As China narrows the artificial intelligence (AI) gap with the United States, Beijing is promoting a narrative centered on "open source" and "inclusive AI" as an argument for its vision of global governance.

During his keynote speech at the World Artificial Intelligence Conference (WAIC) in Shanghai on Friday, Chinese President Xi Jinping expressed China's ambition to take a leading role in setting global standards in AI. He criticized what he described as the "overextension of the concept of national security."

Xi did not openly reference the United States. However, Beijing's effort to position itself as an advocate of a new AI order and challenge US dominance was unmistakable.

This year's Shanghai summit sent several important signals, says Rogier Creemers, assistant professor at Leiden University in the Netherlands specializing in China's digital technology policy. The WAIC indicates that artificial intelligence is now one of Beijing's political priorities, and that China intends to compete with the United States by supporting open-source AI models and promoting AI-driven industrialization, Creemers told DW. He also noted China hopes to win support from countries in the Global South for this effort.

Beijing closing the AI gap

China is steadily narrowing the gap with the United States in AI model development, and its efforts at AI application in industrial processes are slowly maturing. 

Chinese AI model developers have continued to keep pace on the global stage, largely by training open-weight models on limited but advanced computer hardware, Paul Triolo, senior vice president for China and tech policy lead at DGA-Albright Stonebridge Group, told DW.

Triolo believes deployment of AI across Chinese enterprises may be progressing faster than in the United States.

In addition, there is less fear of AI among the Chinese population than there is in the US, facilitating more rapid uptake of AI applications at the consumer level.

Earlier this year, Chinese state media claimed that the performance gap between Chinese and US AI models had "largely disappeared."

Several experts broadly agreed with this assessment, saying that certain benchmark tests of China's large language models (LLMs) such as GLM-5.2 and Kimi K3 showed the gap narrowing or even fully closing.

For most applications, Triolo noted, China's leading open-weight AI models are cheaper to run and more than capable of handling most tasks.

At the same time, advanced proprietary models from leading US companies such as Anthropic, OpenAI and Google still retain advantages in areas such as security and scalability.

Has the US 'AI firewall' worked?

In January 2025, the US Commerce Department expanded AI-related export controls through its "Framework for Artificial Intelligence Diffusion," covering advanced computing chips, certain AI model weights and access to large-scale AI computing infrastructure.

The US also boosted export controls on advanced chips and semiconductor manufacturing equipment intended for China.

Laila Khawaja, research director at Gavekal Research, told DW that the US has been partially successful.

The controls have indeed impacted China's ability to manufacture chips domestically, cheaply, and at scale.

As for export controls on advanced AI chips, however, Khawaja notes that a significant amount of chips is still entering China through the gray market, while Chinese firms can still access computing resources through overseas cloud services.

According to Triolo, the evidence so far suggests the US strategy of slowing down China's advanced AI development has failed at many levels. At the same time, it caused massive collateral damage to US companies, especially to to their budgets for research and development and their supply chains, such as those delivering rare earths.

China's open-source AI push

In his keynote speech, Xi emphasized the importance of "encouraging open-source development."

Chinese state media have likewise promoted the message that China "does not pursue technological blockades."

Analysts say Beijing is increasingly framing "open source" and "inclusive AI" as its own narrative to counter the United States' largely closed-source AI ecosystem.

Chinese AI models are largely open-weight, with leading offerings such as DeepSeek, Zhipu's GLM 5.2 and Kimi K3 released under MIT licenses that allow users to download, deploy, modify, distribute and commercialize them.

Creemers noted that, for the time being, Chinese AI models are quite open in weight, but slightly less open in source.

According to Creemers, China hopes to encourage the rest of the world to adopt Chinese AI models, thereby creating overseas markets for Chinese technology and ensuring that China remains part of the next wave of technological development.

China's WAICO challenges Pax Silica

In Shanghai, Xi also announced that the World AI Cooperation Organization (WAICO), an initiative launched by China, has already attracted 29 member countries.

He described the organization as "an important milestone in the history of AI development" and said it responds to calls from Global South countries for a greater role in AI governance. 

WAICO is seen as an initiative to counter the US-led Pax Silica international initiative which aims to secure ​global AI and critical mineral supply chains.

The real competition, according to Triolo, lies in which parts of the US and Chinese AI technology stacks are ultimately adopted by countries in the Global South.

He believes China holds a significant advantage due to lower costs and the extensive digital infrastructure already built by Chinese companies such as Huawei across many developing countries.

Looking ahead

Overall, experts believe that the US will continue to lead in AI hardware design and compute capacity, while China could seize the lead on model innovation and will come to lead on AI diffusion.

Creemers argued that China's biggest structural disadvantage remains the lack of trust from countries in the Global North.

At the same time, US export controls continue to leave Chinese companies facing shortages of advanced AI computing hardware, limiting their ability to rapidly scale inference capacity to meet growing demand.

Looking ahead, Creemers argues that the rise of China's AI sector represents a broader historical shift.

"We need to understand that we are now facing, for the first time in modern history, a non-Western technological superpower," he said.

In his view, the future will not be defined by a singular global technological sphere, but rather by "a complex of overlapping technological networks and processes."

Edited by: Keith Walker