Chinese artificial intelligence startup DeepSeek has begun developing its own AI chips to break through U.S. export controls. The move is seen as an effort to reduce its chip dependence on Nvidia and, further, to strengthen China's own AI ecosystem.
DeepSeek is in talks with chip design, foundry and memory companies about producing its own chips, Reuters reported Friday. The chips are intended for inference rather than for training new models. Inference is the AI computing stage in which a trained model generates responses for users.
DeepSeek is seeking to develop its own semiconductors because of strong U.S. export control policies. The company has used Nvidia and Huawei chips, and its inference model R1 in particular was trained on the Nvidia H800. The H800 was designed for the Chinese market, but the United States banned its export in late 2023. DeepSeek and other Chinese companies became unable to secure Nvidia's latest chips, and the Chinese government has been urging its technology firms to develop domestic products. Against this backdrop, if DeepSeek succeeds in developing its own chips, China would gain hardware control in addition to its advanced software competitiveness.
Chinese authorities are also moving to protect the AI ecosystem, including by pursuing measures to limit foreign companies' access to domestic AI models. In relation to this, China's Ministry of Commerce recently held several meetings and in-depth discussions with major tech companies such as Alibaba and ByteDance, as well as startups including Zhipu AI. According to a person familiar with the matter, the meetings discussed regulation broadly, including for both closed and open models, but the timing and whether it will be implemented have not yet been decided.
The discussions are read as an extension of China's efforts to keep its domestically produced AI within the country. Like the United States, China is treating advanced AI as a core national asset requiring control. In particular, the administration of U.S. President Donald Trump has been strengthening sanctions, wary of the possibility that AI products could be misused by the military and intelligence agencies of countries of concern such as China and Russia.
Last month, the U.S. administration ordered that foreigners be barred from accessing Anthropic's most advanced models, Fable and Mythos. In response, Anthropic blocked the two models for all users worldwide, citing the impossibility of verifying nationality in real time. Fable, which was designed for the general public, had its export controls lifted after new safeguards were put in place. However, Mythos, intended for cybersecurity experts, is still provided only to certain "trusted" U.S. institutions. Chinese authorities are concerned that Mythos could be used to exploit software vulnerabilities and that the United States could take advantage of it.
