DeepSeek begins in-house AI chip development for inference

Chinese AI startup DeepSeek has initiated an in-house artificial intelligence chip development project, focusing specifically on inference tasks. The effort, as reported by Reuters citing sources, aims to lower inference costs through custom-designed processors, thereby reducing the company's dependence on overseas suppliers such as NVIDIA.

The project is said to be in its early stages and concentrates on inference rather than model training. As generative AI adoption accelerates, inference has emerged as a rapidly growing segment of the AI market. Unlike training, which requires concentrated bursts of computing power, inference demands continuous service for large volumes of user requests, making cost efficiency, power consumption, and reliability critical.

Sources indicate DeepSeek began the chip development project about a year ago and has recently intensified hiring efforts, recruiting experienced chip engineers for architecture, verification, and software enablement. DeepSeek is recognized as one of China's fast-growing foundation model companies, and with rising demand for its services, computing infrastructure has become a significant operating expense.

Developing custom AI chips is a strategic priority for many leading AI companies globally. For DeepSeek, this initiative could lead to long-term cost reductions and improved deployment efficiency, strengthening its competitive position. However, chip development is a capital-intensive and time-consuming undertaking, typically taking over a year from design to mass production, suggesting any impact on the AI chip market will not be immediate.

DeepSeek is also reportedly seeking its first round of external funding, with estimates suggesting a target of around $7 billion. If successful, investment in chip development and other AI infrastructure is expected to become a key priority.