Chinese AI models gain popularity among U.S. companies due to lower costs

Chinese AI models accounted for a much smaller share of usage

Chinese artificial intelligence models are gaining popularity among U.S. companies as businesses look for cheaper alternatives to expensive AI systems from leading American firms.

Recent models from Chinese companies such as DeepSeek and Z.ai are competing closely with advanced AI systems developed by U.S. companies including OpenAI and Anthropic.

The growth comes as companies face rising costs for using advanced AI models. Many businesses are now focusing on affordable solutions that can deliver strong performance.

Data from OpenRouter, a platform that gives developers access to multiple AI models, shows increased use of Chinese AI systems.

The share of tokens used by U.S. companies on Chinese models has remained above 30% each week since February and reached as high as 46%.

Before this increase, Chinese AI models accounted for a much smaller share of usage. The average over the previous year was about 11%, while usage dropped to 4.5% in the first half of 2025.

Experts say lower prices are a major reason behind the growing demand. Kyle Chan, a fellow at the Brookings Institution’s John L. Thornton China Center, said companies are becoming more careful about AI spending as costs continue to rise, “Companies are now becoming more cost-conscious,” Chan said.

Many developers are also exploring open source and open weight AI models because they allow more flexibility. These systems give users access to parts of the technology, allowing them to inspect, adapt and use the models for different tasks.

In June, AI startup Lindy moved all of its traffic from Anthropic’s Claude models to DeepSeek. The company said the change could save millions of dollars within months.

Other Chinese models are also seeing rapid adoption. Z.ai’s GLM 5.2 model gained attention after its launch, with developers quickly increasing their use of the system.

Harpreet Arora, head of agentic infrastructure at Vercel, said companies are choosing cheaper models when they can complete tasks effectively, “Price is doing the work here,” Arora said.

According to experts, some Chinese open source models can cost 60% to 90% less than leading models from OpenAI and Anthropic.

Despite lower prices, Chinese AI systems are improving in performance. Analysts estimate that leading Chinese models are only several months behind top American AI systems.

Justin Summerville of Open Router said newer open source models are capable of handling most AI tasks, except the most complex ones.

Experts believe businesses will continue using a mix of American and Chinese AI models depending on cost, performance and control requirements.