An action initiative for jointly building a globally open artificial intelligence ecosystem was released at the main forum of this year’s World Internet Conference Wuzhen Summit in China’s Zhejiang Province, setting out 18 proposed actions.
The initiative calls on countries to jointly develop and share computing-power pools and high-quality datasets, to establish seed funds for open-source projects, and to support Global South countries in building localised applications on open-source foundations — the part of the agenda bearing most directly on smaller economies such as Sri Lanka.
Speaking at the summit’s opening ceremony on November 3, Alibaba Group co-founder and chairman Joe Tsai said the company had open-sourced more than 470 Qwen models, and that its ModelScope platform now serves more than 33 million developers across 239 countries and regions, holding over 120,000 open-source models and nearly 23,000 AI applications — around 95 percent of them built by individual developers. Chinese firms are open-sourcing models and lowering access costs as demand for computing power outstrips supply, he said, citing prices as low as one yuan per million tokens.
Those figures came from companies promoting the ecosystem. One independent benchmark exists: an assessment published by Hugging Face earlier in 2026 found models developed in China accounted for 41 percent of downloads on the platform over the preceding year.
Adoption claims cited at the summit included a July cooperation agreement under which Georgia will use Chinese open-source models to build government AI tools, according to Sandro Tsirekidze of its Digital Governance Agency, and a Ugandan non-profit’s use of Qwen models across 68 African languages. Bernd Holznagel, a professor at the University of Münster, said 40 to 50 percent of German small and medium-sized enterprises now use Chinese-developed language models.
The cost and language arguments matter for countries without frontier computing capacity. “Open source means that ‘last-mile’ applications — minority languages and niche vertical domains — will actually have people working on them,” said Liu Dayiheng, who heads Alibaba’s Qwen large language model project. Sri Lanka pressed a related case at July’s World AI Conference in Shanghai, where Deputy Minister of Digital Economy Eranga Weeraratne argued developing nations deserve a direct role in writing global AI rules.
Not independently verified: the percentages and platform figures rest on participants’ statements, not audited data. The Daily Mirror carried the account as a bylined international feature; no other verified newsroom published same-day coverage LankaNewz could date and confirm.
