Nvidia NASDAQ:NVDA fell 1.63% intraday after Reuters reported that DeepSeek, the Hangzhou-based AI startup behind the globally viral R1 reasoning model, is developing its own AI chip. The effort is focused on inference (the stage where a trained model generates responses), and remains early-stage, with the company reaching out to chip-design, foundry, and memory partners over roughly the past year. DeepSeek has quietly increased hiring of chip-design engineers without posting public job listings, three people familiar with the matter said.
U.S. export controls banned Chinese companies from accessing Nvidia's most advanced chips, and DeepSeek has been forced to rely on a mix of Nvidia's older H800 (which Washington banned in late 2023) and Huawei's Ascend processors, which still lag Nvidia's best by a wide margin. An in-house inference chip would reduce that dependency. Huawei currently holds roughly half of China's $50 billion domestic AI chip market, but that position is already under pressure from Alibaba NYSE:BABA and Baidu, which are building their own chips and gaining share.
DeepSeek would not be alone in the push. OpenAI launched Jalapeno, its first custom inference chip developed with Broadcom NASDAQ:AVGO, last month, and Anthropic has been weighing a similar move.