Woofun AI reports that DeepSeek has been secretly developing proprietary AI inference chips for approximately one year. The initiative focuses specifically on model inference rather than training, with the company actively recruiting chip design engineers to accelerate progress. While DeepSeek currently depends on NVIDIA and Huawei hardware, successful deployment aims to decrease external supplier reliance and optimize hardware cost control. The project remains in early stages, facing constraints from U.S. export controls on manufacturing and memory acquisition.