Copilot+ PC: Every PC with a Dedicated GPU Will Be Included, Even Without a Dedicated NPU

Microsoft is starting to expand access to AI features executed locally on Windows 11, reducing one of the requirements that had so far characterized Copilot+ PCs. An update to the company’s development platform allows the use of local language model APIs even on systems equipped with NVIDIA GeForce RTX 30 Series GPUs or later, as long as they have at least 6 GB of VRAM.

This development emerged within the experimental version of Windows App SDK 2.2, which introduces support for AI processing via dedicated GPUs and not exclusively through NPUs (Neural Processing Units). For the moment, this feature is aimed at developers and is available in an experimental environment, but the signal sent by Microsoft is quite clear.

When Copilot+ PCs debuted in 2024, Microsoft had indicated the presence of an NPU as a fundamental element for accessing the AI capabilities integrated into the operating system. Apart from the minimum memory and storage requirements, the NPU was the distinguishing factor of this new category of computers. However, over the months, it has become clear that modern GPUs also possess considerable processing capabilities for machine learning workloads. Although NPUs guarantee greater energy efficiency, GPUs offer high parallel computing power and have been used for years in executing AI models.

Thanks to this update, developers can use the Windows.AI.Text API even on PCs not belonging to the Copilot+ family, provided they use compatible hardware. The available functionalities include text summarization, rewriting and editing content, converting text into structured formats, and generating prompts—tasks that until now were closely associated with systems equipped with an NPU.

At the center of this implementation is Phi Silica, the AI model optimized by Microsoft for execution directly on the device. The model can be downloaded via Windows Update when requested by a compatible application and leverages the available local hardware, including supported GPUs. Processing therefore occurs directly on the computer, with reduced reliance on cloud services. This allows for decreased data transfer to external servers, improved application responsiveness, and offers greater assurances regarding information management—an aspect particularly relevant for businesses and developers.

For now, accessing these capabilities requires installing the experimental release of WinAppSDK 2.2 Experimental 9, the Insider build of Windows 11, and activating developer mode. Support remains confined to a testing and development context.

Furthermore, the opening to GPUs does not yet involve all AI features introduced with Copilot+ PCs. Tools like Windows Recall and Click to Do continue to require the presence of a dedicated NPU. The extension currently pertains to the level of the API for local language models and not the entire AI ecosystem of Windows.

Despite these limitations, the change indicates a new direction for Microsoft. The company no longer sees NPUs as the only way to bring local AI to Windows 11, thus broadening the range of compatible systems and allowing millions of PCs equipped with modern NVIDIA RTX GPUs to access a growing portion of AI-based functionalities.