TechFlow reports, July 21, OpenRouter data shows that the share of Tokens used by US enterprises from Chinese models has risen from less than 10% a year ago to nearly 60% in early July, peaking at 63%, with DeepSeek being the main driver. This means Chinese models are no longer just for developers to try out, but are entering the actual workflows of US enterprises by virtue of low prices, open weights, and sufficiently strong programming and agent capabilities.
However, this does not mean Chinese models have captured 60% of the US enterprise AI market. The statistics in the chart represent the number of Tokens on the OpenRouter platform, excluding traffic from enterprises directly calling vendors such as OpenAI and Anthropic; meanwhile, low-price models are more likely to undertake large-volume, long-chain tasks, which naturally amplifies the Token share.
This also means enterprises are reserving high-end models for complex tasks, assigning a large amount of standardized work to Chinese models that are "capable enough and lower cost," and AI competition is shifting from single model rankings to multi-model orchestration and cost per unit task.




