China’s push to dominate the AI ​​world may be paying off, experts told CNBC , with the mainland’s AI models, particularly popular, catching up – and even surpassing – their US counterparts in performance.

AI has become a new front in the US-China trade war, with both sides viewing it as a strategic technology. Washington continues to restrict Beijing’s access to cutting-edge chips due to national security concerns. This has led China to pursue its own approach to advancing its own AI models, including relying on open-source technology and developing its own software and chips.

Similar to the US market leaders, Chinese AI companies are also developing large language models (LLMs), which are trained on large amounts of data and are the foundation of applications like chatbots.

Unlike OpenAI's models, however, mainland companies develop open-source LLMs, where developers can download and write applications for free without permission from the author.

china ai model xinhua
A visitor learns about the world's first professional, multimodal large language model (LLM) for lunar science during the China International Big Data Industry Expo in Guiyang, Guizhou, China, on August 29. Photo: Xinhua

On the LLM Hugging Face repository, the Chinese language models are the most downloaded. According to Hugging Face machine learning engineer Tiezhen Wang, Qwen – Alibaba’s AI model system – is the most popular. He said Qwen has been gaining popularity due to its outstanding performance.

Qwen comes in different sizes—or parameters. Larger models are more powerful but come with higher computational costs, while smaller ones are cheaper. Either way, it’s one of the best models out there, according to Wang.

Startup DeepSeek is also emerging with its DeepSeek-R1 model. It competes with OpenAI's o1 model.

These companies claim their models can compete with other open-source solutions like Meta's Llama, as well as closed-source LLMs like OpenAI's on a number of different functions.

Grace Isford, managing partner at investment firm Lux Capital, commented that over the past year they have seen the rise of Chinese open-source models for AI.

In today’s complex geopolitical landscape, open source LLMs have another benefit for Chinese companies: they can be used around the world, not just within the country’s borders. It helps them become global players in AI.

Today’s AI models are being compared to operating systems like Windows, Android and iOS, with the potential to dominate a market. Xin Sun, senior lecturer in Chinese and East Asian enterprise at King’s College London, said Chinese companies see LLM as central to the future technology ecosystem.

Their business models will rely on developers joining the ecosystem, developing new applications based on their LLM, attracting users and data and then generating profits through various tools.

Still, limited access to advanced chips casts a dark cloud over China’s AI prospects. AI models require massive computing power.

Nvidia is now the world’s largest supplier of AI chips. Most leading AI companies train their systems on Nvidia’s cutting-edge chips, but China does not.

Over the past year, the US has tightened controls on exports of semiconductors and advanced semiconductor manufacturing equipment to China, meaning Nvidia’s most powerful chips cannot be sold to the mainland.

However, major Chinese tech platforms have stockpiled large numbers of Nvidia GPUs and are leveraging domestic GPUs from Huawei... to continuously improve the model.

In fact, Chinese companies are constantly trying to find ways to replace Nvidia, with Huawei being one of the leading players. Baidu and Alibaba are also investing in semiconductor design.

According to Isford, China is systematically investing in and developing its entire domestic AI infrastructure that does not revolve around Nvidia. Therefore, whether Nvidia chips are banned from being sold to the mainland or not, it cannot prevent Beijing from investing in and building its own infrastructure to develop and train AI models.

(According to CNBC, SCMP)