According to a report by the American consumer news and business channel CNBC on July 7, as Chinese artificial intelligence models narrow the gap with their leading American counterparts in terms of performance and at significantly lower costs, they are becoming increasingly favored by American companies.
Reports indicate that models released by Chinese companies such as DeepSeek and Zhipu are widely considered to be highly competitive compared to leading American AI systems like Anthropic and OpenAI. At the same time, the parameter costs of many advanced models from American AI laboratories are increasing, resulting in companies facing unexpected high costs associated with using these technologies.
Data shows that since February 8th, through the OpenRouter platform, which allows access to various AI models, the proportion of tokens used by American companies in Chinese AI models has exceeded 30% each week, reaching a peak of 46%. In the past 12 months, the average proportion was only 11%, and in the first half of 2025, it dropped to as low as 4.5%.
According to Kyle Chan, a researcher at the Brookings Institution in the United States, "As AI costs soar, Chinese AI models are now particularly attractive to American companies. Previously, US companies prioritized adopting AI, regardless of model choice, but they are now prioritizing cost."
He further stated that although the operating costs of Chinese models are usually only a "small fraction" of those of American competitors, their performance is "close to that of top American models." He estimated that Chinese models are currently six to nine months behind top American competitors.
AI startup Lindy shifted its entire traffic from Anthropic's Claude model to DeepSeek in June. CEO indicated that this move will save the company millions of dollars within a matter of months, "We did it, you can see the cost curve dropping sharply and falling off a cliff."
On the platform Vercel, which allows developers to deploy and run applications and websites, DeepSeek's share of gateway terms also increased during the month of May to June.
Vercel's intelligent infrastructure chief, Harpreet Arora, told CNBC that ZipGPT released GLM-5.2 in June has become the fastest model to track all models by 2026, with its "first full week after release saw daily word volume grow about 27 times and the number of customers using it rose roughly 80 times."
“Price is the key,” said Arora. “When a task doesn’t require the best model, teams will start choosing models that are cheapest but still good enough. The recent wave of models from China is winning this competition.”
According to US media reports, as open-source and open-weight models become more popular in China, the US government is increasingly seeking to regulate its most powerful AI models and considering measures to prevent the rapid adoption of alternative solutions abroad.
According to analysts responsible for data and analysis at OpenRouter, Chinese open-source models are 60% to 90% cheaper than models from Anthropic and OpenAI. On the AI agent platform LaunchLemonade, Claude and ChatGPT remain dominant, but GLM 5.2 has entered the top five on this platform.
We see an increasing trend of companies preferring to use cheaper, more customizable AI technology stacks with greater flexibility. Given the current situation of open source and open-weight models, this often means choosing Chinese solutions, according to Yacine Jernite, Head of Machine Learning at Hugging Face, as per CNBC.
