Microsoft has begun automatically pushing an update that fine-tunes how Windows 11 handles on-device AI tasks on laptops and tablets powered by Qualcomm Snapdragon chips. The KB5089617 component update, released this week, targets the Qualcomm QNN Execution Provider — a piece of the Windows AI stack that translates machine learning models so they can tap into the dedicated neural processing units (NPUs) inside newer Snapdragon processors.

For most people, the update will install without any fanfare. But it represents another step in Microsoft’s effort to make local AI acceleration on Windows on Arm devices more reliable and performant, especially as Copilot+ PCs become more common.

What this update actually contains

KB5089617 is not a typical cumulative update. It doesn’t touch Windows security patches, the Start menu, or File Explorer. Instead, it updates a single AI component: the Qualcomm QNN Execution Provider, bringing it to version 2.2604.2.0. The update applies to devices running Windows 11 version 24H2 or version 25H2, and it appears in your update history under the label “Windows ML Runtime Qualcomm QNN Execution Provider Update (KB5089617).”

The package replaces an earlier component update, KB5077526, and requires that the latest cumulative update for your version of Windows 11 already be in place. If your system isn’t fully patched, Windows Update will take care of that prerequisite first, then deliver KB5089617 without you needing to lift a finger.

There’s no off‑switch for this update — it’s delivered automatically, and there’s no separate manual download from the Microsoft Update Catalog because it’s a component update, not a standalone installer. If you’re connected to the internet and using a supported Qualcomm‑based PC, you’ll get it as part of the normal Windows Update flow.

Who should care — and why

If you’re an everyday Windows user, this update will likely be invisible. You won’t see a new feature or a performance boost on your desktop. What you will have is a more up‑to‑date foundation for any application that relies on the ONNX Runtime to run machine learning models on Qualcomm hardware.

For developers working on Windows apps that use AI, the update matters more directly. ONNX Runtime is the execution engine that many frameworks — from PyTorch to TensorFlow — can target to achieve cross‑hardware compatibility. When an ONNX model is loaded, ONNX Runtime can pick the “execution provider” best suited for the underlying hardware. On a Snapdragon PC, that often means the Qualcomm QNN Execution Provider, which converts the ONNX model into a QNN graph consumable by Qualcomm’s AI Engine Direct SDK and then runs it on the NPU, GPU, or DSP. By keeping this provider current, Microsoft ensures that Windows-supplied libraries are in sync with the latest optimizations and fixes from Qualcomm, so developers don’t have to bundle their own version unless they need specific custom behavior.

IT administrators managing fleets of Windows 11 devices — especially those deploying Copilot+ PCs or testing AI‑accelerated workloads — should care because consistency matters. If you’re troubleshooting why an in‑house AI application isn’t seeing NPU acceleration on a Snapdragon device, checking that KB5089617 is installed is now on your diagnostic checklist. Since it’s a component update rather than a full OS feature update, it can slip under the radar if you’re only tracking Patch Tuesday rollouts.

The bigger picture: how on‑device AI is evolving on Windows

KB5089617 didn’t appear in a vacuum. It’s the latest in a series of quiet updates that Microsoft has been issuing to keep the AI plumbing on Windows systems current. The Qualcomm QNN Execution Provider first became available as part of the Windows ML Runtime component updates in early 2024, following the announcement of the Snapdragon X Elite and X Plus platforms and the push for Copilot+ PC features.

The reasoning is simple: for AI experiences like Windows Studio Effects, Cocreator in Paint, or Live Captions to feel instantaneous, the heavy lifting must happen on the device rather than in the cloud. That requires tight integration between the machine learning models, the ONNX Runtime, and the silicon underneath. Qualcomm’s AI Engine, which contains a Hexagon NPU, is one of the primary hardware targets for this PC‑class AI acceleration, and the QNN Execution Provider is the translator that makes that possible.

Previous updates — notably KB5077526 — laid the groundwork, and KB5089617 builds on that with unspecified “improvements” according to Microsoft’s support document. While the changelog is short on specifics, such component updates often include performance optimizations, bug fixes for edge cases in model compilation, or compatibility enhancements for newer ONNX opsets. They might also address issues reported by early adopters or developers building applications that target the Snapdragon NPU.

It’s worth remembering that the same execution provider model applies to other chipsets, too: Intel has its OpenVINO Execution Provider, and AMD has MIGraphX. The idea is that Windows, via ONNX Runtime, can automatically leverage whatever accelerator is available. But Snapdragon PCs, being Microsoft’s poster children for the Arm‑based AI PC wave, get particularly close attention.

What to do right now

For the vast majority of people, the answer is: nothing. Windows Update will handle it. But a little awareness can help when things don’t go as planned.

Check if it’s installed
Open Settings > Windows Update > Update history. Look for:
- Windows ML Runtime Qualcomm QNN Execution Provider Update (KB5089617)

If it’s there, you’re set. If not, make sure your system has the latest cumulative update installed first. Click “Check for updates” and let Windows do its thing. On a metered connection, you might need to give permission for the download.

For developers and testers
If you’re working with ONNX Runtime on a Snapdragon device, you can verify the version of the QNN Execution Provider by querying the available providers in your code. For example, in Python:

import onnxruntime as ort
print(ort.get_available_providers())

If ‘QNNExecutionProvider’ appears and you can load a model, the system‑wide update is in effect. If you’re packaging a specific ONNX Runtime build with your app, the provider version you ship will take precedence, but the system update ensures that any Windows‑level ML tasks pick up the latest stable code.

For IT admins
Add KB5089617 to your tracking list. Since it’s a component update, it might not appear in all WSUS or Configuration Manager syncs the same way full updates do, but it will reach devices via Windows Update for Business if you’ve configured it to include driver and component updates. If you’re troubleshooting AI acceleration, ensure that update history shows this KB.

Outlook: a quiet cadence of AI plumbing updates

KB5089617 won’t grab headlines, and that’s by design. It’s infrastructure — the kind of update that keeps the AI engines humming without user intervention. Expect to see more of these component updates as Windows on Arm matures and as more applications ship with ONNX‑based models. Microsoft’s decision to deliver them through Windows Update rather than requiring manual SDK installations signals that they view these AI runtimes as part of the operating system fabric, not optional add‑ons.

The next milestone to watch is the public release of Windows 11 version 25H2, which will presumably ship with an even newer QNN Execution Provider baked in. In the meantime, if you’re running a Snapdragon‑powered device and diving into local AI workloads, know that Microsoft is keeping the plumbing up to date — often without you ever noticing.