Open models and practical diffusion: China’s strategy for AI dominance
AI competition between the United States and China has been presented as a race that will be won by whoever has the most advanced models. Nevertheless, China has opted for a different strategy, which is to create open models and practical AI applications to achieve greater immediate economic impact and international influence.
In Silicon Valley conferences, discussions, and corridors, the magic word now being repeated everywhere is AGI (Artificial General Intelligence). For the giants in the field, like OpenAI, DeepMind, Anthropic, and xAI, developing AGI—an intelligence that performs better than humans in reasoning ability and broad skills—is the ultimate goal they aspire to, and to which they devote enormous efforts and epic levels of spending. Developing ever more advanced and innovative AI models is believed to be the way of moving one step closer to this goal. Whoever leads in AGI, it is repeatedly stressed, will have an unmatched economic, military, and social advantage. The United States government itself, in both Biden and Trump administrations, also considers that achieving AGI is the great national objective in the AI field, and that the USA must manage to reach it before China does. The AI competition tends to be compared with the nuclear arms race: the U.S. Administration has called its current AI plan the new “Manhattan Project”.
However, the promise of AGI is based on a questionable premise. From the U.S. point of view, the power that creates the most advanced technologies—for example, the one that develops the most cutting-edge AI models—will automatically gain the greatest geopolitical power. Authors such as Jeffrey Ding, however, have found that in major geopolitical transitions linked to General Purpose Technologies (like the steam engine, electricity, or the Internet), the power that gained the most geopolitical influence through them was not the one that invented these technologies, but the one that broadly disseminated and practically applied them across its economy—achieving greater impact in terms of growth and productivity gains. The results of these types of technology usually appear after ten or twenty years, so the economic and subsequent geopolitical effects tend to be long-term.
Unlike the United States, Beijing approaches AI competition more as a matter of widespread adoption and creation of practical applications than as an AGI race. As the economist Keyu Jin has noted, in recent decades China has focused not so much on technological breakthroughs (from 0 to 1) as on scaling up and efficiently bringing existing technologies to market (from 1 to n). It plans to pursue this strategy in the field of artificial intelligence as well. A few months ago, the Chinese government launched the “AI+” diffusion plan, which aims to diffuse and integrate AI by applying it in sectors including manufacturing, governance, healthcare, social welfare, and consumption. In a Chinese Communist Party Politburo AI study session last April, Xi Jinping emphasised the need of “deep integration” of artificial intelligence in the economy and society, and a “strong focus on applications”. Chinese scholars in the technology sector, such as Huang Ping, have also stressed the need to prioritise “innovative resilience” over “breakthrough innovations”. Although there are obstacles that hinder the spread of AI in the Chinese economy—such as the financial troubles of local governments or the poor condition of Chinese private investors—the efforts are clearly moving in this direction.
In promoting this diffusion, China has opted for “open models” of artificial intelligence, which have the advantage of being free-of-charge, open source, and highly flexible. In the field of open AI models, Chinese companies currently outperform those in the United States, with models such as the popular Deepseek and Alibaba’s Qwen leading the rankings. These Chinese open models are especially useful for startups with limited resources, and are widely used by new ventures in both the United States and Japan. Models like Qwen stand out for their flexibility and efficiency, and the fact that they offer different sizes that can be integrated into everything from tiny drones to industrial machinery, as well as functioning as generative models along the lines of ChatGPT. Unlike the chip industry, which has benefited from large Chinese government subsidies, these open models have emerged from a network of competitive private companies and startups originated in technological ecosystems of cities like Hangzhou, where local authorities have nurtured an environment of innovation and flexibility for the private sector.
The Chinese government itself has repeatedly presented its model for AI development as an alternative to that of the United States, arguing that AI should be an “international global good” that supports the development of the Global South, in contrast with the zero-sum competition advocated by Washington. Besides decreasing its reliance on models coming from the United States, the Chinese advancement of free, open-source AI models and practical applications is the mechanism by means of which Beijing is attempting to influence the development and governance of artificial intelligence on a worldwide scale. As Chinese analysts like Liu Shaoshan have indicated, the idea is that, if the world embraces Chinese models ahead of American ones, Beijing could lead the way in setting standards and values around this technology. Di Dongsheng describes this approach in terms of the Maoist strategy of “encircle the cities from the countryside” (农村包围城市) which, in this new context, implies expanding in the Global South before conquering Western markets. On a global scale, diffusion may once again matter more than breakthrough innovations.
The widespread adoption of Chinese open models could contribute to the integration of AI into the global economy, bringing productivity gains that economies such as Europe’s badly need. At the same time, these models raise concerns regarding their built-in censorship of politically sensitive topics in China—for example, if they were used extensively in education or decision-making—and the suspicion that they might contain backdoors, although the very nature of open models, which offer more transparency and control than closed ones, mitigates this issue. In the long term, the fundamental issue is that large-scale adoption of open Chinese models could position China as a leader in this technology with the consequent power to influence the standards, development, and values intrinsic to them. Technological innovations are not neutral and the path they take depends on political decisions. The evolution of the open Internet we now take for granted came about because of the influence and economic and political interests of the United States. An artificial intelligence globally led by Beijing would, for better or worse, be an AI with “Chinese characteristics”, with all the ensuing geopolitical implications.
Keywords: AI, United States, China, open models, AGI, Silicon Valley, technology, geopolitic
All the publications express the opinions of their individual authors and do not necessarily reflect the views of CIDOB or its donors
E-ISSN 2014-0843