DeepSeek Reportedly Developing Its Own AI Inference Chips

DeepSeek is developing its own data center inference chips, marking the Chinese AI startup’s planned expansion into semiconductor design as it looks to reduce its dependence on external hardware suppliers. According to Reuters, the company has spent about a year working on the effort, meeting with potential hardware partners and hiring engineers to support the project.

The initiative is focused on chips designed for inference rather than AI model training. By targeting the hardware used to serve models after they have been trained, DeepSeek is aiming to gain greater control over the infrastructure required to run its large language models.

Reuters reported that one of the primary goals is to reduce reliance on both Nvidia and Huawei. Nvidia remains the dominant supplier of AI chips for companies across North America and Europe, but U.S. export restrictions have limited the company’s ability to sell its most advanced data center hardware in China. Within the Chinese market, Huawei has emerged as the leading supplier, controlling about half of the country’s data center chip market.

DeepSeek is not the only Chinese technology company pursuing custom AI silicon. Alibaba and Baidu have also launched efforts to develop their own chips as Chinese firms seek alternatives to foreign hardware and greater control over their AI infrastructure.

The move mirrors a broader trend among leading AI developers globally. Earlier, OpenAI and Broadcom introduced Jalapeño, OpenAI’s first inference chip designed for deployment at scale. Anthropic has also been exploring custom chip development, although it has not announced any public milestones.

For companies building increasingly large AI services, custom silicon offers more than an alternative to third-party suppliers. It can reduce dependence on dominant chip vendors while giving developers tighter integration between hardware and software. It also provides greater flexibility as demand for AI computing infrastructure continues to grow and competition for data center capacity remains intense.

DeepSeek has not disclosed additional details about its chip program, including manufacturing partners, development milestones, or a timeline for deployment. Reuters reported that the project remains in development after roughly a year of work.

This analysis is based on reporting from ars technica.

Image courtesy of newsimg.koreatimes.co.kr.

This article was generated with AI assistance and reviewed for accuracy and quality.