Microsoft has quietly delivered KB5089167, a machine‑learning runtime update for Windows 11 that sharpens AI acceleration on supported AMD graphics hardware. The update, version 2.2604.1.0, installs automatically through Windows Update and targets PCs running Windows 11 24H2 or the upcoming 25H2.

Inside the Update: What KB5089167 Actually Changes

KB5089167 is not a cumulative update, a security patch, or a graphics driver. It is a component package for the AMD MIGraphX Execution Provider—the bridge between Windows’ built‑in ONNX Runtime / Windows ML inferencing engine and AMD’s MIGraphX graph compiler. Its job is to let Windows‑managed AI workloads hand off supported model operations to an AMD GPU for hardware acceleration.

According to Microsoft’s support article, the update raises the MIGraphX Execution Provider AI component to version 2.2604.1.0 and supersedes the earlier KB5083461. It requires the latest cumulative update for Windows 11, version 24H2 or 25H2, before it will appear in Windows Update. Once installed, you’ll see it under Settings > Windows Update > Update history listed as:

Windows ML Runtime AMD MIGraphX Execution Provider (KB5089167)

There is no new app, control panel, or Start menu entry. The component works silently in the background, called upon by applications that use Windows ML or ONNX Runtime with the AMD MIGraphX provider enabled.

Who Actually Benefits – and How

For everyday users

If you don’t run local AI workloads—tools that classify images, transcribe speech, generate embeddings, or upscale media using ONNX models—you may never notice this update. It installs automatically and stays out of the way. If you do use such apps, however, KB5089167 may improve compatibility or performance when the app leverages your AMD GPU for inference.

For power users and enthusiasts

Check your update history on a machine with a supported AMD GPU. The entry confirms the MIGraphX provider is current. If you compare AI inference times before and after the update, bear in mind that the model, driver version, app settings, and power plan all influence results. This update only refreshes one piece of the pipeline.

For developers

Anyone working with ONNX Runtime on Windows can now rely on a system‑managed MIGraphX execution provider at version 2.2604.1.0. When you create an inference session and select DML, CPU, or other providers, the updated MIGraphX provider becomes available automatically—provided the host meets the prerequisites. It’s wise to log the KB5089167 state, driver version, and ONNX Runtime version when profiling model performance on AMD hardware.

For IT administrators and support teams

KB5089167 is part of the Windows AI component servicing stream. It will deploy to managed Windows 11 24H2/25H2 devices with AMD GPUs through Windows Update for Business or WSUS if you allow “Other Microsoft products” updates. Include it in test matrices if your organisation validates AI‑enabled applications that use Windows ML or ONNX Runtime. The update history name (“Windows ML Runtime …”) often raises questions; reassure users it’s a legitimate Microsoft component, not malware or an unwanted app.

How We Got Here: The Rise of System‑Managed AI

Modern Windows AI workloads are layered. An app might use a model trained in PyTorch or TensorFlow, export it to ONNX format, and then run it locally via Windows ML or ONNX Runtime. Instead of forcing every app to bundle its own hardware‑specific acceleration stack, Microsoft has been moving toward a shared platform model. Windows ML, powered by ONNX Runtime, can acquire execution providers—for CPUs, GPUs, and NPUs—as separate, serviceable components.

AMD’s MIGraphX is a graph inference engine that compiles and optimizes trained models for AMD hardware. The MIGraphX Execution Provider acts as the ONNX Runtime plugin that translates the model graph into operations MIGraphX can accelerate. Other providers exist for NVIDIA CUDA, Intel OpenVINO, and Qualcomm NPUs, but on Windows the DirectML provider has been the primary GPU‑agnostic path. The MIGraphX provider, however, is tuned specifically for AMD GPUs and can offer better performance than the generic DirectML path for certain models—when the driver and model operators align.

KB5089167 follows KB5083461, which introduced an earlier version of the provider. This pattern mirrors the way Microsoft services other platform components—like the Linux kernel in WSL or the web experience pack—independently of full OS feature releases. By decoupling AI runtime updates from cumulative updates, Microsoft can push improvements faster without waiting for Patch Tuesday.

The prerequisite of the latest cumulative update matters because those packages often include servicing stack changes, updated Windows ML components, or ONNX Runtime host updates that the execution provider depends on. Without them, KB5089167 won’t show up.

What to Do Right Now

  1. Keep Windows Update current. If your device meets the requirements, KB5089167 will appear automatically. No manual download is needed.
  2. Check update history. Navigate to Settings > Windows Update > Update history and look for the entry. If it’s missing, install any pending cumulative updates, reboot, and scan again.
  3. Verify your AMD GPU driver. KB5089167 is not a driver, but having the latest AMD Radeon Software is still important for overall compatibility. Check AMD’s website or the Radeon Settings app for updates.
  4. Test your AI app. If you rely on a specific application that uses ONNX Runtime or Windows ML, run a typical workload after the update and confirm nothing breaks. If you hit an issue, start by narrowing down whether the app actually uses the MIGraphX provider. Many apps default to CPU or DirectML; you may need to check the app’s documentation or configuration.
  5. Don’t uninstall without a reason. Removing AI platform components can cause more problems than it solves, especially as more apps begin depending on shared runtimes. Only consider uninstalling if you can reproduce a specific failure and after isolating KB5089167 as the trigger.

Outlook: More Pieces in the AI Acceleration Puzzle

KB5089167 is a small but telling piece of Windows’ AI strategy. As local inference becomes central to apps—from Recall in Windows 11 to third‑party tools like Stable Diffusion and Whisper—keeping hardware‑specific acceleration current through Windows Update reduces friction for developers and users alike. Expect more execution‑provider updates as the ecosystem evolves, alongside tighter integration with NPUs and hybrid AI engines. For AMD GPU owners, this silent update ensures their hardware stays ready for the next wave of on‑device intelligence.