Microsoft is reportedly testing Copilot+ AI features with discrete GPUs instead of NPUs

Microsoft is reportedly testing a new feature that allows Copilot+ AI capabilities to run on discrete GPUs instead of dedicated NPUs, marking a strategic shift from its previous NPU-centric approach. ... Read original →

Microsoft’s pivot to discrete GPUs for Copilot+ marks a tactical retreat from its NPU-centric doctrine. Technically, this empowers NVIDIA to embed RTX GPUs into the local AI inference stack via Windows App SDK, eroding Qualcomm’s NPU leverage and redirecting advanced-node investments toward general-purpose compute. Compliance-wise, while sidestepping NPU supply chain risks—especially those tied to Taiwan, China—it heightens reliance on GDDR6/7 memory amid soaring prices, pressuring OEM margins and likely triggering OS-level resource throttling. Strategically, this move counters Apple’s vertical integration by lowering AI adoption barriers to stem Windows-to-macOS migration; Linux may exploit open-source inference frameworks to leapfrog. Over the next 12–24 months, if Copilot+ scales across the RTX 30-series installed base, it will redefine 'AI PC'—yet expose Windows’ weak heterogeneous scheduling, forcing a runtime architecture overhaul.