The Chinese Communist Party is attempting to gain access to big data resources and AI rule-making authority through open source, low prices, and market scale. Experts say that what truly determines the future of AI competition is no longer just the models themselves, but also factors such as security, computing power, ecosystem, and institutional trust.
The 8th World Artificial Intelligence Conference (WAIC) and the high-level meeting on global AI governance were held on July 17 in Shanghai. According to Xinhua News Agency, Chinese leader Xi Jinping attended the conference for the first time and delivered a speech, emphasizing the importance of seizing historical opportunities, open source, cooperation, and sharing.
The Chinese state media outlet “People’s Daily” published a commentary stating that technology should not be a dividing wall between countries and should not be used for zero-sum games. “Sina Finance” quoted “Observer Net” as saying that the U.S. AI bullying must be rejected.
Reports from “Deutsche Welle” suggest that Xi Jinping aims to establish a new global AI order, lead the development of global AI standards, and challenge the U.S. dominance. BBC reports quoted analysts saying that “open source” has become a key term in China’s AI strategic narrative.
“Deutsche Welle” also points out that China’s release of open source AI models like DeepSeek challenges the U.S.’ dominance in closed-source AI products in the hope of replacing them.
While the AI conference was being held in Shanghai, both the U.S. and China were preparing for their first government-level dialogue on artificial intelligence.
In late 2023, when China launched DeepSeek, there was a significant reaction from Western societies. However, the response to China’s AI development, including hosting the AI conference and attempting to reshape AI regulations, seems to have been relatively calm.
Assistant Researcher Yang Yikui from the Taiwan Institute of National Defense and Security stated that through propaganda and public opinion manipulation, the Chinese Communist Party has created a strong narrative around the rise of Chinese AI with the launch of DeepSeek. Currently, the Western world seems indifferent towards Chinese AI, which was inevitable.
Yang Yikui further stated, “The impact of DeepSeek on the U.S.-China AI competition is more virtual than real, and the cognitive operation and data security risks of DeepSeek will limit its expansion on an international scale.”
Assistant Researcher Wang Xiuwen from the Taiwan Institute of National Defense and Security also mentioned that, despite China’s models emphasizing open source, there are concerns about data security. She added, “The ban on the use of Chinese AI products by the U.S. government and companies is primarily based on national security considerations.”
On January 20, 2025, DeepSeek Company released the open-source DeepSeek-R1 large language model, claiming its performance is comparable to some of the most powerful AI developed by U.S. enterprises, and can be downloaded completely for free by anyone. However, the U.S. proposed banning the use of this model on government devices soon after.
Chinese AI products are very cheap, even free. What is the purpose of being cheap or free? Why does the U.S. prohibit the use of Chinese AI products on government devices?
Yang Yikui stated that the AI competition between the U.S. and China is gradually shifting towards who can control the vast international AI application market and related ecosystems. The ability to dominate the major share of AI applications in the future and possibly make other countries, enterprises, and industries reliant on their technology will give them the power to set rules, technical standards, and industry norms.
He mentioned that China continues its strategy of government subsidies and price competition, aiming to rapidly expand the application scale of AI models and platforms through low prices or even free offerings. At the same time, they are actively promoting the integration of AI technology with industrial manufacturing supply chains, aiming to acquire more industry data through large-scale practical applications, continually improve model efficiency, and seize the opportunity for deep integration of Chinese AI technology with multinational industrial manufacturing supply chains.
Wang Xiuwen also commented that the main reasons for the cheap prices of Chinese AI products are strong support and subsidies from the government and distilling data mainly from original large models in the U.S., eliminating the expensive costs of starting from scratch. The main purpose of these “cheap” offerings is to gain the primary market share and establish an application ecosystem based on Chinese AI standards.
However, the U.S. government views Chinese AI models with a high degree of caution. Yang Yikui mentioned that Chinese AI models and applications may collect a large amount of personal information from users. Influenced by Chinese laws and the national security system, data controlled by Chinese AI companies may be required to be provided to the Communist government or intelligence agencies under certain circumstances.
He emphasized that the U.S. is also concerned that China’s active subsidization of AI enterprises could create unfair market competition. If the U.S. market is fully open, Chinese AI models could rapidly spread due to their price advantage, increasing the risk of Chinese companies gaining access to, misappropriating, or improperly using precise U.S. industrial and manufacturing data.
He further added that in the long run, this could lead to the gradual dependence of the U.S. on Chinese AI platforms, software systems, or related ecosystems in key technological areas such as smart manufacturing, automated logistics, autonomous driving, industrial robots, and other critical fields, thereby affecting U.S. industrial security, supply chain resilience, and national security.
Nevertheless, he stated that China lags behind the U.S. in cutting-edge developments and advanced technologies in the AI field. It is unable to compete with the U.S. in frontiers and ideological areas when it comes to AI.
Using Chinese AI could not only lead to the theft of related information but also pose risks such as misinformation and distorted information in search results.
Yang Yikui remarked that studies indicate that Chinese AI models, in addition to the risks of data misuse and leaking personal information, also have internal security vulnerabilities, posing risks to the development of new software and enterprise applications.
Testing conducted by the Center for Artificial Intelligence Standards and Innovation (CAISI) under the U.S. Department of Commerce found that the hijacking risk of DeepSeek’s R1-0528 model compared to U.S. AI models was 12 times higher. DeepSeek’s R1-0528 seems more susceptible to malicious hijacking, leading to activities such as phishing emails, downloading malicious software, and stealing user login information.
Regarding the “jailbreak” tests, DeepSeek generated malicious content exceeding 90%, while the success rate for “jailbreaking” U.S. AI models ranged from 5% to 12%.
Research also suggests that the DeepSeek model may possess tendencies to align with specific positions and narratives of the Chinese government. The ideological scrutiny mechanism of Chinese AI models may be incorporated into the model architecture, training data, and weight distribution. Hence, even if users are outside China, the model may retain its original political biases and content moderation actions.
Wang Xiuwen added that using Chinese AI for searches might yield results with pre-set value biases, such as favoring communism over capitalism or deliberately avoiding certain sensitive issues.
The Artificial Intelligence Index Report 2026 from Stanford University’s Human-centered Artificial Intelligence Research Center (Stanford HAI) indicates that the world has entered a period of dual-polar AI competition between the U.S. and China, with each side having its own strengths.
As per the report, in February 2025, DeepSeek-R1 briefly kept pace with top U.S. models, but by March 2026, the top model from the U.S.-based Anthropic had a slight 2.7% lead over China.
The U.S. leads in Frontier Models with entities like OpenAI, Anthropic, Google DeepMind, xAI, and more holding the world’s top AI models and influential patents, maintaining being the global core of AI innovation. Additionally, the U.S. leads in GPU, ecosystem, and capital markets.
On the other hand, China has an advantage in the number of published papers, citations, patent outputs, and the installation volume of industrial robots, or even leads in some areas.
The report suggests that while the U.S. and China have their respective strengths in the AI field, the performance gap between their AI models has largely been eliminated. The focus of AI competition is shifting from model capabilities to computing power, funding, talent, energy, and ecosystems.
The top 5 strongest models in the world (Anthropic, Google, OpenAI, xAI, Alibaba) are all from the U.S., while China’s DeepSeek ranks 6th.
Looking ahead, countries will strive to establish independent systems for models, chips, computing power, data, and regulations. The report has identified “AI sovereignty” as a new analytical framework.