The geopolitical battleground for artificial intelligence dominance is shifting from proprietary, closed-door systems to the open-source arena, and China is rapidly gaining ground. According to a recent analysis by MIT Technology Review, Chinese open-weight AI models accounted for 17.1% of global downloads on the Hugging Face platform in the year ending August 2025. This marks the first time China has surpassed the United States, which held a 15.86% share, in this critical metric of global AI influence. The surge is driven by leading Chinese labs such as DeepSeek, Moonshot, Alibaba (Qwen), MiniMax, and Zhipu aggressively releasing their models as open-weight packages.
This strategy of open-sourcing powerful foundation models went mainstream following DeepSeek’s release of its R1 model in January 2025. By making the weights freely available, Chinese developers have effectively bypassed traditional commercial and geopolitical barriers, allowing their technology to proliferate rapidly, particularly across the Global South. Founders and developers from Nairobi to São Paulo are increasingly building their applications on Chinese foundations, drawn by the combination of high performance and accessibility.
(Related: Alibaba Kills the Free Tier After Building the World’s Most Downloaded Open-Source AI)
The Strategic Value of Open Weights
The proliferation of Chinese models is not merely a matter of developer preference; it carries significant strategic implications. Alibaba’s Qwen family of models currently boasts the highest number of user-generated variants on Hugging Face, surpassing the combined totals of Google and Meta as of last month. This widespread adoption embeds Chinese technological standards and architectures into the global AI ecosystem.
The impact is evident in national-level AI initiatives. Singapore’s AI Singapore program recently selected Alibaba’s Qwen over Meta’s Llama as the foundation for its latest regional model, citing superior performance in Southeast Asian languages. Similarly, Malaysia announced last year that its sovereign AI ecosystem would run on DeepSeek’s architecture. These decisions highlight how open-source models are serving as a potent tool for Chinese soft power, establishing technological dependencies in key emerging markets.
However, the export of these models also exports certain domestic constraints. Chinese models carry a distinct content moderation imprint, having been trained to avoid generating outputs that conflict with government policy. While this may be less relevant for enterprise applications or coding tasks, it raises questions about the cultural and political biases embedded in the AI systems increasingly powering the Global South.
A Workaround for Chip Restrictions
Beyond geopolitical influence, the open-source strategy serves as a crucial workaround for US export controls on advanced semiconductors. By releasing their models to the global developer community, Chinese labs benefit from external feedback, bug fixes, and performance optimizations. This crowdsourced refinement helps compensate for the constrained computing power available domestically, allowing Chinese developers to iterate more rapidly than they could in isolation.
(Related: The US-China AI Science Split Is Accelerating, and Open Source May Be the Last Bridge Left)
This dynamic has not gone unnoticed in Washington or Silicon Valley. In February 2026, Anthropic accused several Chinese labs of illicitly extracting capabilities from its Claude model through a process known as distillation. The allegation underscores the porous nature of the open-source ecosystem and the difficulty of containing AI advancements once they are released into the wild. As Chinese models continue to gain traction globally, the US faces a complex challenge: how to maintain its technological edge without stifling the open collaboration that has historically driven AI progress. It should be noted that some Chinese players are now shifting from open to closed-source models. For example, EastFrontier recently reported that Alibaba is closing the door and killing the free tier after building the world’s most downloaded open-source AI Qwen.