Microsoft Q&A cannot confirm whether this issue has been escalated or whether this memory consumption is expected, I would like to ask a practical question.

Is there any officially supported way to completely remove Windows AI Fabric, WindowsWorkload.Manager, WorkloadsSessionHost, Recall, Click to Do, and other related Windows AI components from the computer?

I do not use Microsoft AI services or built-in Windows AI features. I do not use Copilot, Recall, or Click to Do. I therefore do not want Windows to preload or maintain several gigabytes of AI infrastructure in the background.

Disabling WSAIFabricSvc is only a workaround. The components remain installed, Windows may re-enable the service after an update, and the user has no clear control over the entire AI stack.

As a licensed Windows user, I believe I should be able to decide whether optional AI functionality is installed on my computer and allowed to consume its memory, storage, CPU, GPU, or NPU resources.

Please clarify:

Is there an official Microsoft script or supported procedure to remove the entire Windows AI infrastructure?

Can WindowsWorkload.Manager be uninstalled safely?

Is there a supported Group Policy or registry policy that permanently disables all Windows AI components?

Does Microsoft plan to provide a single Windows setting to remove or disable all AI features system-wide?

If complete removal is not supported, why are optional AI components allowed to consume several gigabytes of memory when the user does not use them?

I would prefer a supported removal method rather than repeatedly disabling system services after Windows updates.

Windows 11 25H2: WSAIFabricSvc launches 8 WorkloadsSessionHost processes and consumes about 5 GB of RAM

Windows 11 25H2 – WorkloadsSessionHost consuming 4–5 GB of RAM on Ryzen AI 7 350 – WSAIFabricSvc workaround

1 answer

Your answer