Days before a contentious debate in the American AI industry over the Chinese Kimi K3 and other open-weight models, thousands of world leaders, technology executives and AI researchers gathered at the Palexpo convention center in Geneva for the United Nations’ AI for Good summit. The event, hosted by the UN agency that sets international technology standards, combined exhibition halls and public displays with quieter diplomatic conversations about the governance and global distribution of AI.
Notably, leaders of US frontier labs were largely absent; one of the most visible American executives at the conference was Marc Benioff, CEO of Salesforce. Instead, a substantial Chinese delegation of government officials and corporate executives used the four-day forum to argue that free or low-cost Chinese open-source models will be the practical choice for many countries.
Wang Jian, a prominent Chinese computer scientist, former Microsoft Asia executive and chief architect of Alibaba’s cloud business who now runs a government-backed AI research institute, told Semafor that China’s AI can serve as a "resource" for other countries similar to energy: offering choice matters for the rest of the world. Wang spoke in a quieter room under a banner reading "Group of Friends of Global Governance," where Chinese officials and executives engaged ministers and delegations from Pakistan, Russia, Zambia and other Global South countries.
That Chinese-led coalition, formed last year, was presented as a counterweight to US and European influence. Delegates emphasized that open-source AI could support developing economies by providing lower-cost access to advanced models and by offering training on their use. The implicit pitch: as the United States retreats from parts of the multilateral system, China can present itself as a more dependable partner.
Observers see the approach as a software-centric version of China’s broader diplomatic playbook. Just as cheaper Chinese solar panels and electric vehicles helped shape global market dynamics and standards, Beijing is distributing its AI models widely, offering them at lower costs than many Western competitors and promising capacity-building abroad.
This shift in dynamics comes amid a broader move by some countries to seek technological independence from the United States, a trend that intensified after recent US export controls on Anthropic’s Mythos and Fable models disrupted non‑US companies and governments that relied on those frontier models. A week after the Geneva summit, Chinese leader Xi Jinping emphasized open source and global cooperation at the World AI Conference in Shanghai, urging opposition to an overly broad national security framing in AI and pledging to align global rules and technical standards through a new 29‑country World AI Cooperation Organization. The United States is not a member of that China-led group; Washington has convened its own coalition, Pax Silica.
The US position has stressed selective sharing with trusted allies and protecting American intellectual property, with Treasury Secretary Scott Bessent warning that "open source is not open season on American IP." A US State Department spokesperson told Semafor that China’s approach often follows a pattern of initial partnership promises that lead to dependency on Chinese infrastructure and the adoption of substandard technical standards.
Some officials remain cautious about relying on Chinese models. Nkundwe Mwasaga, director general of the ICT Commission in Tanzania (a country whose national fiber network was built by Chinese firms with Chinese state loans), said America still "leads the pack," though China is "just catching up." Mwasaga is developing a sovereign AI model that local startups can build using either Chinese or American models.
Experts at the conference also raised concerns about transparency. Many Chinese open-source models have not disclosed their training data, and studies have shown that some models can reflect Beijing‑aligned framings on sensitive issues. That can be problematic for agencies seeking neutral technical assistance — for example, an economic agency researching semiconductor policy might worry that a model’s responses reflect Chinese perspectives on Taiwan’s chip industry.
Not all actors are wary: Singapore and Saudi Arabia’s Aramco, among others, appear less concerned about adopting Chinese models for certain localized uses. Some analysts note that Western models also embed cultural assumptions and that the availability of open weights in Chinese models allows users to adapt or adjust content restrictions.
China’s open-source AI diplomacy is part of a broader effort to build an alternative global order and inscribe its values into digital infrastructure. But analysts say that for Chinese firms to secure long-term international adoption, they will need to address trust and transparency concerns. Meanwhile, US companies such as Microsoft, Google and Meta remain active globally, and the Pax Silica initiative has attracted allies aiming to reduce dependence on China.
What to watch
- China is promoting open-source models and lower-cost access as a way to gain market share and influence standards abroad.
- Xi Jinping’s Shanghai speech framed open source as a historic opportunity and opposed overbroad national security restrictions in AI.
- US export controls and IP protections are pushing some countries to seek alternatives to American-provided models.
- Transparency about training data and model behavior remains a central unresolved issue influencing which countries rely on which providers.
The global competition over AI governance and supply chains continues: China’s open-source push offers rapid distribution and affordability, while questions about transparency and strategic dependency will shape how countries choose partners in the coming years.
