Two developments at the recent World Artificial Intelligence Conference (WAIC) in Shanghai sent the same signal. China launched the World Artificial Intelligence Conference Organization (WAICO) to institutionalize its vision of more open and inclusive global AI governance, while Moonshot AI unveiled Kimi K3, China’s most powerful open-weight model to date. Together, they suggest that China is likely to remain committed to open-source and open-weight AI — for now.
WAICO institutionalizes China’s preferred AI governance message: AI should be accessible to all countries, development should be prioritized alongside safety, and global coordination should remain closely linked to the United Nations. While President Xi Jinping’s WAIC address, the 2026 WAIC chair’s statement and the announcement establishing WAICO did not explicitly commit China to open-source AI models, they emphasized openness, inclusion and broader access to AI, especially for developing countries.
This approach reflects China’s broader AI development strategy. Open models allow Chinese companies to attract developers, accelerate industrial applications and expand overseas adoption, especially in countries that lack advanced AI capabilities. More importantly, they enable Beijing to internationalize its domestic AI strategy by extending China’s open-source ecosystem beyond its borders. Through application-driven innovation, “AI Plus” initiatives and AI-enabled industrial upgrading, China is seeking to build an ecosystem around its own technologies while promoting AI as an international public good rather than an exclusive technology controlled by a handful of countries. In doing so, Beijing advances its longer-term objective of expanding global AI infrastructure and increasing the international reach of Chinese AI.
Rather than serving simply as another international AI forum, WAICO institutionalizes China’s strategy by providing China with a permanent platform to shape global AI governance. Drawing on China’s previous experience creating international organizations, such as the Asian Infrastructure Investment Bank and the New Development Bank, WAICO appears designed to combine institution building with practical cooperation. China is likely to provide much of the financial support for capacity building, technical assistance and AI infrastructure projects in participating countries, particularly across the Global South.
This open-source diplomacy, however, is unlikely to persuade established Western powers, which remain committed to smaller, values-based AI governance frameworks. The geopolitical alignment is difficult to miss. In this respect, China’s message of openness and public goods resembles Xi’s earlier defence of free trade at Davos days before Donald Trump’s inauguration in 2017: rhetorically attractive, but received in the West with muted responses and considerable skepticism.
Compared with initiatives such as the Global Partnership on AI (GPAI) and the Council of Europe Convention on AI, WAICO stands out in three respects: its emphasis on AI capacity building for developing countries, its promotion of openness and open-source ecosystems, and its close alignment with the United Nations. Together, these features reveal China’s broader geopolitical ambitions. By advocating openness, opposing what it describes as the over-securitization of AI, and promoting a UN-centred governance framework, Beijing seeks to position itself as a champion of equitable access to AI while implicitly criticizing US-led export controls and technology restrictions. Rather than simply participating in existing AI governance frameworks, WAICO advances a competing vision — one that prioritizes technology diffusion, development and the interests of the Global South.
WAICO’s membership reinforces this geopolitical reading. Unsurprisingly, its 29 initial signatories are exclusively non-Western states, spanning Russia, Brazil, Indonesia and Pakistan alongside other members of BRICS, the Shanghai Cooperation Organization, as well as 10 African countries. The absence of Western members suggests that WAICO may become not so much a universal AI governance platform as a China-centred coalition focused on development, infrastructure and resistance to Western technology controls.
WAICO’s near-term priorities translate this broad rhetoric into a more concrete institutional strategy. The organization will focus on capacity building — including Xi’s pledge to provide 5,000 training opportunities for developing countries and to establish AI cooperation centres with the Association of Southeast Asian Nations, the African Union, BRICS and other regional bodies — while supporting an open-source ecosystem designed to extend China-centred AI infrastructure abroad and anchoring its governance agenda in UN-centred multilateralism. Together, these priorities reflect China’s view that development deserves as much attention as safety, a position it also advanced through the AI + International Cooperation Initiative proposed at the UN Global Development Initiative high-level meeting in September 2025.
This UN alignment is already visible. With the UN Secretary-General António Guterres attending and endorsing the event, WAICO is likely to promote goals that overlap with the UN agenda, including AI for good, human-centred AI. The alignment was also reflected in the China-backed UN resolution on strengthening international cooperation in AI capacity building at the 78th UN General Assembly in 2024, as well as in China’s proposed “AI Capacity-Building Action Plan for Good and for All” released in September 2024. Both prioritize global AI connectivity and digital infrastructure, especially in the Global South, and the application of AI to empower various industries.
Open Weights as Business Strategy
Yet institutional initiatives alone do not explain why China continues to embrace openness. Equally revealing are the decisions made by China’s leading AI companies. The release of Kimi K3 and DeepSeek’s continued commitment to open-source development provide perhaps the clearest evidence that openness remains a strategic choice for China’s AI industry — not merely a political slogan. For now, China appears likely to maintain both the concept and practice of open source, but the durability of this commitment depends on whether Chinese firms continue to see openness as strategically useful once the technology gap narrows.
Kimi K3, released during WAIC, is one of China’s most powerful AI models and further narrows the gap with leading US models such as Claude Fable 5 and GPT 5.6 Sol. Despite the exaggerated claims that Chinese AI models are overtaking their US counterparts, Kimi has acknowledged that K3 still lags behind leading American models. More importantly, Kimi announced that K3 would remain open weight.
DeepSeek appears to be following the same path. According to the recent comments by founder Liang Wenfeng, the company intends to continue pursuing open-source development, potentially including its most advanced models. Liang argues that openness does not weaken DeepSeek’s competitive position because releasing model weights does not eliminate the formidable technical and financial barriers to deploying frontier AI systems. The barrier to entry remains extremely high even when a model is open source: deploying it effectively is difficult and running it at very low cost is even harder. After all, deploying frontier AI systems effectively and at low cost requires infrastructure, engineering and computing resources that cannot simply be copied by downloading an open model.
Liang’s view provides an important corporate-level counterpart to China’s official rhetoric. Open source helps Chinese firms expand adoption, strengthen ecosystems and compete with leading US companies without necessarily sacrificing commercial returns. Liang argues that open source and commercialization are not mutually exclusive: DeepSeek can support competitors and community deployment while still preserving advantages in cost control, engineering and operational efficiency. The AI market is sufficiently large that no single company can dominate it, making ecosystem expansion more valuable than maximizing short-term licensing revenue.
For now, open source serves China’s strategic interest on every front. While China is still catching up, openness helps diffuse Chinese models, build ecosystems, reduce dependence on Western platforms and gain support from developing countries. But if Chinese firms further close the gap or achieve leadership, they may have stronger incentives to protect model weights, talent and intellectual property.
Yet China’s openness has limits. Beijing is also working to protect AI-related intellectual property, talent and technologies. Reports of efforts to prevent talent from leaving the country, the forced cancellation of Meta’s proposed $2 billion acquisition of Manus, and Financial Times’ reporting on export controls restricting foreign access to China’s most advanced AI model weights all point in the same direction. If China achieves, or comes close to achieving, AI technological leadership, the incentives may shift from diffusion to protection.
