Driven by soaring costs, U.S. companies turn to Chinese AI models
As the pricing of models from leading American AI manufacturers continues to rise, more and more American companies are beginning to adopt Chinese artificial intelligence models. These open source and open weight models from China are close to the top cutting-edge products in the United States in terms of performance, but the cost of use is 60% to 90% lower. They cover the vast majority of conventional AI tasks and are therefore favored by enterprises. According to statistics from the AI model aggregation platform OpenRouter, since February 8 this year, the proportion of words used by American companies to call Chinese AI models has exceeded 30% every week, reaching a peak of 46%; while the average of the previous 12 months was only 11%, and it was as low as 4.5% in the first half of 2025. Models recently released by Chinese companies including DeepSeek and Zhipu are widely considered to be highly competitive with leading American systems such as Anthropic and OpenAI. US AI startup Lindy switched all its traffic from Anthropic's Claude model to DeepSeek in June. The company's CEO said that the move will save millions of dollars within months, and the performance of many core use cases has not dropped but has improved. Industry data shows that in the first week of its launch, the latest large model GLM 5.2 released by Zhipu saw an average daily word element call volume increase of 27 times and the number of customers increased by 80 times, making it the fastest model to be launched on relevant platforms in 2026. Kyle Chan, a researcher at the Brookings Institution, pointed out that as the cost of AI soars, Chinese AI models are particularly attractive to American companies. In the past, companies prioritized the adoption of AI regardless of model costs, but now they are more price-sensitive. Analysts estimate that Chinese models are now about six to nine months behind top U.S. rivals, but typically cost a "fraction" of the latter.