Despite strategic competition and political friction between the United States and China in artificial intelligence, analysis of papers from the Neural Information Processing Systems (NeurIPS) conference shows the two countries still collaborate in fundamental AI research. A Wired.com investigation of more than 5,000 NeurIPS papers found that 141 (about 3%) of the works presented involved authors affiliated with both U.S. and Chinese institutions, a consistent share compared with prior years. This collaboration stands out given the broader narrative of rivalry between the two nations in developing algorithms, models, and compute infrastructure.
The study also highlighted how AI technologies and tools travel across borders. The transformer architecture, introduced by researchers at a U.S. lab, underpins hundreds of papers from Chinese institutions. Meta’s Llama family of open models appeared in more than a hundred NeurIPS submissions with Chinese authors, while Alibaba’s Qwen model was included in numerous papers with U.S. contributors. These patterns show that research communities in both countries adopt and build on shared methods, despite political rivalry.
Experts note that part of this collaboration reflects the mobility of talent and long-standing academic networks. Many Chinese-born researchers study or work in American universities before returning home or maintaining ties with colleagues across the Pacific. As a result, personal and professional networks often outlast national policies and contribute to sustained joint research efforts.
NeurIPS itself is seen as a platform that encourages international cooperation. Organizers and participants say that working on shared problems and maintaining professional relationships helps advance the field worldwide. At the same time, political pressures in both countries increasingly challenge openness in research. Still, the presence of joint AI papers suggests that science maintains channels of exchange even as policymakers discuss restrictions and export controls.