China's efforts to lead the global artificial intelligence (AI) race appear to be yielding significant results. Technology analysts and industry experts indicate that Chinese AI models are increasingly popular and, in some cases, performing at or even exceeding the levels of their U.S. counterparts.
 |
| Blackdovfx | Istock | Getty Images |
AI has become a critical arena of competition between the United States and China, as both nations regard it as a strategically vital technology. Washington has implemented restrictions to limit China’s access to advanced semiconductor chips, which are essential for powering AI systems. These measures stem from concerns over national security and the potential misuse of AI technologies.
In response, China has devised its own strategy to improve the appeal and performance of its AI models. This approach includes embracing open-source technologies, developing domestic high-speed software, and producing locally manufactured chips to sustain its AI development.
China’s Leading Role in LLM Development
Similar to major U.S. tech companies, Chinese AI firms are heavily investing in large language models (LLMs), which are trained on vast datasets and serve as the foundation for applications like chatbots.
Unlike OpenAI, whose models like GPT-4 power ChatGPT but operate under restrictive licensing, many Chinese companies are focused on developing open-source LLMs. These open-weight models allow developers to download and build upon them freely, without complex licensing hurdles.
Tiezhen Wang, a machine learning engineer at Hugging Face, a leading LLM repository, highlighted that Chinese LLMs are the most frequently downloaded models on the platform. Wang pointed to Qwen, a family of AI models developed by Chinese e-commerce giant Alibaba, as the most popular.
“Qwen’s rapid rise in popularity is due to its excellent performance on competitive benchmarks,” Wang told CNBC in an email. He also noted that Qwen’s open licensing framework makes it highly attractive for companies because it eliminates the need for extensive legal reviews.
Qwen is available in multiple sizes, or parameters, which determine its computational capacity. Large-parameter models are more powerful but come with higher costs, while smaller models are more affordable to operate. “Regardless of the model size, Qwen is consistently one of the best-performing options on the market right now,” Wang added.
In addition to Alibaba, other Chinese players are making waves. For instance, DeepSeek, a Chinese startup, has developed the DeepSeek-R1 model. Last month, the company claimed that R1 rivals OpenAI’s GPT-4o in reasoning capabilities and complex problem-solving.
These advancements have fueled claims that Chinese LLMs can compete with leading open-source models like Meta’s Llama, as well as closed models like OpenAI’s GPT series.
“In the past year, we’ve seen the rapid rise of open-source contributions from Chinese AI developers. These models offer strong performance, high efficiency, and lower operating costs,” Grace Isford, a partner at Lux Capital, told CNBC.
Open Source as China’s Path to Global Dominance
China’s emphasis on open-source LLMs serves multiple strategic goals. Open-source frameworks foster innovation by granting developers broad access to the technology and encouraging collaboration. Moreover, it helps Chinese firms build global communities around their AI models.
Other firms, like Meta and the European startup Mistral, have also released open-source AI models. However, geopolitical tensions between Washington and Beijing have given Chinese firms a unique opportunity to leverage open-source platforms for global adoption.
“Chinese companies aim to see their models widely used outside China,” said Paul Triolo, a partner at the global advisory firm DGA Group. “By adopting an open-source strategy, they can position themselves as global players in the AI market.”
While AI models remain a focal point, the larger question revolves around the applications built on top of these foundational systems — and who will dominate the next generation of global internet technology.
“If you consider base AI models to be foundational, the real opportunity lies in their application. This includes accelerating advancements in science, engineering, and technology,” Isford explained.
Some analysts draw comparisons between today’s AI models and operating systems like Microsoft’s Windows, Google’s Android, and Apple’s iOS. These operating systems dominate their respective markets on PCs and mobile devices, hinting that AI models could follow a similar trajectory.
“The stakes are enormous,” said Xin Sun, a senior lecturer at King’s College London. “Chinese companies view LLMs as the core of future technology ecosystems. Their business models rely on attracting developers to their platforms, encouraging them to create new applications, and generating profits through diverse revenue streams — including cloud services.”
U.S. Chip Restrictions and the Uncertain Future of Chinese AI
Training LLMs requires immense computational power, typically provided by advanced graphics processing units (GPUs). Nvidia is the global leader in designing these high-performance chips.
However, U.S. export restrictions have barred the sale of Nvidia’s most advanced GPUs to China. Over the last year, Washington has tightened these controls to limit China’s access to cutting-edge semiconductor technology. In response, Nvidia has developed less powerful, export-compliant chips specifically for the Chinese market.
Despite these limitations, Chinese firms have managed to maintain progress and release advanced AI models.
“China’s major tech platforms have stockpiled Nvidia GPUs, allowing them to continue improving their AI systems,” Triolo said. “Additionally, domestic Chinese companies, like Huawei, have developed alternative GPUs to support AI growth.”
Huawei has emerged as a frontrunner in China’s quest to build domestic AI hardware. Tech giants like Baidu and Alibaba are also investing heavily in semiconductor design to reduce reliance on foreign technology.
Nevertheless, experts believe that the hardware gap between the U.S. and China will widen over time. Nvidia’s next-generation Blackwell-based GPUs, scheduled for release next year, will likely remain unavailable to China due to U.S. sanctions.
“The restrictions will create a growing challenge,” Triolo warned.
Lux Capital’s Isford noted that China is systematically building its own AI infrastructure. “China is investing heavily in domestic high-performance chips and alternative technologies, like those from Baidu,” she said.
“While restrictions on Nvidia chips pose a hurdle, they will not stop China from developing its own infrastructure to support AI innovation,” Isford concluded.
Final Thoughts
China’s advancements in AI, particularly in the development of large language models, are positioning it as a global leader in the field. Its focus on open-source technology is enhancing the accessibility and appeal of Chinese AI models, allowing them to compete on an international scale.
However, as U.S. export restrictions tighten, the long-term sustainability of China’s AI ambitions will depend on its ability to overcome hardware challenges and develop cutting-edge domestic alternatives. Regardless, China’s momentum in the AI space demonstrates its commitment to becoming a dominant force in this critical technology sector.
