DeepSeek is building its own AI chip to cut reliance on Nvidia and Huawei, report

DeepSeek is making one of its biggest strategic moves yet. The Chinese AI startup that stunned the industry early last year with its low-cost reasoning models is now developing its own AI chip, signaling a desire for greater control over the hardware behind its technology rather than relying on suppliers like Nvidia and Huawei.

According to a Reuters report citing three people familiar with the project, DeepSeek has spent the past year quietly developing a custom inference chip, marking a shift from its earlier focus on AI models. The effort places the company alongside OpenAI, Anthropic, Google, Amazon, and other AI leaders investing heavily in custom silicon to reduce dependence on external chipmakers and lower the cost of running AI systems at scale.

The news sent Nvidia shares down about 1.6% in premarket trading, reflecting investor attention to a broader trend reshaping the AI infrastructure market.

The chip DeepSeek is developing is intended for inference rather than training. Inference is the stage where trained AI models generate responses for users, and it has become one of the fastest-growing areas of AI computing, with chatbots, coding assistants, enterprise tools, and autonomous agents handling millions of requests every day.

“The chip is designed ‌for inference — the stage of AI computing in which a trained model generates responses for users — rather than for training new models,” Reuters reported, citing the sources.

If the project succeeds, it would represent a major milestone for a company that has become one of China’s most closely watched AI startups. It could shift the competitive balance inside China’s AI chip market, where Huawei has emerged as the primary domestic alternative after U.S. export restrictions limited access to Nvidia’s most advanced processors.

“Nvidia is at zero in China and staying there. DeepSeek has almost no chance of selling silicon outside of China unless it gets access to leading-edge manufacturing,” said Analyst Richard Windsor of Radio Free Mobile, adding that the development does not affect the chipmaker.

DeepSeek shot to international prominence after releasing a pair of efficient AI models that surprised researchers and investors with their performance relative to their cost. The company’s R1 reasoning model, released in early 2025, triggered a selloff across U.S. technology stocks as investors questioned whether future AI systems would require the massive computing budgets many companies had projected.

That success established DeepSeek as one of China’s leading AI companies. Until now, its reputation has centered on software rather than hardware.

DeepSeek’s custom AI chip could reshape China’s battle against Nvidia’s dominance

China’s AI chip market has changed quickly over the past two years. U.S. export controls have blocked Chinese companies from purchasing Nvidia’s high-end processors, creating an opening for Huawei. The Chinese technology giant now supplies a large share of the country’s AI infrastructure, serving DeepSeek along with many of China’s largest technology companies.

Huawei’s lead is no longer guaranteed. Alibaba and Baidu have each invested heavily in their own AI processors, giving major cloud providers greater control over the chips running their AI platforms. DeepSeek now appears ready to follow the same strategy.

Sources told Reuters the startup began exploring chip development roughly a year ago and has quietly expanded its semiconductor hiring. Instead of advertising positions publicly, the company has recruited chip engineers through private channels and has held discussions with chip design firms, foundries, and memory suppliers as it builds the project.

The effort remains in its early stages. Developing a competitive AI processor demands years of engineering work, billions of dollars in investment, and access to manufacturing technology that remains heavily restricted for Chinese companies.

DeepSeek has faced hardware limitations before. Founder Liang Wenfeng acknowledged in a rare 2024 interview that U.S. export controls had become a major obstacle for the company. DeepSeek previously disclosed that the foundation model behind its R1 reasoning system was trained on Nvidia’s H800 processors, chips designed specifically for the Chinese market, which Washington later prohibited from being sold.

More recently, DeepSeek has leaned further into Huawei’s ecosystem. In April, the company introduced its V4 model optimized for Huawei’s Ascend chips. Huawei later said its processors handled part of the training workload for V4-Flash, a lighter version of the model. That announcement helped drive a wave of demand for Huawei’s Ascend 950 chips from Chinese technology companies.

DeepSeek’s decision to build an inference chip reflects a broader shift taking place across the AI industry. As AI applications move from research labs into everyday products, inference has become the largest source of computing demand. Running AI models for millions of users requires processors built for efficiency rather than raw training performance, creating a growing market for specialized chips.

The same trend has pushed several global AI developers into semiconductor design. OpenAI recently unveiled Jalapeno, its first custom inference chip developed with Broadcom. Meanwhile, Reuters previously reported that Anthropic has also explored building its own AI processors. Owning more of the hardware stack gives AI companies greater control over cost, supply, and long-term product development.

DeepSeek’s hardware ambitions arrive as the company opens another new chapter. After years of rejecting outside investors, Reuters reported in June that the startup is preparing its first funding round, seeking about $7 billion at a valuation between $52 billion and $59 billion.

If both efforts move forward, DeepSeek would no longer be viewed simply as the Chinese startup that surprised Silicon Valley with efficient AI models. It would be positioning itself as a full-stack AI company, building both the intelligence and the chips needed to run it.