Microsoft is reportedly experimenting with a broader deployment strategy for Windows AI features by enabling certain local AI workloads to run on discrete graphics cards rather than relying exclusively on dedicated neural processing units. The feature was discovered in an experimental version of the Windows App SDK and currently requires a Windows Insider Experimental Channel build along with Developer Mode enabled. Early support appears focused on NVIDIA GeForce RTX 30-series and newer graphics cards equipped with at least 6 GB of video memory. The implementation centers around Microsoft's Language Model APIs, which provide developers with access to local AI inferencing capabilities. These APIs can be used for tasks such as text summarization, content rewriting, editing assistance, and other language-based operations. Previously, Microsoft's AI roadmap heavily emphasized Copilot+ PCs equipped with dedicated NPUs, creating a relatively narrow hardware requirement for many local AI experiences.
By extending support to discrete GPUs, Microsoft could dramatically increase the number of Windows systems capable of running local AI workloads. GeForce RTX graphics cards already include Tensor Core hardware designed for accelerating AI inference, making them well suited for this type of processing. The move effectively allows developers to leverage existing gaming and workstation hardware without requiring users to purchase new NPU-equipped devices.
The experimental feature should not be interpreted as an immediate removal of all Copilot+ hardware requirements. Current information suggests that the change primarily targets developer-facing APIs rather than flagship consumer features such as Recall. Nevertheless, the development indicates that Microsoft is exploring a more flexible approach to AI acceleration across the Windows ecosystem.
The broader strategy aligns with Microsoft's recent investments in Windows ML. The company has been developing infrastructure capable of utilizing both GPUs and NPUs through dedicated execution providers, enabling applications to take advantage of available hardware acceleration while maintaining a consistent development platform. This approach reduces fragmentation and simplifies AI application deployment across different classes of Windows hardware.
For PC enthusiasts, the change could prove significant. Many RTX-equipped systems already possess substantial AI compute resources, and Microsoft's latest testing suggests those resources may become increasingly relevant for future Windows AI capabilities. While the feature remains experimental and may evolve before any public rollout, it demonstrates a potential shift away from strict NPU dependency and toward a more inclusive AI hardware model.
Source: Windows Latest