Microsoft tests AI features on Nvidia GPUs as Copilot+ exclusivity begins to fade

Microsoft Opens AI Features to Discrete GPUs Beyond NPU Requirements

Microsoft is testing a shift in its AI strategy by allowing Microsoft AI features to run on discrete GPUs rather than exclusively on Neural Processing Units. According to updated documentation spotted on GitHub, the Windows App SDK now enables local AI workloads on non-Copilot+ devices equipped with Nvidia GPUs 1

1

. The experimental feature supports Nvidia GeForce RTX 30-series cards and newer models with at least 6GB of VRAM, marking a departure from the company's previous NPU-only approach 2

2

.

The capability requires users to download the experimental Windows App SDK 2.2, run a Windows Insider Experimental Channel build, and enable Developer Mode 3

3

. While these requirements create barriers for casual users, the development signals Microsoft's willingness to broaden access to local AI inferencing beyond the strict hardware requirements that defined Copilot+ PCs since their June 18, 2024 launch.

Language Model APIs Powered by Phi Silica Model

The Windows.AI.Text APIs now enable text-based AI tasks on supported hardware, utilizing Microsoft's compact Phi Silica model that runs entirely on-device 2

2

. Instead of pre-installing the model on all systems, Windows can download it through Windows Update when an application requires it. This approach treats AI models as downloadable Windows components rather than premium features exclusive to specific device categories 5

5

.

Developers can now build applications that leverage text summarization, content rewriting, text-to-table conversion, code generation, and prompt generation capabilities 3

3

. The Windows 11 AI framework processes these tasks locally, reducing reliance on cloud services and keeping sensitive data on the user's machine. This matters for enterprise users and privacy-conscious consumers who prefer on-device processing over external server dependencies.

Why Microsoft Is Reversing Course on Copilot+ Exclusivity

When Copilot+ PCs launched, Microsoft positioned NPUs as essential for local AI experiences, requiring devices to have Neural Processing Units capable of 40 TOPS of AI performance alongside 16GB of RAM and SSD storage 5

5

. However, research from 2024 revealed that consumers weren't purchasing AI PCs specifically for their features—they bought them simply because newer models happened to include AI capabilities 1

1

.

The situation has intensified in 2026 as AI data center-driven shortages of memory and storage chips push computer prices higher, causing PC sales to collapse and threatening the availability of entry-level laptops by 2028 1

1

. Limited adoption of NPU-equipped devices would restrict the reach of Copilot+ AI capabilities on GPUs, potentially allowing Microsoft to differentiate Windows 11 as the platform loses users to macOS and Linux.

GPUs Deliver Power While NPUs Offer Efficiency

The technical reality is that GPUs have always been capable of handling AI workloads—often with more raw processing power than NPUs 2

2

. Modern graphics cards excel at the parallel processing required for machine learning models, which is why AI enthusiasts have relied on them for years to run local language models and image generators . The primary advantage of NPUs lies in energy efficiency rather than performance, making them valuable for battery-powered laptops but less critical for desktop systems or gaming laptops with discrete GPUs 1

1

.

This created an awkward situation where a gaming PC with an RTX 4070 had sufficient horsepower for AI tasks but couldn't access Microsoft's native AI framework due to the absence of an NPU 5

5

. The restriction appeared arbitrary to many users who questioned why Microsoft limited these capabilities when capable hardware already existed in millions of machines.

What Remains Exclusive and What's Next

Not all Copilot+ features are migrating to GPU-based systems. Windows Recall, Click to Do, and certain AI-powered creative tools still remain tied to NPU-equipped devices 2

2

. The current expansion focuses specifically on Language Model APIs for text-based tasks rather than the complete suite of AI integrations 4

4

.

Yet the trajectory appears clear. At Microsoft's recent Build conference, the Copilot+ brand received no mention as the company shifted focus toward promoting AI agents as the next evolution in Windows 11 4

4

. Andrew Hill, Microsoft's corporate vice president of Surface, indicated the company is becoming increasingly flexible about running local AI models where appropriate while relying on cloud processing when needed 3

3

. This pragmatic approach suggests Microsoft will continue expanding AI availability across a broader range of Windows 11 devices, making the distinction between Copilot+ PCs and standard systems increasingly blurred.