China's DeepSeek, whose AI model sent markets tumbling, is now developing its own chips — Reuters

Chinese startup DeepSeek is developing its own AI chip, three sources familiar with the matter told Reuters. According to the news agency, this move could reduce the AI company’s reliance on chips from Nvidia and Huawei, which it uses to train and run its artificial intelligence models.

According to sources, the chip is being developed for inference—the stage of computation in which a pre-trained model generates responses for users—rather than for training new AI models. Work on this project began about a year ago, one of Reuters’ sources clarified. However, according to the agency, development is still in the early stages: the company is in talks with external partners in the fields of chip design, contract manufacturing, and memory production, and is also hiring semiconductor engineers.

DeepSeek did not promptly respond to a request for comment from Reuters.

The agency notes that developing a competitive processor typically takes years and requires massive investments.

In trading on July 7, Nvidia shares fell 1.72%; year-to-date, they are up 3%.

If successful, DeepSeek’s expansion into chip development would mark a major strategic shift for the company, which is already considered China’s leading AI firm, Reuters notes. DeepSeek, for example, will be able to address issues related to export restrictions that prohibit Chinese companies from purchasing Nvidia’s most advanced chips, among other things.

The situation could also potentially exacerbate the problems faced by Chinese tech giant Huawei, the agency notes. Although Huawei’s offerings lag significantly behind the cutting-edge developments of the American company Nvidia, the U.S. ban on their export to China has helped Huawei capture about half of the domestic—Chinese—AI chip market, valued at $50 billion, Reuters explains. For example, DeepSeek and a number of other leading Chinese players in the industry, among others, supply Huawei’s products. However, the tech giant’s position is gradually weakening as its competitors, Alibaba and Baidu, develop their own AI chips and gain market share.

DeepSeek gained worldwide recognition over a year ago after releasing two AI models in late 2024–early 2025: the free V3, as well as R1—an open-source neural network that mimics the human thought process. The Chinese company’s developments performed on par with American market leaders in tests. In addition, the Chinese AI startup reported at the time that developing R1 cost the company several times less than its American counterparts (less than $6 million) and took only two months. By comparison, training similar models from OpenAI or Google, according to some reports, cost between $100 million and several billion dollars at the time. Chinese AI models caused a sensation in the market at the time and triggered a trillion-dollar stock sell-off in the U.S.: DeepSeek’s results called into question the wisdom of Western corporations’ significant spending on AI infrastructure.

This article was AI-translated and verified by a human editor