Of 'open weights' and unbeatable prices: how Chinese AI is advancing across the Global South while the West debates
🕒 Published on Zendoric: June 26, 2026 · 09:00
A report by The Wire China portrays Beijing's strategy to make its AI models prevail first in Southeast Asia and the Global South. The figure that sums up the shift: the U.S. share of downloads on Hugging Face fell from 60% to 16% in four years. The battle isn't being fought on capability alone, but on cost and accessibility.
The report by Alex Colville in The Wire China on June 21, 2026, describes with precision a technological offensive that often goes unnoticed in the West. The opening scene sums it up well: the corridors of Jakarta's airport plastered with Huawei and Alibaba ads. This is not improvised marketing, but the visible face of a strategy articulated from the top of the Chinese Communist Party.
The central figure is striking. According to the article, four years ago U.S. models accounted for 60% of downloads on Hugging Face, the leading open-model platform; by November 2025 that share had fallen to 16%. The driver of the collapse is economic: against the closed code and subscriptions of Anthropic or OpenAI, the Chinese models from Alibaba (the Qwen family) and DeepSeek can be used for free or through an API at a fraction of the price. The remark from an Indonesian developer quoted in the text condenses it: even granting that Claude is of better quality, 'they will always choose the cheapest one.'
A technical clarification the report itself underscores is worth making, because it helps frame the debate without falling into simplifications. Talking about 'open source' is imprecise: the correct term is 'open weights,' because Chinese companies publish the parameters but not the training data or other core information. The difference matters, and yet Chinese licenses tend to be more permissive about modifying the models, something highly valuable for developers in emerging countries.
The geopolitical dimension is explicit in the sources the article cites. Xi Jinping spoke in 2018 of the 'lead goose effect'—whoever heads the flight sets the direction—and the report draws a parallel with the old Maoist strategy of 'surrounding the cities from the countryside': China aims for its AI to predominate by advancing first through the periphery. The new Five-Year Plan, adopted in March 2026, and the State Council's 'AI+' policy, which argues that AI should not be 'a game for rich countries and people,' provide institutional backing for that narrative.
The account, however, is not one-directional, and the article has the merit of not hiding it. Qwen's latest models have moved to closed code with a subscription, MiniMax has published licenses designed to capture the most profitable uses, and only labs such as Z.ai, Moonshot AI and DeepSeek are for now maintaining their open bet. It is the classic tension between commercial imperative and geopolitical discourse, and it is best to attribute each claim to its sources without assuming which way the balance will tip.
The sensible conclusion is not alarmist but strategic. The competition is shifting from the purely technical terrain—where the parameter scale the article attributes to models like DeepSeek V4 is impressive—toward the terrain of cost, licensing and physical infrastructure, such as the dozens of availability zones that Alibaba, Huawei and Tencent say they have deployed in the region. For the countries of the Global South, more supply and lower prices are good news in terms of access. The open question, legitimate and still unanswered, is what standards and dependencies will accompany that adoption over the long term.