Chinese models reach a majority of tokens on OpenRouter
On OpenRouter, Chinese models processed between 57% and 67% of OpenRouter traffic during the week beginning 14 September. The February range was 6% to 13%. Vercel separately reported a rise to 55 percent of its model use in August from 11 percent in January. These platform figures show a shift in developer use, while spending remains weighted toward high-end models from US labs. The platforms do not measure all AI activity.
Artificial Intelligence··Evening
OpenRouter counts a majority of Chinese-model tokens
On OpenRouter, Chinese AI models handled between 57% and 67% of token traffic during the week beginning 14 September, according to platform data given to CNBC. February’s lower bound was 6% and its upper bound 13%. OpenRouter connects developers to models from many providers, so the figures describe activity on one routing platform rather than all use of artificial intelligence. About half of the tokens on that service came from US firms. The reported range also includes firms in Europe and countries that OpenRouter groups as the Global South.[1]
Vercel reports its own shift
Vercel told CNBC that Chinese models made up 55 percent of its model use in August, up from 11 percent in January. Its measure is separate from OpenRouter’s token count and uses a different period, so the two percentages should not be added or read as a single market share. Among firms OpenRouter places in the Global South, the Chinese-model token share was 67 percent. OpenRouter’s Peter Walker said new Chinese releases can do demanding coding work for less than many models from US labs; that is his assessment of the products, not an independent performance test.[1]
Spending still favors US frontier models
The rise in token share has not made Chinese models the leaders in spending. CNBC reported that high-end models from US labs still draw most expenditure, even as cheaper alternatives handle more tasks on the measured platforms. Vercel’s Harpreet Arora said price matters after a model clears the quality threshold for a job, while firms still choose US frontier models for harder work. CNBC also reported that two US House committees are examining the rise of Chinese models. Their inquiry is a political response to the usage change, not a measurement of model capability.[1]
