Microsoft is bringing DeepSeek to Windows Copilot+ PCs, here’s how

The DeepSeek-R1-Distill-Qwen-1.5B model will be the first to arrive on Copilot+ PCs, optimised for NPUs (neural processing units). Microsoft plans to release larger 7B and 14B versions later, enabling developers to build AI-powered applications that operate efficiently on local hardware. Copilot+ PCs require a minimum of 16GB RAM, 256GB storage, and NPUs with at least 40 TOPS (trillions of operations per second).

Microsoft stated that the models are designed to utilise low-bit processing, ensuring they run efficiently on-device without relying heavily on cloud-based computing.

DeepSeek AI expands to Microsoft Azure AI Foundry

In addition to integrating DeepSeek into Copilot+ PCs, Microsoft is also adding the R1 model to its Azure AI Foundry platform. This move places DeepSeek alongside AI models from OpenAI, Mistral AI, and Meta.

Despite the integration, Microsoft is reportedly investigating whether DeepSeek used OpenAI’s technology without authorization. OpenAI has previously accused several Chinese AI startups, including DeepSeek, of distilling its models.

Security concerns
Privacy and security concerns surrounding China-based DeepSeek have been raised by AI researchers. Microsoft addressed these concerns in a community post, stating that DeepSeek R1 underwent extensive security evaluations, including automated model behavior assessments and risk mitigation procedures.

DeepSeek’s advancements in AI efficiency have disrupted the market, leading to a selloff in AI-related stocks, including NVIDIA. The company’s R1 model requires less computational power than competitors, raising questions about its methods and impact on the AI landscape.