The Chinese start-up DeepSeek wants to follow in the footsteps of the largest artificial intelligence firms and become independent of hardware suppliers. According to Reuters, the company is working on its own AI chip designed for inference – the stage at which models generate responses for users.
The project is still at an early stage. DeepSeek is in talks with firms specialising in chip design, manufacturing and memory, and is quietly increasing its recruitment of engineers specialising in semiconductors. The company has not confirmed this information.
If the project proves successful, it will mark a significant shift in the company’s strategy. Until now, DeepSeek has focused primarily on developing high-performance AI models using chips from Nvidia and Huawei. Having its own processor would allow the company to better control costs and develop the infrastructure needed to support a growing number of users.
The chip is to be designed with inference in mind, rather than model training. This is no coincidence. As services based on generative AI become more widespread, an increasing proportion of the demand for computing power relates specifically to running pre-trained models. In this segment, it is not only performance that matters, but also operating costs and energy efficiency.
DeepSeek is therefore part of a global trend. OpenAI, in collaboration with Broadcom, is already developing its own chips, and Anthropic is also exploring similar plans. The aim is to reduce dependence on Nvidia, which remains the dominant supplier of AI accelerators.
For the Chinese firm, however, the motivation is broader. US export restrictions make it difficult for Chinese companies to access state-of-the-art chips, which is why Beijing has been supporting the development of a domestic semiconductor ecosystem for several years. DeepSeek, which previously used both Nvidia H800 chips and Huawei Ascend processors, is seeking to reduce its reliance on both suppliers.
This does not, however, mean that the project will be successful. Designing a competitive AI processor requires many years of work, significant financial investment and access to advanced manufacturing capacity and memory, all of which remain limited. Therefore, whilst an in-house chip may increase DeepSeek’s independence, it will not solve the industry’s biggest problem: access to the global supply chain for the most advanced semiconductors.
