The global AI landscape is witnessing a significant shift, as Chinese models have now surpassed their U.S. counterparts in weekly usage for the fourth consecutive week.
This isn't just a small lead; it's a structural change. In the fourth week of May 2026, data from the AI model router OpenRouter showed that Chinese models accounted for nearly 32% of global token usage, while U.S. models stood at just over 17%. The total volume of tokens processed reached a new all-time high, indicating that the entire market is growing rapidly, with Chinese models capturing the lion's share of that new growth.
The primary catalyst for this surge is the incredible price-performance ratio of new Chinese models. First, the standout model is DeepSeek-V4-Flash. Released in late April, it has quickly become the most used model globally. Its pricing is a game-changer—at roughly $0.125 per million tokens, it's nearly 20 times cheaper than some competitors. This ultra-low cost makes it highly attractive for developers building large-scale applications, directly stimulating massive adoption.
Second, this isn't just about cheap tokens; it's about where they're being used. The growth is fueled by a burgeoning ecosystem in China. AI is being deeply integrated into everyday applications, from AI agents like OpenClaw that automate complex tasks to "super-apps" like WeChat. This creates a constant, high-volume demand for AI processing that goes beyond simple chatbot demos, turning into real, recurring usage.
Third, there's a crucial hardware story unfolding in the background. As the U.S. tightens export controls on advanced chips, Chinese tech companies have accelerated their shift to domestic hardware, like Huawei's Ascend processors. The fact that models like DeepSeek V4 are optimized for these homegrown chips is significant. It reduces reliance on foreign technology, mitigates supply chain risks, and allows for more predictable and cost-effective scaling. This self-reliant supply chain provides a stable foundation for the entire ecosystem's growth.
In essence, the recent dominance of Chinese AI models is not a fluke. It's the result of a multi-pronged strategy combining aggressive pricing, deep ecosystem integration, and a strategic pivot to hardware independence. The trend started months ago and has now solidified into a new market reality.
- Token: The basic unit of data processed by an AI model. For text, one token is roughly equivalent to 4 characters or 0.75 words.
- OpenRouter: A platform that acts as a "router" for AI models, allowing developers to access and switch between various models from different providers through a single interface.
- AI Agent: An autonomous program that can perceive its environment, make decisions, and take actions to achieve specific goals, often by using other tools or AI models.