How China's open-source AI project is challenging US technological supremacy

How China's open-source AI project is challenging US technological supremacy

China's artificial intelligence industry has rolled out a series of consecutive open-source—or "open-weight"—models this year, beginning in January with the DeepSeek R1 reasoning model and followed by Alibaba's Qwen and others like Moonshot, Z.ai and MiniMax. These models can be downloaded, tuned, and integrated into corporate systems for free, a strategy that was used in an effort to hasten global adoption. In contrast, the majority of US AI champions made their tech proprietary—until OpenAI, under competitive pressure, released its first open-source model, gpt-oss, in August, the Wall Street Journal reported.

The struggle for global standard-setting

The rivalry echoes past tech wars where one standard eventually won out, such as Microsoft Windows in desktop computers or Android in smartphones. In AI, the most advanced tech is not necessarily going to win if someone else's is easier to use, everywhere, and everything. The Trump administration's AI action plan in July warned that Chinese open-source models "could become global standards" and called for development of "leading open models based on American values.".

Stakes commercial and strategic

Open-source AI generates little direct income—development costs are hundreds of millions—but companies that create huge user bases can then leverage complementary products, as Google does for its free Android operating system in advertising, YouTube, and application sales. Chinese government sees open source as insurance against US technology controls and encourages development in AI, operating systems, and chip design. Washington is afraid that Chinese AI dominance can be used as a geopolitical weapon, just as rare earths or US-made chips have been used in the trade war.

Business adoption and usability

Open-source AI appeals to businesses which are comfortable modifying models with sensitive data within the firm. Singaporean OCBC Bank uses about 10 open models like Google's Gemma, Qwen for programming, and DeepSeek for market research. Research firm Artificial Analysis says Alibaba’s latest Qwen model now outperforms OpenAI’s gpt-oss in overall capability, though it’s almost twice the size—making it potentially less efficient for simpler tasks. OpenAI counters that its model excels on reasoning tasks at lower cost, with Amazon Web Services citing its cost-effectiveness over DeepSeek R1 on AWS infrastructure.

Language and cultural advantages

Asian engineers describe Chinese models to work better with local languages and cultural nuances. Japanese software developer Shinichi Usami chose Qwen for a chatbot that went shopping, noting it grasped implicit intent better and replied more politely in Japanese than US counterparts. Chinese models trained on huge volumes of Chinese-language data learn to generalize across close regional languages.

A fierce, survival-of-the-fittest marketplace

China's artificial intelligence sector, already highly competitive in closed-source offerings, is now racing to go open source. Companies are prioritizing user takeup over near-term revenues, hoping to establish "stickiness" before monetizing on apps, cloud platforms, or specialist tools. Most of them will likely disappear in a "Darwinian" cleansing, analysts say, but survivors might become serious international players. As Andrew Ng of DeepLearning.AI pointed out, extreme competition in the initial stage might eventually lead to some of the world's most formidable companies.