China's AI Opening Strategy and Global Standard Competition: The Open Source Narrative of WAIC and Western Anxiety

Executive Summary

Following the DeepSeek shock, China has elevated the distribution of open-weight models to a national strategy, as evidenced by Alibaba's Qwen achieving 3 billion downloads and the participation of purchasing delegations from 177 countries at WAIC 2026. This indicates that the strategy is leading to actual commercial dissemination. Faced with restricted access to advanced semiconductors, China has shifted the competitive front by expanding its developer ecosystem, while the United States has yet to establish a coherent alliance system in response. The next 12 to 18 months are likely to see a path where narrative competition becomes institutionalized, with ongoing friction between camps rather than a dual coexistence. Given that South Korea is already intertwined with both ecosystems in semiconductor, AI infrastructure, and standard policies, it is necessary to establish a selective dual-track principle where national core infrastructure prioritizes stability, while the private industrial sector allows for cost efficiency. To this end, both the government and enterprises must undertake a short-term assessment of their dependence on foundational models and evaluate the risks of adopting open-weight models.

I. Issue Situation Analysis

China's AI Opening Strategy and Global Standard Competition: The Open Source Narrative of WAIC and Western Anxiety

1. Background and Progress

China's open-source AI strategy was elevated to a national strategic level following the DeepSeek (DeepSeek-R1) shock in early 2025. The success of DeepSeek, which created cracks in the US-led AI development narrative with its low-cost, high-efficiency model, instilled confidence in the Chinese government. Subsequently, the release of follow-up models such as Alibaba's Qwen, Zhifubai's Z.ai, Moonshot AI's Kimi, and Minimax in an open-weight manner established the construction of an open ecosystem as a direction for the entire industry rather than individual corporate choice.

The stage on which this trend was completed as a national narrative was the World Artificial Intelligence Conference (WAIC) held in Shanghai in July 2026. At this event, President Xi Jinping publicly promised technology sharing and support for enhancing AI capabilities in developing countries. China launched the World Artificial Intelligence Cooperation Organization (WAICO) at this event, attended by representatives from 102 countries and international organizations, including 29 countries such as Russia. Kyle Chan, a research fellow at the Brookings Institution, analyzed this as a dual strategy for China to enhance its image as an inclusive AI leader while simultaneously reducing the world's technological dependence on US AI. The scale of the event was also significant, with 400,000 on-site visitors, purchasing delegations from 177 countries, and an intended purchase amount of approximately 20.36 billion yuan, a 25% increase from the previous year.

2. Current Situation

Chinese state media, including the Global Times, highlight the achievements of the open strategy by presenting specific figures regarding the overseas spread of domestic AI models. Reports indicate that Alibaba's Qwen surpassed 3 billion downloads globally in the last six months, ranking first in major rankings, ahead of Meta and Google. The Global Times stated that despite US export controls and political constraints, Chinese AI companies are expanding overseas developer adoption by leveraging open-source strategies and cost competitiveness. The same media quoted domestic experts who argue that global AI competition should be based on innovation capabilities, characterizing the US approach as framed by geopolitics.

This narrative has led to a direct criticism of Washington's response strategy. The Global Times compared Reuters' report on the US-led AI alliance concept to a "stone thrown into calm waters," asserting that the US AI separation strategy has "arrived dead on arrival." This is a preemptive discourse response that marks the US's attempts to rally allies to exclude Chinese AI as ineffective.

There is a sense of anxiety within the Western camp regarding this narrative offensive. The Brookings Institution has specifically pointed out why Washington fears China's open-source AI. Foreign Policy highlighted that AI technology in China is not subject to social debate as it is in the US, leading to broader acceptance, and diagnosed that this acceptance gap results in differences in the speed of dissemination. The Diplomat cited a Wall Street Journal report indicating that concerns are being raised within the US that advanced AI regulations may actually give Beijing a cyber warfare advantage. The point that Zhifubai's Z.ai demonstrated performance comparable to Anthropic's model in software vulnerability detection was cited as a supporting example of this concern.

There are also fractures in public opinion. A Pew Research survey found that one-third of Americans perceive that Chinese AI has surpassed the US, indicating that the 'American superiority' narrative maintained by Silicon Valley and Washington is being shaken at the public perception level. There are also fractures within the open-source camp. At the AI4 conference, Western AI researchers such as Geoffrey Hinton, Fei-Fei Li, and Andrew Ng advocated for the necessity of maintaining open-source, which contrasts with the trend of major US labs seeking to strengthen closed approaches due to safety concerns.

3. Key Actors and Positions

Chinese Government and State Mediatreat open-source as a national strategic tool aimed at both technology dissemination and securing normative leadership. The launch of WAICO and the commitment to support developing countries are seen as attempts to build an alternative axis to the Western-centric AI governance order. The Global Times and Xinhua have named the practical governance system implemented by Shanghai as the "Chinese solution" and presented it as a model worth referencing for global AI governance. However, Chinese media themselves acknowledge practical limitations such as "technological walls" and "scene walls" in the field of embodied intelligence, indicating a temperature difference between external narratives and domestic industrial realities.

Chinese AI Companies(Alibaba, DeepSeek, Zhifubai, Moonshot AI, Minimax) have strong practical motivations to secure overseas developers through open-weight releases. In a situation constrained by US export controls on advanced semiconductors, the strategy of leveraging cost competitiveness and openness to increase global adoption rates also serves as a practical means of circumventing sanctions.

US Government and Major AI Labs(OpenAI, Anthropic) maintain a closed attitude in the development of advanced models, citing safety as a justification, but they face internal criticism that this approach may actually provide a reflexive benefit to China. The AI separation strategy through rallying allies reported by Reuters is mocked by Chinese media as an ineffective attempt, and the actual response from allied countries remains uncertain.

US Public Opinion and Academiaexhibit fractured perceptions. The results of Pew Research surveys and analyses from think tanks and media such as Brookings and Foreign Policy reveal a temperature difference from Washington's official optimism. The advocacy of open-source by prominent researchers such as Hinton and Fei-Fei Li shows that there are differing opinions regarding the closed strategy within the US AI community.

Developing Countries and Middle Powersare actors exposed to selection pressure in this competitive structure. The presence of 102 countries participating in WAIC and 29 countries joining WAICO suggests that low-cost open-weight models are being accepted as practical alternatives by countries lacking infrastructure and capital.

4. Key Issues

The first issue is the interpretative difference regarding whether open-source truly signifies 'inclusivity' or is merely a strategic means to secure standard leadership. The Chinese side emphasizes the former, while Western analysts, including those from Brookings, weigh the latter more heavily.

The second is the tension between openness and safety. The spread of open-weight models carries security concerns of uncontrollability, which serves as the justification for US AI labs to adhere to closed approaches, while simultaneously acting as a dilemma that hinders competition in technological dissemination.

The third is the effectiveness of export controls. Many sources commonly point out that US semiconductor and technology export controls not only fail to prevent the spread of China's open-source models but paradoxically promote the development of low-cost, high-efficiency models and accelerate overseas adoption.

The fourth is the gap between public perception and actual technological capability. The changes in perception within the US, as shown by Pew Research, may reflect the results of discourse competition rather than an actual reflection of technological gaps, and this shift in psychological balance itself operates as an independent variable in the standard competition.

II. In-Depth Issue Analysis

China's AI Opening Strategy and Global Standard Competition: The Open Source Narrative of WAIC and Western Anxiety

1. Fundamental Cause Analysis

China's transition to open-source was a structural response rather than a voluntary choice to US export controls. With access to NVIDIA's advanced GPUs blocked, Chinese companies concluded that they could not compete directly with OpenAI and Google using closed proprietary models. Instead, they opted for a workaround by lowering learning costs and publicly disclosing model weights to attract the global developer community to their ecosystem. The low-cost, high-efficiency demonstration of DeepSeek-R1 proved the effectiveness of this approach. The subsequent following of the same path by Alibaba's Qwen, Zhifubai, and Moonshot AI is a result of convergence across the entire industry rather than individual corporate strategy.

Market logic also plays a role. The constraints of needing to produce competitive models with relatively inexpensive computational resources without advanced semiconductors paradoxically aligned with the diffusion strategy of open-weight distribution. A Chinese industry expert quoted by the Global Times explained that open-source and cost advantages are the key drivers leading to overseas developer adoption. While US regulations intended to impose a blockade, they inadvertently provided incentives for China to circumvent through open models.

At the same time, the political calculus is clear. President Xi Jinping's public promise of technology sharing and the launch of WAICO with 29 countries, including Russia, is an attempt to form an alternative axis to the US-centric AI order by encompassing a majority of developing countries. As Kyle Chan from the Brookings Institution pointed out, this is a dual strategy aimed at building an inclusive image while reducing dependence on US technology.

2. Structural Context

Political Structure: The differences in political systems between the US and China underlie this competition. Foreign Policy pointed out that AI technology in China is not subject to social debate and is widely accepted, unlike in the US. In the US, political debates surrounding job replacement, safety, and freedom of expression slow the pace of AI dissemination, while China has a structure that allows for state-led adoption. This acceptance gap leads to a gap in dissemination speed, according to Foreign Policy.

Economic Structure: The existing analysis by EAI indicates that the US and China AI ecosystems have already entered a decoupling path. The US maintains a consistent advantage in talent, infrastructure, and private investment, significantly outpacing China in cumulative private investment from 2013 to 2023. However, the advantage in investment scale does not directly translate to an advantage in dissemination. Instead of falling behind in the competition for closed high-performance models, China has shifted the front by broadening its developer base through open-weight distribution. The performance indicators from WAIC 2026—purchasing delegations from 177 countries and an intended purchase amount of approximately 20.36 billion yuan—demonstrate that this strategy of broadening the base is indeed translating into commercial success.

Security Structure: The security implications of open-weight models are dual-edged. A report from the Wall Street Journal, cited by The Diplomat, raised concerns that US advanced AI regulations may actually hand over a cyber warfare advantage to Beijing. The fact that Zhifubai's Z.ai is on par with Anthropic's model in software vulnerability detection was presented as evidence. Conversely, statements from AI researchers such as Geoffrey Hinton, Fei-Fei Li, and Andrew Ng, reported by TechCrunch, emphasize the safety research and validation functions that openness provides, rather than the risks of uncontrollability inherent in open-weight models. The clash between these two perspectives marks the fracture point in the security discourse surrounding open-source AI.

3. Historical Precedents and Comparative Cases

China's opening narrative resembles a script that has already been demonstrated in the telecommunications infrastructure sector. The way Huawei secured market share by supplying 5G equipment at low prices to developing countries parallels how Chinese AI companies are now distributing open-weight models for free to broaden their developer base, both aiming for standard preemption through low costs. However, while 5G led to infrastructure dependency through the sale of physical equipment, AI open-weight aims for dependency through the diffusion of intangible model weights. The low entry barriers also support a rapid dissemination speed, as evidenced by Qwen's 3 billion downloads within six months.

Similar narrative warfare structures can be found in the US-Soviet competition during the Cold War. At that time, the Soviet Union positioned itself as an alternative to the US-led order by emphasizing technological aid and ideological solidarity with the Third World, which structurally overlaps with how China is now building an image as an "inclusive AI leader" targeting developing countries. However, while the Cold War competition was centered around military alliances, the current competition unfolds in a much more fluid and non-state terrain of developer communities and technological standards. The participation of 29 countries, including Russia, in WAICO indicates that attempts at factionalization are also occurring in this new landscape.

Another reference point is the early experience of the US in leading technological standards on the internet. The precedent of the US seizing TCP/IP and web standards, which shaped the framework of global internet governance for decades, serves as a reference axis for understanding why both the US and China are so sensitive to the diffusion of AI models and the leadership of standards. Once developers and infrastructure settle into a specific ecosystem, the costs of transition increase exponentially, a learning effect that applies to both countries.

4. Key Variables in Issue Development

The first variable is the effectiveness of the US response strategy. The Global Times has already declared the US-led AI alliance concept reported by Reuters as "dead on arrival." Whether this assessment is exaggerated or reflects actual signs of allied countries' withdrawal will be determined in the future by the actual procurement decisions of major countries in Europe and Asia. If the US demands that allied countries restrict the use of Chinese open-weight models, the backlash from developers and companies that have already adopted models like Qwen for cost competitiveness will be crucial.

The second variable is the perception fracture within the West. The Pew Research survey indicates that one-third of Americans perceive that Chinese AI has surpassed their own, suggesting that psychological superiority is being shaken, regardless of the actual technological gap. As this perception spreads, political skepticism regarding the effectiveness of export controls may grow within the US, potentially leading to pressure for regulatory relaxation.

The third variable is whether the security risks of open-weight models materialize. If Z.ai's cyber vulnerability detection capabilities lead to actual misuse cases, the skepticism towards open-source in the Western camp will gain strength. Conversely, if the safety research contributions emphasized by researchers like Hinton, Li, and Ng are highlighted, the justification for regulating open models will weaken. Which way this balance tips will influence each country's policies on the import and utilization of open-weight models.

The fourth variable is the actual choices of the Global South. The fact that 102 countries participated in WAIC and 29 countries joined WAICO is more of a diplomatic gesture. The extent to which these countries actually adopt Chinese open-weight models in their public sectors and industrial infrastructure will be a litmus test for the practical success of China's narrative as an "inclusive AI leader."