Microsoft has released KB5089169, a background update that refreshes the AMD Vitis AI Execution Provider to version 2.2604.1.0 on Windows 11. The component update, delivered automatically to devices running version 24H2 or 25H2, is designed to improve how applications tap into the neural processing unit (NPU) inside recent AMD Ryzen AI processors. While it won’t add a flashy new app or a settings toggle, this refresh is a critical servicing moment for anyone using an AMD Ryzen AI-powered PC.

At its core, KB5089169 replaces an earlier provider update (KB5079258) and is part of Microsoft’s push to treat local AI acceleration as a managed platform service, not something each application has to bundle and maintain on its own.

What Actually Changed – and How It Works

Unlike a feature update that changes how Windows looks or behaves, this is a pure component refresh. Microsoft’s support page doesn’t publish a detailed changelog; instead, it frames KB5089169 as an update that “includes improvements to the AMD Vitis AI Execution Provider AI component.” In practice, that means the bits behind ONNX Runtime and Windows Machine Learning (Windows ML) on AMD hardware have been retooled.

The Vitis AI Execution Provider is a bridge: when an app asks the system to run an AI model, Windows ML and ONNX Runtime can route supported operations to the AMD NPU rather than piling everything onto the CPU or GPU. Think of it as a translator. Without a current provider, the NPU sits idle, and all the heavy lifting falls to the CPU or GPU. With KB5089169, that translation layer gets a refresh.

Specifically, the update bumps the provider to version 2.2604.1.0 and is available only for Windows 11 24H2 and 25H2 editions. It requires the latest cumulative update to be installed first; Microsoft says the component will download and install automatically through Windows Update. There is no manual installer in the Microsoft Update Catalog. Once installed, it appears in Settings > Windows Update > Update history as “Windows Runtime ML AMD NPU Execution Provider Update (KB5089169).” The proper title might be “AMD Vitis AI Execution Provider update version 2.2604.1.0,” but both refer to the same package.

According to ONNX Runtime documentation, execution providers like Vitis AI are what allow the runtime to work with different hardware acceleration libraries. The CPU provider acts as a fallback, while hardware-specific providers accelerate compatible portions of a model on GPUs, NPUs, or other accelerators. By updating this provider, Microsoft and AMD are ensuring the latest optimizations and compatibility fixes are present for AI workloads.

What It Means for You

The impact of KB5089169 depends on your role. Here’s how it breaks down.

If you’re an everyday user with an AMD Ryzen AI PC: You likely won’t notice anything different. This is a behind‑the‑scenes tweak. However, if you use apps that lean on local AI—such as Windows Studio Effects, certain video‑conferencing enhancements, or image‑generation tools optimized for Windows ML—you might see subtle improvements in responsiveness or efficiency. No promises, because AI acceleration depends on many factors, but keeping the execution provider current is step one. Do not expect a new performance dashboard; the real gains, if any, will show up in the apps that already know how to use the NPU.

For developers building with ONNX Runtime and Windows ML: This update is important. If you target AMD Ryzen AI hardware, KB5089169 changes the floor underneath your application. The Vitis AI Execution Provider is what ONNX Runtime uses to compile and run models on the NPU. With version 2.2604.1.0, you should:

  • Retest your ONNX models. The provider update could affect first‑run compilation times, operator support, and performance characteristics.
  • Re‑examine cache directories. According to AMD’s Ryzen AI documentation, compiled NPU artifacts should not be reused across different provider versions. If your app does its own cache management, invalidate or version‑tag the cache when the provider changes.
  • Verify fallback behavior. The Vitis AI EP automatically partitions a model graph, sending supported subgraphs to the NPU and the rest to the CPU. After this update, the partitioning might behave slightly differently. Check that your hardware‑acceleration path is still active.
  • Double‑check model quantization. The provider is optimized for INT8 and BF16 formats. If you were already using those, performance should be consistent; if not, now is a good moment to quantize.

Microsoft’s Windows ML documentation encourages developers to let the system dynamically acquire the latest execution providers, avoiding the need to bundle every vendor SDK inside an app. KB5089169 is a live example of that model.

For IT administrators managing fleets: On managed Windows 11 24H2 or 25H2 devices, this is yet another component update. It will arrive via Windows Update automatically, but only if the devices have the latest cumulative update. If your patching policies delay cumulative updates, this provider update will also be delayed. Action items:

  • Verify that AMD Ryzen AI devices in your fleet have received KB5089169 (look for the “Windows Runtime ML AMD NPU” entry).
  • Note that KB5089169 supersedes KB5079258; update any internal inventory records.
  • If you support developers or power users who depend on local AI, consider adding the provider version to your hardware baseline documentation.
  • Check Windows Update policies to ensure these component updates aren’t blocked unintentionally.

How We Got Here: From Siloed AI to Managed Platform

For years, AI on Windows was fragmented. Every app that wanted hardware acceleration had to ship its own runtime, its own model, and its own driver‑level optimizations. That duplicated effort and made consistent updates a nightmare. Microsoft’s answer is Windows ML, a platform‑level AI inferencing framework that uses ONNX Runtime under the hood. Execution providers plug into ONNX Runtime to target specific hardware: Intel OpenVINO, NVIDIA TensorRT, Qualcomm QNN, and AMD Vitis AI, among others.

The shift began in earnest with Windows 10, but it’s Windows 11—and the spread of dedicated NPUs in laptops and desktops—that turned execution providers into something that must be serviced system‑wide. AMD’s Ryzen AI, Intel’s Core Ultra series, and Qualcomm’s Snapdragon X each brought their own NPU silicon, and each required its own execution provider to unlock local AI acceleration. In 2024, Microsoft started releasing out‑of‑band component updates like KB5079258 for the Vitis AI provider, signaling that these were no longer static drivers but dynamic pieces of the AI stack.

KB5089169 continues that trend. It replaces KB5079258, meaning it is now the active provider version for supported systems. The fact that it requires the latest cumulative update is a reminder that staying current on Patch Tuesday opens the door to these AI enhancements. If you defer monthly updates by weeks, you might be running a stale execution provider that doesn’t match the latest ONNX Runtime improvements delivered through those same cumulative updates.

What to Do Now

If you own an AMD Ryzen AI PC:

  1. Open Settings > Windows Update and click Check for updates.
  2. Install any pending cumulative updates for Windows 11 24H2 or 25H2.
  3. Wait for the automatic component update to arrive—no manual download needed.
  4. Once installed, confirm it appears under Update history as “Windows Runtime ML AMD NPU Execution Provider Update (KB5089169).”
  5. Restart your PC if required (though typically, component updates don’t demand a reboot).

If you’re a developer:

  1. Update your test machines to the latest cumulative update and confirm KB5089169 is present.
  2. Reproduce your ONNX Runtime inference sessions with VitisAIExecutionProvider and check performance metrics.
  3. Purge any on‑disk compiled model caches that were generated with the old provider version.
  4. Review AMD’s Ryzen AI documentation for any new notes on supported operators, model compilation flags, or quantization guidance.
  5. Consider adding a runtime check for the provider version; if it’s 2.2604.1.0 or higher, you know you’re on the updated path.

If you’re an IT pro:

  1. Monitor Windows Update compliance reports for KB5089169 on eligible devices.
  2. If devices aren’t receiving it, verify that they have the latest cumulative update installed and that no policy is blocking “other Microsoft products” updates (though this is a component, not a traditional driver, it should flow with the regular servicing channel).
  3. Update your software asset management to note the replacement of KB5079258.

What’s Next

This isn’t the last you’ll see of NPU‑related component updates. As Qualcomm Snapdragon X, Intel Core Ultra, and AMD Ryzen AI chips push more AI tasks onto the NPU, Microsoft will continue to service execution providers through Windows Update. The long‑term goal: make local AI acceleration as boring and reliable as printing.

Microsoft is also building a “Windows Copilot Runtime” that will further abstract AI hardware, but execution providers will remain the low‑level workhorses. Keeping them updated will be essential for both performance and security. For now, KB5089169 is a small but tangible step toward that boring ideal. If you’re on a Ryzen AI machine, let the update install and go about your day. If you’re building AI software for Windows, take an hour to re‑validate your models. The invisible platform shift is underway, and this time it comes with a KB number.