DeepSeek Is Building Its Own AI Chip to Cut Reliance on Nvidia and Huawei

DeepSeek, the Chinese company behind the model that rattled tech stock markets in January 2025, is now designing its own artificial intelligence chip. According to Reuters, citing three people familiar with the matter, the effort has been underway for roughly a year and aims to reduce the company's dependence on Nvidia and Huawei processors.

What Reuters Found

The Hangzhou-based company, regarded in China as a national AI champion, reportedly launched the chip project as a quiet response to the growing risk of being cut off from foreign semiconductor suppliers. Talks with outside partners span chip design firms, foundries and memory manufacturers. Chip engineers are being recruited without public job postings, a move meant to keep the project out of the spotlight.

The choice of segment matters: DeepSeek is targeting an inference chip, not one for training models. It's a practical call, since inference makes up a growing share of AI companies' operating costs once models are deployed and handling millions of user queries a day. An in-house inference chip would let DeepSeek cut the cost of running its models without having to buy further batches of Nvidia or Huawei cards.

Hardware and Export Barriers

Designing a competitive AI chip usually takes years and requires enormous capital, and DeepSeek is starting essentially from scratch compared with Nvidia or even Huawei. On top of that, US export restrictions cut Chinese chip designers off from the most advanced foreign foundries capable of manufacturing chips on the latest lithography processes.

Separate US restrictions have also limited China's access to HBM (high-bandwidth memory), a component critical to the performance of inference chips. Without access to state-of-the-art foundries and HBM memory, even a well-designed DeepSeek chip might fail to match Nvidia cards in performance, limiting the real-world payoff of the whole effort.

Implications for Huawei and China's Market

If the project succeeds, it would mark a major strategic shift for a company so far known solely for its language models, not hardware. DeepSeek's move into semiconductors also adds pressure on Huawei, which has spent years trying to build a domestic alternative to Nvidia chips through its Ascend line but still faces manufacturing constraints stemming from its lack of access to advanced lithography processes.

For Beijing, every step toward chip self-sufficiency carries weight beyond any single company. China has spent years investing in building a domestic semiconductor supply chain in response to US export restrictions, and a successful DeepSeek chip would demonstrate that even companies outside the traditional chip industry can meaningfully reduce reliance on Nvidia.

What It Means for the Global Market

For companies and developers using DeepSeek's models outside China, the news changes nothing practical for now, the project is at an early stage and could take years to reach production. But it carries signal value: it shows that US export pressure is pushing more Chinese AI companies into hardware, not just software, which could reshape the balance of power in the global AI chip market over the longer term.

For Polish companies and institutions using open source models, including the DeepSeek family, the key issue remains the availability and cost of compute. If Chinese model providers eventually field their own cheaper inference hardware, it could translate into lower prices for hosted APIs down the line, though that's a distant prospect that hinges on clearing the export and manufacturing barriers described above.

Sources: Reuters via Japan Times (japantimes.co.jp), Bloomberg (bloomberg.com), Reuters via Taipei Times (taipeitimes.com)