Microsoft Expands Windows AI Features to Nvidia RTX GPUs, Moving Beyond NPUs

Microsoft has begun testing an expansion of its Windows AI features to include devices equipped with dedicated Nvidia RTX graphics cards, moving beyond its previous reliance on specialized Neural Processing Units (NPUs). This shift, detailed in an experimental release of the Windows App SDK, suggests a technical pivot in how the company delivers local AI capabilities to Windows 11 users, potentially broadening the footprint of features previously exclusive to Copilot+ branded hardware.

The experimental Windows App SDK 2.2 allows specific AI-driven tools—including text summarization, content rewriting, and code generation—to leverage the processing power of Nvidia RTX GPUs. Previously, Microsoft’s Copilot+ initiative was centered on the NPU as the primary engine for on-device AI, a strategy that restricted access to these features to a specific subset of newer, NPU-equipped laptops. By enabling GPU-accelerated AI, Microsoft is effectively utilizing the high-performance compute resources already present in millions of existing Windows PCs.

Expanding AI Beyond the NPU

The core of this development lies in the Windows App SDK 2.2 Experimental 9, which provides the necessary framework for developers to offload AI tasks to the GPU. While NPUs are designed for power efficiency, particularly for background tasks like Windows Studio Effects, GPUs generally offer significantly higher raw AI performance, often measured in TOPS (trillions of operations per second). According to technical documentation regarding the current Windows App SDK, this experimental support targets Nvidia GeForce RTX 30 series GPUs or newer, provided the system has at least 6GB of VRAM.

This approach addresses a long-standing technical reality: many consumer PCs already possess the hardware necessary to run sophisticated AI models, even if they lack the specific NPU architecture Microsoft initially championed for its Copilot+ branding. By enabling these features on GPUs, Microsoft is signaling a more pragmatic approach to AI deployment, prioritizing the availability of tools across a wider range of hardware configurations rather than maintaining a strict, monolithic requirement for NPU-only silicon.

Hardware Requirements for Current Testing

For users interested in testing these capabilities, the barrier to entry remains high, consistent with Microsoft’s standard testing protocols for unreleased features. Participation requires a PC running a Windows Insider build, an Nvidia GeForce RTX 30 series GPU with at least 6GB of VRAM, and the manual activation of Developer Mode within Windows settings. These prerequisites ensure that the experimental SDK is tested primarily by developers and power users capable of managing potential system instability.

The integration of GPU support is expected to impact features within the Microsoft Photos app, such as Super Resolution upscaling and AI-powered object extraction. While these features were originally optimized for NPU hardware, the move to include GPUs suggests that the software architecture is becoming more modular, allowing the operating system to dynamically assign tasks to the most appropriate processor available on the motherboard.

Strategic Shifts in Microsoft’s AI Roadmap

This technical shift aligns with broader discussions regarding Microsoft’s AI strategy. During the recent Microsoft Build conference, company leadership indicated an increasing openness to a hybrid model for AI, where local models run on-device when hardware allows, while more intensive tasks are offloaded to the cloud. This strategy reflects a departure from the “NPU or bust” narrative that accompanied the initial launch of the Copilot+ PC category.

By leveraging the GPU, Microsoft is effectively democratizing access to its latest AI tools for users who have invested in high-end gaming or workstation hardware. For the millions of users with robust GPUs but older CPU architectures, this development represents a significant increase in the utility of their existing machines. While there is no official timeline for a general release of these features outside of the experimental SDK, the current trajectory suggests that the company is moving toward a more inclusive approach to AI hardware compatibility.

As Microsoft continues to refine its Windows App SDK, further updates are expected to be released through the official Windows Developer channels. Users interested in tracking the progress of these features can monitor the official Microsoft Windows App SDK repository on GitHub for future experimental builds and release notes. We encourage readers to share their experiences with these new AI features in the comments section below as they become more widely available.