DeepSeek is reportedly recruiting talent to develop its own AI chips due to difficulties in accessing Western products.
Digitimes cited sources in the semiconductor industry in China as saying that DeepSeek has launched a major recruitment campaign to look for semiconductor design talent. The company is specifically targeting experts in high-performance computing (HPC) and algorithm optimization to accelerate both technology development and commercialization efforts.
DeepSeek has not officially announced plans to develop chips for edge or cloud applications, but the hiring move is a clear indication of the company's ambitions in the AI chip industry, according to sources.
"The global competition for large language models (LLMs) is getting fiercer, and machine autonomy has become a necessity. This puts DeepSeek in a similar position to OpenAI in developing its own processors," one source said.
Another source said DeepSeek's current strategy is to use low-end, low-end GPUs and domestic computing processors, such as Huawei Ascend, to maximize AI training capabilities. Recently, 16 Chinese AI chip companies, including Huawei, Moore Threads, and Hygon, have announced their own chip models that are compatible or integrated with the DeepSeek service, ensuring access from cloud providers such as Huawei Cloud, Tianyi Cloud, and Tencent Cloud.
In addition to domestic chips, DeepSeek still uses GPUs purchased from Nvidia. SemiAnalysis previously reported that the Chinese company may have access to about 10,000 Nvidia H800 GPUs that have reduced performance to comply with the US ban and about 10,000 H100 GPUs. Speaking to CNBC, Scale AI CEO Alexandr Wang also revealed that he has information that DeepSeek owns 50,000 Nvidia H100 chips.
"Domestic chips currently cannot help DeepSeek keep up with the world, especially in large-scale training and inference applications," a source said. "Therefore, DeepSeek's initiative to develop its own chips can help reduce external risks, improve the performance and power consumption of the company's AI models, and create a differentiated solution that fits specific requirements."
However, DeepSeek is also facing many challenges in developing its own chips, including the need to establish a solid foundation in technology R&D and meet international standards for the development, design and manufacturing of core technologies. Therefore, the company's long-term strategy for self-developed chips remains uncertain.
DeepSeek did not immediately respond to a request for comment.
Developing custom chips in-house is a trend in the semiconductor industry to reduce dependence on Nvidia - a company that holds more than 80% of the AI chip market share. Most recently, according to Reuters, OpenAI is collaborating with Broadcom and TSMC to develop the first self-designed inference chip, expected to be released in 2026.
DeepSeek was founded in 2023 by Liang Wenfeng, co-founder of AI investment fund High Flyer. The company began developing DeepSeek in April 2023, before releasing its R1 inference model on January 20.