Microsoft Unveils DeepSeek R1 Models for Copilot+ PCs on Windows 11

Microsoft has announced that the DeepSeek R1 models will be integrated into Copilot+ computers running Windows 11, promising to enhance and streamline the user experience.
Microsoft and DeepSeek R1: AI Powers Windows 11
Technology titan Microsoft has unveiled plans to incorporate DeepSeek R1, a widely-used artificial intelligence model, into its Windows 11 operating system. The goal? To enable developers to create AI-enhanced applications that operate seamlessly on devices.
Phased Rollout
The DeepSeek R1 models will initially debut on PCs equipped with Qualcomm Snapdragon X processors. Availability will expand to machines with Intel Core Ultra 200V (Lunar Lake) processors, and eventually to other hardware configurations. The DeepSeek R1 7B and 14B variants will be accessible through Microsoft’s AI Toolkit.
Optimization for the NPU
Building on the experience with Phi Silica, a smaller language model for Windows 11, Microsoft has fine-tuned the DeepSeek models for the Neural Processing Unit (NPU). “We’re leveraging key lessons and techniques from this endeavor, including how we segment different model components to strike an optimal balance between performance and efficiency,” Microsoft emphasizes. The company continues to leverage the Windows Copilot Runtime (WCR) to adapt to the entire Windows ecosystem using the ONNX QDQ format.
Specific Hardware Requirements
To take advantage of these AI-optimized applications, Windows 11 devices need to have a Neural Processing Unit (NPU) with at least 40 TOPS (trillions of operations per second), 16 GB of DDR5 RAM, and 256 GB of storage space. These specifications are already standard for many PCs equipped with Snapdragon X Elite, X Plus, and the newly announced Snapdragon X.
Despite concerns about massive data collection and plagiarism allegations against DeepSeek R1, Microsoft appears committed to maximizing the potential of this AI model on its Copilot+ PCs. It remains to be seen how the company will implement this AI model across its devices.