Chinese AI models are gaining ground with U.S. companies as OpenAI, Anthropic costs surge

Chinese AI models are increasingly appealing to U.S. companies, driven by their competitive performance and significantly lower costs. Recent innovations from Chinese firms, such as DeepSeek and Z.ai, are closing the gap with leading American competitors like OpenAI and Anthropic, particularly as prices for advanced AI models rise sharply in the United States. Since February, the share of tokens utilized by U.S. companies on Chinese AI models through OpenRouter—a platform that grants access to various AI systems—has consistently remained above 30%, peaking at an impressive 46%. This is a stark contrast to the previous year's average of just 11%, which dipped to 4.5% in the first half of 2025. As U.S. regulators contemplate stricter controls on powerful AI technologies, the rise of open-source and open-weight models from China presents a viable alternative. Notably, in June, OpenAI announced it would limit the deployment of new models at the government's request, while export restrictions on Anthropic's models were lifted after negotiations. Kyle Chan, a fellow at the Brookings Institution, noted that the soaring costs of U.S. AI models are prompting companies to re-evaluate their strategies. "Previously, American firms prioritized AI adoption without much regard for cost, but now they are becoming more budget-conscious," Chan commented. Companies are increasingly experimenting with accessible open-source models, which allow developers to inspect and modify parts of the AI system. This contrasts with closed models from major players like OpenAI and Anthropic, which keep their code proprietary. For example, AI startup Lindy transitioned entirely from Anthropic's Claude models to DeepSeek, realizing substantial cost savings almost immediately. The rapid adoption of Chinese models is evident, with Z.ai's GLM 5.2 being recognized for its remarkable performance shortly after its release. This model alone experienced a staggering 27-fold increase in daily token volume and an 80-fold rise in users within its first week on the market. According to Justin Summerville from OpenRouter, Chinese open-source models can be between 60% and 90% cheaper than their U.S. counterparts. While models like Claude and ChatGPT still lead in overall usage, GLM 5.2 has quickly climbed into the top five on platforms geared toward regulated industries. As the performance of Chinese AI models continues to improve, they are becoming increasingly viable for various applications, with estimates suggesting they are now only six to nine months behind the leading American technologies in terms of capability. With the potential for significant cost savings and adaptability, many companies are weighing the practicality of integrating these models into their operations, despite the ongoing debate about the implications of relying on foreign AI technologies.