Microsoft brings NPU optimized DeepSeek-R1 AI to Copilot+ PCs
Microsoft is set to integrate “NPU-optimized” versions of the DeepSeek-R1 AI model into Copilot+ PCs, starting with Snapdragon X devices. Support for Intel Lunar Lake and AMD Ryzen AI 9 processors will follow. The first release, DeepSeek-R1-Distill-Qwen-1.5B, will be available to developers via the Microsoft AI Toolkit, with larger 7B and 14B models coming later.
Copilot+ PCs must meet specific hardware requirements, including at least 256GB storage, 16GB RAM, and an NPU capable of 40 TOPS. Older NPU-equipped devices may not be compatible.
“These optimized models enable developers to build AI-powered applications that run efficiently on-device, leveraging the powerful NPUs in Copilot+ PCs,” Microsoft stated. The company highlighted its advancements in efficient inferencing through its Phi Silica work, reducing power consumption while maintaining performance. The models also use the Windows Copilot Runtime (WCR) with ONNX QDQ format for scalability across Windows devices.
Microsoft detailed the optimizations that make DeepSeek-R1 models work seamlessly on local hardware. A sliding window design improves speed and extends context support, while the 4-bit QuaRot quantization scheme enhances low-bit processing efficiency.
The 1.5B model will soon be available via the AI Toolkit extension in VS Code, allowing developers to experiment locally. Additionally, Microsoft is making DeepSeek-R1 accessible through Azure AI Foundry, offering a secure and scalable cloud platform for enterprise use.
Meanwhile, OpenAI has accused DeepSeek of using stolen proprietary code to develop its AI, reportedly built for under $10 million—far less than the billions spent by US firms.