Reuters reported on the 7th (local time), citing multiple sources familiar with the matter, that DeepSeek, a Chinese artificial intelligence (AI) startup, is pushing to develop its own AI Semiconductor. The move is seen as an effort to reduce dependence on Nvidia and Huawei. As U.S. controls on semiconductor exports to China drag on, Chinese AI corporations are speeding up efforts toward "chip self-reliance."
According to the report, DeepSeek has been in talks for a year with semiconductor design corporations, foundries (contract semiconductor manufacturing), and memory corporations to develop in-house an AI chip optimized for AI inference. DeepSeek has also increased hiring of chip design engineers in recent months, but recruited technical staff privately without posting on job platforms, the sources said.
Reuters noted that major AI corporations have recently chosen a strategy of developing their own semiconductors to strengthen control over AI infrastructure and reduce dependence on Nvidia, and that DeepSeek has joined this trend. Earlier, OpenAI, the developer of ChatGPT, unveiled the "Jalapeño," an inference-focused AI chip co-developed with Broadcom, and Anthropic is also reviewing the development of its own AI chip.
Unlike the two corporations, DeepSeek is directly affected by U.S. export controls. As it has become harder for Chinese AI corporations to secure Nvidia's latest AI chips, they have increased adoption of China-made semiconductors and expanded in-house chip development.
DeepSeek has so far used Nvidia's lower-spec AI chips alongside Huawei semiconductors. It has said that the R1, an inference AI model that drew global attention early last year, was developed with Nvidia's lower-spec AI chip, the H800. Since then, it is said to have gradually increased the share of Huawei chips.