Key facts
- Chinese AI models hold majority share in global top 10 by token usage
- Token usage measures real-world deployment, not just benchmark scores
- Finding comes despite US chip export restrictions targeting Chinese AI
- Potential implications for India's AI platform choices and strategic dependencies
Chinese AI models now hold a majority share of the global top 10 by token usage, a concrete measure of real-world deployment, according to data reported by NDTV Profit — a finding that upends the conventional narrative of Western AI dominance.
Token usage tracks how much a model is actually being used to process and generate text, code, and other outputs across applications. It is distinct from benchmark performance or media coverage, making it one of the more reliable indicators of practical adoption at scale. The new data suggests that Chinese models — likely including offerings from companies such as Baidu, Alibaba, ByteDance, and DeepSeek — have achieved mass deployment that exceeds their Western counterparts in aggregate.
The development is particularly striking given the intense US restrictions on semiconductor exports to China, which were explicitly designed to slow Chinese AI progress. That Chinese models are nonetheless leading on token usage suggests either that the restrictions have not had their intended effect at the application layer, or that the sheer scale of domestic Chinese deployment is inflating the numbers — either way, a significant geopolitical data point.
For India, which is currently deciding which AI models and platforms to integrate into government services, enterprise software, and consumer applications, the data raises questions that go beyond pure capability. Relying on Chinese AI models at scale would create dependencies that sit uneasily alongside India's strategic posture toward Beijing — making this a question of technology policy as much as a question of which product works best.