The global contest over artificial intelligence shifted from a race of models to a contest of governance structures this week, as China opened its annual World AI Conference in Shanghai by launching a 29-nation intergovernmental body and releasing what it called the world’s largest open-weight AI model. The World Artificial Intelligence Cooperation Organisation (WAICO), founded with members including Indonesia, Brazil, South Africa and Pakistan, positions itself as a multilateral alternative to Washington’s AI Opportunity Statement, which has drawn 35 mostly higher-income countries. On the same day, Beijing-based Moonshot AI released Kimi K3, a 2.8-trillion-parameter model that, in early benchmarks, matched or exceeded the coding and agent-task performance of second-tier US proprietary systems from Anthropic and OpenAI, though it still trails their frontier models.

The dual announcements crystallise a strategy that analysts in European capitals describe as “AI communism” in distribution: open-weight models that can be freely downloaded, modified and run on local servers, lowering cost and dependency on American cloud APIs. While US labs keep their most capable models closed, citing safety and commercial reasons, Chinese firms are making large-scale systems available to developers worldwide. The full Kimi K3 weights are to be released globally by the end of July. The approach is already reshaping market expectations: investors in Silicon Valley and on Wall Street are reassessing whether proprietary US models can sustain premium pricing if open alternatives close the capability gap, a dynamic reminiscent of the DeepSeek shock of early 2025.

Viewed from Jakarta, the WAICO launch is a diplomatic win. Indonesia, represented by Coordinating Minister Airlangga Hartarto, secured a commitment to discuss a role in the secretariat and used the bilateral meetings to advance trade access, investment in a 100 GW solar programme, and the Two Countries Twin Parks industrial scheme. For Global South nations, China’s promise of 5,000 AI training projects and open-weight models offers a pathway to build local AI capacity without the compute and licensing costs of US systems. Yet the openness has clear boundaries. Domestically, new regulations that took effect last week ban AI companions from encouraging emotional dependence and prohibit such services for minors, forcing ByteDance, Alibaba and Tencent to disable personalised chatbot features. The crackdown, aimed at reversing a declining birth rate by pushing citizens back toward real-world relationships, underscores that Beijing’s open-source push abroad coexists with a tightly controlled information environment at home.

The next factual milestones are the full public release of Kimi K3’s weights at the end of July and the institutionalisation of WAICO’s secretariat, where Indonesia is seeking a formal post. In Washington, the Trump administration faces a choice: whether to impose regulatory friction on the use of Chinese open-weight models by US firms, as some former officials have urged, or to rely on the still-superior performance of proprietary American systems. The answer will shape whether the AI landscape fragments into two distinct regulatory and technological spheres.