DeepSeek Moves to Develop In-House AI Chip in Bid to Reduce Reliance on Nvidia, Huawei

Chinese artificial intelligence firm DeepSeek is taking a significant step toward greater technological independence by developing its own AI inference chip, a move that could lessen its dependence on industry heavyweights Nvidia and Huawei while strengthening its position in China's rapidly evolving AI sector. According to three people familiar with the matter, the Hangzhou-based startup has been quietly working on the project for about a year. The custom chip is being designed specifically for AI inference —the stage where trained models generate responses to users—rather than for training new AI models. If successful, the initiative would mark a major strategic shift for DeepSeek, a company that has earned global recognition for developing highly efficient AI models but has largely avoided commercializing its technology or venturing into semiconductor development. The development also signals increasing competition within China's AI ecosystem, where technology firms are racing to reduce reliance on foreign chipmakers amid tightening U.S. export restrictions. Investor sentiment reflected the news, with shares of U.S.-based chip giant Nvidia slipping about 1.6 percent in premarket trading following reports of DeepSeek's chip ambitions. Commenting on the development, Richard Windsor, an analyst at Radio Free Mobile, downplayed the immediate impact on Nvidia. "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," he said, adding that the development does not materially affect the U.S. chipmaker. DeepSeek shot to international prominence more than a year ago after releasing two highly efficient AI models that gained widespread attention across the global technology industry, surprising competitors in Silicon Valley and policymakers in Washington. The company has since become one of China's leading AI startups. Until now, DeepSeek has relied on chips from both Nvidia and Huawei to power its models. Its R1 reasoning model, which triggered a major selloff in U.S. technology stocks in January 2025 due to its impressive low-cost performance, was trained using Nvidia's H800 chips—processors specifically designed for the Chinese market before they were later banned under expanded U.S. export controls. In recent months, however, the company has increasingly adopted Huawei's Ascend AI chips. Earlier this year, DeepSeek unveiled its V4 model optimized for Huawei's Ascend processors, while Huawei confirmed its chips were also used in training the lighter V4-Flash model. The launch reportedly fueled strong demand for Huawei's Ascend 950 processors among Chinese technology companies. Sources familiar with DeepSeek's latest plans said the startup has begun discussions with chip-design firms, semiconductor foundries and memory manufacturers as part of its effort to build its own processor. The company has also quietly expanded recruitment of chip-design engineers without advertising vacancies on public job platforms. The project comes as several global AI companies seek greater control over the hardware powering their artificial intelligence systems. Last month, OpenAI unveiled its first custom inference chip, Jalapeno, developed in partnership with Broadcom, while Anthropic has also reportedly explored designing proprietary AI chips. For DeepSeek, the strategy carries added geopolitical significance. U.S. export restrictions continue to limit Chinese firms' access to Nvidia's most advanced AI processors, while Beijing has encouraged domestic technology companies to accelerate the development of homegrown alternatives. DeepSeek founder Liang Wenfeng acknowledged the challenge posed by chip restrictions during a rare interview in 2024, describing export controls as a significant obstacle for the company. Industry experts note that while inference chips represent one of the fastest-growing segments of AI computing—driven by rising demand for AI-powered applications—developing competitive semiconductors remains an expensive and technically demanding undertaking. Chinese companies also face additional hurdles due to restrictions on access to cutting-edge manufacturing facilities and high-bandwidth memory technologies essential for advanced AI chips. DeepSeek's semiconductor ambitions also coincide with a broader shift in its business strategy. After years of avoiding outside investors, the company is reportedly preparing to raise approximately $7 billion in its first external funding round, a deal that could value the AI startup between $52 billion and $59 billion , providing additional resources to support its long-term expansion into both artificial intelligence and chip development.