Chinese AI Captures 46% of US Tokens: An Economic Shock

A Rapid Rise of Chinese AI Models

Artificial intelligence models developed in China have recently captured a significant share of the enterprise token market in the United States. According to a survey conducted by CNBC, published on July 7, these models reached 46% of token usage in just one week this summer. This figure marks a dramatic increase from the 4.5% recorded eighteen months ago. Over a twelve-month period, the average stands at 11%. Since February 8, 2026, Chinese models have consistently accounted for at least 30% of the enterprise token volume on OpenRouter, the world's largest neutral LLM router.

The Economic Reasons Behind This Shift

The rise of Chinese models can primarily be attributed to economic reasons. These models offer significantly lower entry and exit costs, making them attractive for companies looking to optimize their spending. Despite the reduced costs, the performance of these models is deemed "good enough" for many practical applications. Among these applications are data extraction, content summarization, augmented writing through retrieval, and agent integration. Usage data at the router level and examples of adoption by companies illustrate this trend. Furthermore, benchmarks indicate that Chinese models are approaching the performance of American models, but at a fraction of the cost.

Security and Compliance: Concerns to Manage

The growing adoption of Chinese AI models raises concerns regarding security and compliance, particularly concerning third-party hosting of services. However, companies can mitigate these risks. One solution is to run open-weight models on managed APIs hosted in the United States. Companies can also utilize hyperscalers or their own infrastructure, ensuring that tokens are not sent to servers located abroad.

A Five-Minute Migration Strategy

To facilitate the transition to these AI models, a practical five-minute migration approach is proposed. It includes several key steps:

• First, conduct A/B testing on prompts via OpenRouter to evaluate performance.
• Next, set up routing with a fallback option, maintaining the quality of flagship models as a safety net for complex cases.
• Adjust routing percentages based on the results of the evaluations.
• Finally, choose model options considering specific workloads and regulatory constraints.

This methodology allows companies to leverage the economic advantages of Chinese models while minimizing potential security risks.