John Thornhill and Caiwei Chen debate the trajectory of artificial intelligence power between the United States and China, grounding the discussion in publication, patent, talent, and model data. Citing the Stanford Artificial Intelligence Index Report 2025, Thornhill notes that by 2023 China accounted for 22.6 percent of citations versus 13 percent for the US and 69.7 percent of AI patents. The US still dominates the top 100 most cited publications, with 50 entries versus China’s 34 in 2023, but that lead is narrowing. Talent distribution shifted markedly between 2019 and 2022: the share of top researchers in the US fell from 59 percent to 42 percent while China rose from 11 percent to 28 percent. Thornhill warns that visa restrictions for H-1B holders may accelerate researcher returns to China, altering the talent balance further.
On technology and deployment, both writers emphasize different strengths. The US produced 40 of the world’s most notable models in 2024 compared with 15 from China, yet Chinese teams have optimized for efficiency and open-weight releases. Examples cited include DeepSeek-V3 and Alibaba’s Qwen 2.5-Max, which outperform some US counterparts on algorithmic efficiency; DeepSeek-V3’s training used 2.6 million GPU-hours. Export restrictions on top GPUs have pushed Chinese labs toward gray markets and repair strategies, but they have also encouraged pooling compute, releasing open-weights, and building competitive multimodal and video models. Air Street Capital data referenced in the conversation shows China has overtaken the US in monthly downloads of AI models, and China already leads in AI-enabled fintech, e-commerce, and logistics. The authors highlight hardware and embodied Artificial intelligence as fertile ground, where China’s manufacturing and robotics expertise could translate into applied advantage.
Policy, education, and social dynamics shape how advantage may convert into lasting influence. Caiwei points to local governments and major enterprises rapidly rolling out reasoning models, and to national plans to embed Artificial intelligence literacy across universities and school curricula. Stanford HAI’s 2025 index found Chinese respondents the most optimistic about Artificial intelligence’s future. Concerns about social control and repression are raised by Thornhill and cited analysts, but Caiwei also notes a new generation of founders who are transnational in outlook. The exchange ends on the point that winning is not just speed or research volume but how widely and deeply technologies spread, and that the open-source versus proprietary model debate-acknowledged even by Sam Altman-remains a key subplot in the US-China competition.
