An analysis said Chinese AI models are shaking U.S. companies’ dominance in the global AI market by leveraging low inference costs.

On Sept. 2, TechRadar reported that research firm Juniper Research said in a recent report that rapid performance improvements and low costs of Chinese AI models are changing developers’ model choices. It said the focus of AI competition is shifting from simply building the best-performing model to how much inference costs needed for real services can be lowered.

The shift is most visible on the AI model brokerage platform OpenRouter. Juniper Research said U.S. companies’ models, including those from OpenAI, Google and Anthropic, accounted for about 70 percent of platform usage last year, but their combined share has now fallen to about 30 percent. OpenRouter’s own data also showed U.S. models accounted for about three-quarters of total token usage in 2025, but usage of Chinese models surged this year and overtook the United States in early June.

The market is also splitting around price and quality. In areas requiring mass processing, such as customer support or content generation, low inference costs matter and Chinese models are becoming more competitive. In contrast, Juniper Research said U.S. cutting-edge models still have an edge in areas requiring high reliability, such as complex reasoning and long-duration tasks, regulated industries and autonomous agent work.

The spread of open-weight models is another variable. Some Chinese models disclose their weights, allowing companies to run them directly on their own servers or local equipment. That broadens options for data control and costs because companies do not have to pay for an external cloud API each time. If this trend continues, it could also put pressure on inference revenue at U.S. AI and cloud companies that have made massive data centre investments, the analysis said.

Still, it is too early to say Chinese models are fully replacing U.S. models. U.S. companies remain strong in markets where performance and reliability matter. Chinese low-cost models and high-performance U.S. models may ultimately coexist depending on use cases, increasing the likelihood that competition in the AI market expands from "who is the smartest" to "who can provide the needed performance at the lowest cost".