Chinese AI Models Have Led Global Volume for 21 Weeks
Data shows Chinese model usage outpaced US counterparts in total token volume for over five months.
Updated on Sept. 21, 2026 in Artificial Intelligence
Live Poll
Does the global growth of Chinese AI models make you more concerned about technology leadership?
Global AI large model call volume hit 129 trillion tokens for the week of September 14-20, 2026. Chinese models accounted for 67.46 trillion tokens, maintaining a lead over US models for 21 consecutive weeks.
Why it matters
The consistent volume gap highlights a significant shift in AI model utilization patterns. This divergence reflects a 10.28 percent increase in Chinese model usage compared to a 34.7 percent decrease in US model volume over the period.
DeepSeekV4.1-Flash captured the top position for individual model performance, reaching 15.8 trillion tokens in weekly volume. This total volume for the model reflects a 219 percent increase, contributing to the global 1.57 percent week-over-week rise in AI model call volume.
The players
DeepSeek
An AI research firm focused on developing large language models and high-efficiency inference stacks.
The details
Call volume represents the total number of tokens processed by large language models during specified operations. Chinese models have sustained higher usage levels for 21 weeks, while US model volume saw a sharp contraction of 34.7 percent during the same reporting interval. The data points specifically to DeepSeekV4.1-Flash, a large language model designed for high-throughput inference, as the primary driver for this shift.
Timeline
September 14-20, 2026: Global AI large model call volume was measured at 129 trillion tokens.
The Tech Race
This performance data underscores the shifting landscape of global AI adoption, where usage patterns in Asia have moved to outpace established Western platforms. The sustained 21-week lead for Chinese models marks a departure from the historical dominance of US-developed models in aggregate token processing.
Developers and organizations relying on large language model infrastructure should note the significant volatility in US model usage compared to the rising trend in Chinese alternatives. Those maintaining cross-regional AI workflows may need to evaluate the stability and availability of these high-volume models for future integration.
The takeaway
The sustained volume lead held by Chinese AI models suggests a fundamental change in where the world's primary model compute is concentrated. Analysts should watch for the next weekly volume report to see if the 34.7 percent decline in US model usage is a temporary adjustment or a persistent trend.
Further reading
For more on the current trajectories in generative technology, see Artificial Intelligence.
Source note: This article includes information reported by Metal.
Live Poll
Does the global growth of Chinese AI models make you more concerned about technology leadership?
