DeepSeek advances its own AI inference chip to cut reliance on Nvidia

Chinese AI start‑up DeepSeek is developing a proprietary inference‑only processor aimed at reducing dependence on external suppliers such as Nvidia and Huawei. The chip is intended for running trained models rather than training them, and the company has already begun talks with manufacturers and quietly recruited engineers for the project. This move reflects a broader industry trend: OpenAI has unveiled its Jalapeño inference chip with Broadcom, Anthropic is reportedly working with Samsung, and fellow Chinese firm Zhipu is also exploring a custom chip. Companies cite lower operating costs, greater control over supply chains and, for Chinese firms, a response to U.S. export restrictions that limit access to high‑end Nvidia GPUs. While DeepSeek’s processor is expected to remain within China due to export controls, its development signals an accelerating push for AI hardware self‑sufficiency in the country.