Media in the U.S.: The adoption rate of Chinese AI models among U.S. enterprises is rapidly rising, primarily due to sharply increased usage costs of top-tier American models such as those from OpenAI and Anthropic.
According to data from the OpenRouter platform, the share of tokens used by U.S. enterprises on Chinese AI models has exceeded 30% weekly since February 8, peaking at 46%, compared to an average of only 11% over the previous 12 months and just 4.5% in the first half of 2025.
Recently released models from Chinese companies like DeepSeek and Z.ai - Advanced AI Chatbot & Agent powered by GLM-5.2 have performance levels approaching those of leading U.S. models—only lagging by 6 to 9 months—while being 60% to 90% cheaper.
For example, the AI startup Lindy has shifted all its traffic from Anthropic’s Claude to DeepSeek, expecting to save millions of dollars within a few months; the GLM 5.2 model behind Z.ai - Advanced AI Chatbot & Agent powered by GLM-5.2 saw a 27-fold increase in daily token volume and an 80-fold rise in customer count within its first week on the Vercel platform.
Meanwhile, the U.S. government is intensifying regulation on the most powerful AI models. OpenAI has already restricted the rollout of new models upon government request, and Anthropic's export controls were tightened before being later lifted. To manage costs, enterprises are increasingly routing non-core tasks to more cost-effective Chinese open-source or open-weight models, creating a new competitive landscape characterized by “nearly equivalent performance, drastically reduced costs.”
Original source: toutiao.com/article/1870064104427532/
Disclaimer: This article represents the personal views of the author(s).