Microsoft has rolled out KB5089618, a quiet but significant Windows Update that services the Qualcomm QNN Execution Provider on Windows 11 version 26H1. The update, which requires no user action beyond staying current on Windows updates, brings the component to version 2.2604.2.0 and replaces the older KB5078978. It is the latest example of how Windows is separating AI hardware acceleration from individual app installers—and, perhaps, a glimpse of how the OS will manage on-device AI going forward.

The Update at a Glance

This is not a run-of-the-mill Patch Tuesday rollup. KB5089618 touches a single, specialized piece of the AI pipeline: the execution provider that lets ONNX Runtime workloads tap into Qualcomm’s neural processing silicon—the Hexagon Tensor Processor and related NPU blocks inside Snapdragon X Series PCs. Specifically, it updates the Windows ML Runtime Qualcomm QNN Execution Provider, and you’ll find it listed exactly that way in Settings > Windows Update > Update history.

There is no manual download link and no preview toggle. Microsoft delivers the package automatically to devices running Windows 11, version 26H1 that already have the latest cumulative update for that release. If your PC is on Windows 11 24H2 or 25H2, you won’t see it—and that’s by design. The QNN Execution Provider update track is platform-locked, so different Windows versions get their own servicing packages.

Who Gets It—and Who Doesn’t

Right now, the audience for KB5089618 is narrow. Windows 11 26H1 is a scoped release built for select new hardware, not a broad feature update for existing machines. “It is designed to support new device innovation in partnership with OEMs and silicon partners,” Microsoft has said, and it is not offered as an in-place upgrade from any other Windows 11 version. That means the update will land almost exclusively on fresh Snapdragon-powered laptops and tablets that ship with 26H1 preinstalled.

If you’re running an Intel or AMD system, or even an older Qualcomm laptop that received Windows 11 24H2/25H2, KB5089618 simply won’t apply. Those devices have their own QNN Execution Provider updates (identified by KB numbers in the KB5077xxx range) that follow a separate servicing cadence. The key takeaway: not seeing KB5089618 in your Update history does not indicate a problem; it just means your machine is on a different servicing track.

What’s Inside: Not Just a Routine Patch

To understand why this update matters, you have to look at the plumbing beneath Windows’ local AI features. ONNX Runtime—the cross-platform inference engine used by Windows ML, Copilot+ PC experiences, and many third-party AI apps—can draw on hardware accelerators through execution providers. The Qualcomm QNN Execution Provider is the bridge that translates ONNX models into something the Snapdragon NPU or HTP (Hexagon Tensor Processor) can run.

The QNN provider uses the Qualcomm AI Engine Direct SDK (the QNN SDK) to construct an optimized graph from the model. Then, a backend library—typically the HTP backend for NPU offload—executes that graph on dedicated silicon. Without this layer, AI models would fall back to the CPU, using more power and delivering slower results.

What KB5089618 does is bring that bridge layer up to date with the current 26H1 servicing baseline. Microsoft’s support note is tight-lipped about exact changes, just promising “improvements to the Qualcomm QNN Execution Provider AI component.” In practice, those improvements could mean better operator coverage, reduced CPU fallbacks, compatibility fixes for newer ONNX model formats, or performance tweaks that squeeze out more throughput or lower latency. You won’t see a splash screen, but apps that lean on Windows ML—Camera effects, Studio Effects, any Store app that calls the Windows.AI.MachineLearning API—may suddenly feel a bit snappier or more responsive.

Why This Matters for Your AI Apps

For everyday users, KB5089618 is a background enhancement. It doesn’t alter the desktop, the Start menu, or the taskbar. But it can improve the day-to-day experience of AI-powered features that run locally. Video‑call backgrounds, real‑time transcription, image generation in Photos or Paint, and intelligent search all stand to benefit from a well-tuned execution provider that reliably routes models to the NPU.

For IT administrators, the update introduces a compliance checkpoint. Since Windows 11 26H1 will be deployed mainly on new Copilot+ hardware, admins should add KB5089618 to their device validation checklist. If a Snapdragon laptop is being tested with a line-of-business AI app, making sure the QNN provider is current eliminates one variable. Also, because KB5089618 supersedes KB5078978, any internal documentation that referenced the old KB needs updating.

For developers, the implications are broader. The update is a concrete example of Microsoft’s strategy to “platform-ize” AI acceleration. Rather than asking every developer to package the QNN SDK inside their app, Windows handles the execution provider as a serviced component. Developers target ONNX Runtime or Windows ML, and the OS picks the right provider. That shrinks app sizes, simplifies distribution, and ensures that AI models work with the latest silicon and driver combinations. But—and this is crucial—developers must still prepare their models correctly. The QNN HTP backend typically demands quantized models (int8 or int16) and may reject operators that the NPU doesn’t support. No Windows Update can fix a model that wasn’t built for the accelerator.

How to Check If You Have It

Verifying installation takes three taps:
1. Open Settings.
2. Go to Windows Update > Update history.
3. Look for Windows ML Runtime Qualcomm QNN Execution Provider Update (KB5089618).

If it’s there, you’re current. If it’s missing, first confirm you’re actually on Windows 11 26H1. Run winver or check System > About; the OS build should start with “26H1.” If you’re on 26H1 and still don’t see the update, install all pending cumulative updates, reboot, and check again. Managed devices may have update policies that delay component deliveries—admins can verify through Windows Update for Business reports.

A Look Behind the Curtain: The Shift to Componentized AI Servicing

KB5089618 didn’t appear in a vacuum. It’s the latest in a line of AI‑specific component updates, following earlier packages like KB5078978 (the one it replaces) and similar provider updates for Intel’s OpenVINO and AMD’s IPU stacks on other Windows versions. Microsoft is carving out a space in Windows Servicing where execution providers live independently of both the OS feature update and the hardware driver.

This approach has real advantages. It allows Microsoft and Qualcomm to push out a compatibility fix for a popular AI model without waiting for a Patch Tuesday or a full driver drop. It also prevents the fragmentation that would occur if every developer shipped a different version of the QNN runtime. As Copilot+ PCs gain features like Recall, Click to Do, and live video translation—all heavily reliant on the NPU—the ability to update the AI runtime quickly and consistently becomes a reliability imperative.

Your Next Steps

  • Home users: Do nothing. Keep Windows Update running, and KB5089618 will install silently when your device is eligible. If you’re curious, check Update history.
  • IT administrators: Add KB5089618 to your compliance reports for 26H1 Snapdragon devices. Note the dependency on the latest cumulative update; ensure servicing pipelines deliver that first. Replace any references to KB5078978 in your knowledge base.
  • Developers: Use the update as a signal to retest your AI workloads. If your app relies on Windows ML, KB5089618 ensures you’re working with the newest QNN stack. If you’re calling ONNX Runtime directly from a custom Python environment or a bundled runtime, remember that the Windows‑serviced provider is separate from any onnxruntime-qnn wheel you pip‑install. Keep your models quantized and validate them with the QNN toolchain to avoid surprises.

What Happens If You Don’t Install It

Because KB5089618 replaces an older version, skipping it means you stay on a QNN provider that is no longer the latest. That won’t break apps immediately, but over time you might miss out on reliability fixes or model‑compatibility updates that newer applications expect. There’s no security vulnerability listed, so the risk is primarily functional. For most users on 26H1, the update will be applied automatically anyway; only aggressively deferred update policies could block it.

The Bigger Picture

KB5089618 is small. But it represents the operating system firmly inserting itself into the AI acceleration stack in a way that benefits everyone. Users get better AI experiences without thinking about drivers. Admins get a clean servicing story. Developers get a platform they can target instead of a puzzle of vendor SDKs. As local AI becomes a feature that sells PCs, expect more updates like this—for more hardware, on more Windows versions—to stream through Windows Update, keeping the AI edge sharp and automatic.