UMVA has learned that a hidden shift inside Windows is quietly reshaping Microsoft’s AI strategy, opening the door for dedicated GPUs to power Copilot+ features.
The experimental Windows App SDK 2.2, now posted on GitHub, lets select AI functions—text summarization, rewriting, code generation, and more—run on Nvidia RTX graphics cards instead of relying solely on an integrated NPU.
This change could also unleash the Photos app’s upscaling and Super‑Resolution tools on machines with modest‑size GPUs, while enabling AI‑driven “erase” and object‑extraction capabilities.
For over two years, enthusiasts have known that GPUs deliver the highest AI throughput in PCs, even though NPUs remain the most power‑efficient engine for tasks like Windows Studio Effects or Paint’s Co‑Creator.
Historically, Copilot+ devices prioritized the NPU and largely ignored the GPU, but the new SDK signals a decisive pivot toward a more flexible, hardware‑agnostic approach.
To test the feature today, users must download the WinAppSDK 2.2 Experimental 9 package, pair it with an Nvidia GeForce RTX 30‑series card sporting at least 6 GB of VRAM, run a Windows Insider build, and enable Developer Mode.
While the setup is a bit of a hurdle, the real story lies in what it foretells: Microsoft is dismantling its monolithic “NPU‑only” AI vision.
In a recent interview at the Build conference, Microsoft’s Surface vice‑president hinted that the company will run local AI models where they make sense and fall back to the cloud for other workloads.
That philosophy aligns perfectly with the experimental SDK, which aims to spread AI capabilities across a wider spectrum of Windows 11 devices.
Millions of older PCs lack an NPU but possess capable GPUs, making this move a logical step that could soon benefit a vast user base, even if broader rollout takes a few months.
