Chinese Startup DeepSeek Developing Custom AI Inference Chip

Chinese startup DeepSeek is developing its own AI chip designed for inference rather than training new models, according to three people familiar with the matter. The chip would reduce the company’s dependence on Nvidia and Huawei processors, which it has used to train and run its models.

DeepSeek reached out to external partners and held discussions with chip-design, foundry and memory companies.

The effort began about a year ago, one of the sources said. DeepSeek has increased hiring of chip-design engineers in recent months without posting openings on public platforms, two of the sources said. All three people declined to be identified because the information is not public.

DeepSeek did not respond to a request for comment. Nvidia shares slipped about 2% in premarket trading after the report. DeepSeek rose to global attention more than a year ago after releasing two highly efficient AI models that gained wide use.

Huawei holds around half of the $50 billion domestic AI chip market in China, though its share has declined as Alibaba and Baidu develop their own chips. DeepSeek has used both Nvidia and Huawei processors, including Nvidia’s H800 for one model and Huawei’s Ascend chips for later versions. Founder Liang Wenfeng said in a 2024 interview that chip export controls were a challenge for the company.

DeepSeek was slated to raise $7 billion in a funding round valuing it at between $52 billion and $59 billion, Reuters reported in June. Other AI developers have pursued similar hardware strategies. OpenAI unveiled its first custom inference chip, developed with Broadcom, in June 2026.

Anthropic has been weighing building its own AI chips, Reuters reported in April. Designing and manufacturing a competitive AI chip typically takes years and significant capital. U.S. export controls bar Chinese companies from the most advanced overseas foundries and have restricted access to high-bandwidth memory needed for inference chips.