Chinese AI startup DeepSeek is working to design its own artificial intelligence chip, according to three individuals with knowledge of the project — a development that could reduce the company’s reliance on processors from Nvidia and Huawei, which it has used to build and operate its widely used AI models.
The chip being developed is intended for inference — the phase of AI computing where a trained model produces responses for users — rather than for the training of new models, the sources indicated.
Should the effort prove successful, DeepSeek’s move into chip design would represent a significant strategic change for a company celebrated in China as a national AI leader, and could add competitive pressure on tech giant Huawei.
DeepSeek gained worldwide attention more than a year ago after releasing two highly efficient AI models that spread rapidly across the globe, catching many in Silicon Valley and Washington off guard.
The company has historically focused on advancing AI model development rather than commercializing its technology.
While Huawei’s chips still fall considerably short of Nvidia’s most advanced offerings, U.S. restrictions on exporting those Nvidia chips to China have allowed Huawei to capture roughly half of China’s $50 billion domestic AI chip market, with DeepSeek and several other major players among its customers.
That dominance is already being tested, however, as tech rivals Alibaba and Baidu develop their own chips and claim a growing share of the market.
DeepSeek’s push into chip development is still in its early stages. The company has been reaching out to outside partners and holding talks with chip-design, foundry, and memory companies, the three sources said. One of the individuals noted the effort started roughly a year ago.
The Hangzhou-based firm has also been quietly bringing on chip-design engineers in recent months, though the hiring has been conducted without any public job postings, two of the sources said.
All three sources asked not to be named because the matter has not been made public. DeepSeek, which has maintained a low profile despite its prominence in China’s AI sector, did not respond to a request for comment.
By pursuing an in-house chip, DeepSeek would be following a broader trend among global AI companies seeking greater control over the hardware powering their systems and less dependency on Nvidia.
OpenAI last month unveiled its first custom inference chip, called Jalapeno, developed in partnership with Broadcom. Separately, Anthropic has been exploring the possibility of building its own AI chips, according to a Reuters report from April.
For DeepSeek, the move carries additional strategic weight. U.S. export controls prevent Chinese companies from purchasing Nvidia’s most advanced chips, and Beijing has been pushing its technology companies to develop homegrown alternatives.
DeepSeek founder Liang Wenfeng acknowledged in a rare 2024 interview with a Chinese media outlet that chip export restrictions posed a challenge for the company.
DeepSeek has relied on both Nvidia and Huawei chips over time. The company has stated that the underlying model powering R1 — its reasoning model whose low-cost performance rattled U.S. tech stocks in January 2025 — was trained using Nvidia’s H800, a chip made for the Chinese market that Washington banned in late 2023.
More recently, DeepSeek has leaned more heavily on Huawei. In April, it released its V4 model adapted to run on Huawei’s Ascend chips, and Huawei confirmed its processors played a role in training V4-Flash, a lighter version of the model. Orders for Huawei’s Ascend 950 chips from major Chinese tech companies surged following that launch, Reuters has reported.
A DeepSeek inference chip would enter the fastest-growing area of AI computing. As AI applications become more widespread, the industry is shifting more of its computing workload from training models to running them — a task that relies on specialized chips that can be less expensive and more energy-efficient than general-purpose processors.
Still, success is far from certain. Designing a competitive AI chip typically takes years and requires substantial investment. Manufacturing presents additional obstacles, as U.S. restrictions block Chinese chip designers from using the most advanced overseas production facilities, while separate controls have limited China’s access to high-bandwidth memory — a key component for AI inference chips.
DeepSeek’s chip ambitions come alongside the company’s first move toward accepting outside investment. The company was set to raise $7 billion in an initial funding round that would value it at between $52 billion and $59 billion, Reuters reported in June — a reversal of its longstanding policy of turning down external capital.
