Beijing: Chinese artificial intelligence company DeepSeek is developing its own AI chip as it seeks greater control over the technology needed to run its powerful models, according to people familiar with the project. The move comes as Chinese technology companies face growing pressure to reduce their dependence on foreign semiconductors and build a stronger domestic technology industry.
The chip project is said to be at an early stage, and DeepSeek has not publicly announced a timetable for its completion. The company is reportedly working with outside partners and recruiting engineers with experience in chip design. The main focus of the project is believed to be an inference chip, which is used to run trained AI models and provide answers to users.
Inference is an important part of artificial intelligence services. After an AI model is trained, it needs large amounts of computing power to answer questions, create images, write text and carry out other tasks. As millions of people use AI services, the cost of running these models can become very high. Developing a specialised chip could help DeepSeek reduce these costs and improve the speed and efficiency of its services.
DeepSeek became one of the most closely watched AI companies in China after its models attracted global attention. The company surprised the technology industry by showing that advanced AI models could be developed with fewer computing resources than many experts had expected. Its success raised questions about the amount of expensive hardware needed to build and operate powerful AI systems.
However, DeepSeek has continued to depend on outside hardware suppliers. Like many AI developers, the company has used advanced processors from Nvidia, the American chip company that dominates the global market for AI computing. United States export restrictions have made it harder for Chinese companies to obtain Nvidia's most powerful chips.
China has responded by increasing efforts to develop its own semiconductor industry. Chinese companies, including Huawei, have been working to create processors that can support advanced AI systems. DeepSeek has also taken steps to make some of its technology work more effectively with Chinese made hardware.
Reports earlier this year indicated that DeepSeek was adapting its AI technology for Huawei's chip systems. This was seen as another sign that Chinese AI companies are looking for alternatives to American technology. However, replacing advanced foreign chips remains difficult because AI hardware requires not only strong processors but also advanced memory, manufacturing technology and highly developed software.
DeepSeek's reported decision to focus first on an inference chip may be a practical approach. Training a large AI model requires enormous computing power and some of the world's most advanced semiconductors. Inference can also be demanding, but companies can design specialised chips that are better suited to their own models and services.
If DeepSeek succeeds, it could gain greater control over both its AI software and the hardware used to operate it. This could help the company lower costs, improve performance and reduce its exposure to restrictions on foreign technology.
The development also reflects the wider technology competition between China and the United States. Washington has introduced restrictions aimed at limiting China's access to some advanced semiconductor technology and chip making equipment. American officials say the controls are related to national security concerns, while Beijing has criticised them as an attempt to slow China's technological development.
These restrictions have encouraged Chinese companies and the government to invest heavily in domestic technology. China is seeking progress across the entire semiconductor supply chain, including chip design, manufacturing, advanced packaging, memory and software.
DeepSeek's reported chip plans are also important because the company has built its reputation on finding more efficient ways to develop AI. Its earlier models showed that software improvements and careful use of computing resources could help reduce the cost of AI development. Creating hardware designed specifically for its own models could be the next stage of that strategy.
Still, designing a chip is only one part of the challenge. DeepSeek would need access to manufacturing capacity and other important components before it could produce the chips on a large scale. It would also need strong software support to ensure that its AI models can run efficiently on the new hardware.
There is currently no confirmed date for the launch or mass production of the reported chip. DeepSeek has also not provided detailed public information about the project. This means the final performance and commercial importance of the chip remain uncertain.
Even so, the reported development shows how quickly China's AI industry is changing. Companies are no longer focusing only on building better AI models. They are also trying to gain more control over the hardware, software and infrastructure needed to support them.
For DeepSeek, developing its own AI chip could mark a major shift from being mainly a model developer to becoming a company with greater control over its entire AI computing system. If successful, the project could strengthen the company's position in China's fast growing AI industry and support the country's broader effort to become more independent in advanced technology.