DeepSeek develops in-house AI chip to reduce Nvidia dependence

Chinese artificial intelligence startup DeepSeek is developing its own inference chip as it seeks to reduce its reliance on Nvidia and Huawei hardware, marking a significant strategic shift for one of China’s leading AI companies, according to people familiar with the matter.

Reuters reported that the Hangzhou-based company has been working on the project for around a year and is holding discussions with chip designers, foundries and memory suppliers. Three sources familiar with the plans said the chip is intended for AI inference, the stage at which trained models generate responses, rather than for training new models. DeepSeek did not respond to Reuters’ request for comment.

The move would place DeepSeek alongside global AI developers seeking greater control over the hardware powering their models. OpenAI last month introduced its first custom inference chip, Jalapeno, developed with Broadcom, while Reuters reported in April that Anthropic has been considering building its own AI chips.

DeepSeek’s effort carries additional strategic importance because US export controls prevent Chinese companies from accessing Nvidia’s most advanced processors. Founder Liang Wenfeng acknowledged in a rare 2024 interview with a Chinese media outlet that chip export restrictions had become a challenge for the business. The company has relied on both Nvidia and Huawei hardware, although it has increasingly adopted Huawei’s Ascend chips after adapting its V4 model to the platform earlier this year.

Reuters reported that DeepSeek has quietly expanded its chip design team in recent months through private recruitment rather than public job listings. If successful, the initiative could increase competition within China’s domestic AI chip market, where Huawei has benefited from US export restrictions but is facing growing competition from companies including Alibaba and Baidu, which are developing their own processors.

The project targets the rapidly expanding inference market, where AI models are increasingly deployed to serve users rather than trained. Designing a competitive AI chip remains expensive and technically demanding, while US restrictions on advanced semiconductor manufacturing and high-bandwidth memory continue to limit Chinese developers’ access to critical technologies.

Reuters reported in June that DeepSeek was preparing to raise $7 billion in its first external funding round, valuing the company at between $52 billion and $59 billion after years of avoiding outside investment.