hina’s open-source AI prowess has been a looming threat. Now, US labs may be seeking it out.

On Thursday, WIRED published an analysis of more than 5,000 research papers from the AI conference NeurIPS, which found that about 3% involve collaboration between US and Chinese institutions. The research also found that Meta’s Llama family was a key element in more than 100 papers from Chinese institutions, while Alibaba’s Qwen models were a part of more than 60 papers from US authors. 

The analysis underscores a broader uptick in the popularity of Chinese open source AI models. Earlier this month, the Qwen family of models surpassed 700 million downloads globally on Hugging Face, outpacing Meta’s Llama models on the platform. The use of these affordable models has begun to take root in Silicon Valley, with many choosing to build on Chinese open-source systems rather than on proprietary US-made ones. 

And tech executives at the World Economic Forum in Davos have echoed this sentiment:

• Arthur Mensch, CEO of Mistral, said in an interview with Bloomberg on Thursday that the idea that Chinese AI lags behind that of the West is a “fairy tale,” noting that the region’s open-source tech is “probably stressing the CEOs in the US.”
• And Tencent Senior Vice President Dowson Tong pushed the narrative that an open ecosystem prioritizing interoperability is the most viable path to reaping real benefits from AI. 

The buzz around China’s more efficient, open AI ecosystems stands in stark contrast to the push in the US to develop powerful proprietary systems domestically. The emphasis from both leading US AI firm OpenAI and the government in recent months position the country’s AI dominance as vital and necessary.