American companies are increasingly switching to Chinese AI models due to the sharp rise in service costs from leading Western providers such as OpenAI and Anthropic.
What Happened
The share of tokens consumed by Chinese models via the OpenRouter platform has exceeded 30% per week since February 2026, reaching peak values of 46%. The main players are DeepSeek V4, Alibaba's Qwen, and Z.ai's GLM 5.2, which cost users 60–90% less than their American counterparts.
Context
The technological gap between top American systems and Chinese developments has narrowed to 6–9 months. The market is seeing a paradigm shift from a "tokenmaxxing" strategy (using the most powerful models for all tasks) toward intelligent request routing to cheaper but efficient solutions.
Why It Matters for the Industry
A global shift is occurring toward multi-model architectures and the development of open weights solutions in China. This creates significant competitive pressure on American developers and stimulates the development of model routing technologies that select models based on task complexity and available budget.
Why It Matters for Users
Developers and startups should consider Chinese models, such as DeepSeek and Qwen, for radical optimization of infrastructure costs. Switching to such solutions allows for a multi-fold reduction in operating expenses (OpEx) with almost no critical loss in task performance quality.
What Remains Unknown / Limitations
There is a difference in risk assessment: while technical specialists focus on cost savings, corporate architects point to potential challenges in compliance and data security when using Chinese providers.
Sources
Author
Look at AI, Editorial Team