Chinese AI models surge in global popularity – and Washington is worried

Chinese AI models have rapidly expanded their share of commercial use in 2026, with usage data from developer gateways showing dramatic shifts: on OpenRouter Chinese models accounted for 57-67% of tokens the week of Sept. 14 (up from 6-13% in February), and on Vercel their share rose to about 55% in August from 11% in January. Firms such as DeepSeek, Z.ai and Alibaba released models that closed performance gaps on tasks like coding and other agentic applications, while being far cheaper per token. Adoption has been strongest in the so‑called Global South (defined by the data provider as 82 countries across Central/South America, Africa and Asia), where two‑thirds of tokens go to Chinese models, though U.S. frontier models still lead benchmarks and attract higher overall spending. OpenAI and Anthropic have responded with lower‑cost models and efficiency improvements of their own.

The shift is drawing scrutiny in Washington: two House committees are investigating commercial use of Chinese models amid worries about national security, technology competition and geopolitical influence. Policymakers have tightened export controls to limit Chinese access to advanced chips and are concerned about remote access to Nvidia hardware and “distillation” techniques that replicate capabilities. Analysts warn that cheap, capable Chinese models could pull businesses and governments into a separate technology sphere of influence, while providers and platforms weigh tradeoffs between cost, performance and regulatory risk.