DIGITIMES Research observes that China is rapidly advancing in the field of LLMs. Major technology giants such as Baidu, Alibaba, and Tencent have launched their self-developed LLMs. Chinese-developed LLMs are continuously improving in terms of parameter count, functionality, and application scenarios, according to the latest report on China's development of AI chips.
However, the development of these LLMs relies heavily on high-performance AI chips. In light of the US chip ban, Chinese companies are compelled to accelerate the R&D of domestic AI chips. At the same time, the Chinese government has introduced policies and measures to support the development of these domestic AI chips.
Furthermore, China's development of domestic AI chips must also address the challenges posed by the dominance of x86 and Arm architectures. China is adopting the RISC-V architecture for its self-developed chip strategy to realize the goal of achieving complete self-sufficiency in AI chip production, the report's study shows.
The three key elements in the development of generative AI include LLMs, data resources, and application software. Among these, AI chips are crucial for both LLM and data resources, while information security impacts data resources and application software. Additionally, the characteristics and demands of specific fields influence the selection of LLM and application software.
Currently, in addition to the trend of increasing parameter counts in China's LLMs and the development of multimodal architectures, China is also advancing rapidly in generative AI. Notably, it leads the world in the number of patents in this field.
Under the impact of the US chip ban, China is accelerating the development of its domestic AI chip technology to reduce reliance on US chips. Companies such as Baidu, Cambricon, and HiSilicon have launched high-performance AI chips, including Baidu's Kunlun series, Cambricon's MLU series, and HiSilicon's Ascend series. China's domestic AI chips have demonstrated superior energy efficiency compared to Nvidia's HGX H20.
To address the challenges posed by chip bans and architectural monopolies, the Chinese government is actively promoting open-source RISC-V architecture as a means to achieve independent technological development. This initiative aligns with the nation's goals outlined in the "14th Five-Year Plan." The government is supporting this movement through funding from the National IC Fund, fostering the growth of RISC-V technical talent, and engaging in the establishment of international standards. These efforts aim to advance the development and commercial application of RISC-V technology in China.
