DeepSeek is reportedly entering the semiconductor industry with plans to develop its own AI chips. The company has started designing an AI inference chip, marking its first major move into custom hardware, reports Reuters. The project has reportedly been underway for about a year and is still in its early stages, with DeepSeek expanding its chip design team and working with semiconductor partners as it looks to reduce its dependence on external suppliers.
Instead of building a processor for training large AI models, DeepSeek is focusing on an AI inference chip. Inference refers to the stage where a trained model is put to work, generating answers, writing code, creating images, translating text, or summarizing documents for users. Since every interaction with an AI assistant requires inference, demand for such computing power is growing rapidly. A chip designed specifically for these workloads can process requests faster, consume less power, and significantly reduce the cost of running AI models for millions of users.
It is worth noting that DeepSeek’s chip project is also closely linked to the changing global semiconductor landscape. The company previously used Nvidia’s H800 GPUs, a version of Nvidia’s AI chips developed for the Chinese market before stricter US export controls were introduced. As those restrictions expanded, Chinese AI companies faced increasing difficulty in accessing the world’s most advanced AI processors. DeepSeek has since shifted many of its newer AI workloads to Huawei’s Ascend AI chips. However, relying on outside suppliers — whether foreign or domestic — still carries risks related to supply, pricing, and future restrictions. Therefore, building its own chip would give DeepSeek more independence and better control over its long-term AI strategy.
The move shows a wider trend across the AI industry. While Nvidia remains the global leader in AI accelerators, many technology companies are now designing their own chips instead of relying entirely on third-party hardware. DeepSeek is entering a highly competitive market, with China’s domestic AI chip industry estimated to be worth around $50 billion. Companies like Huawei, Alibaba, and Baidu are already investing heavily in their own AI processors. Also, developing an AI chip is much more difficult than developing an AI model. It requires expertise in processor design, memory systems, software tools, packaging technology, manufacturing, and cooling, while also securing access to advanced semiconductor production and high-bandwidth memory.