American businesses prefer Chinese AI because of its cheap price

Last month, AI startup Lindy transferred all traffic from Anthropic's Claude model to DeepSeek's product, a Chinese startup that shocked the world when it launched a low-cost model early last year. "We did that and we saw the cost curve go down, almost straight to the bottom," Lindy CEO Flo Crivello told CNBC. He said this decision will save millions of dollars over many months.

Famous cryptocurrency company Coinbase also claims to nearly halve AI costs thanks to using cheaper models, including products from China such as Z.ai's GLM-5.2 (Zhipu AI) and Moonshot AI's Kimi.

Lindy and Coinbase are two of a series of American companies that are switching from expensive domestic models to cheap products from the East. The proportion of tokens (basic input data units of AI models) that US businesses use for Chinese AI through the OpenRouter platform has remained above 30% every week since February 8, sometimes reaching 46%. The average over the previous 12 months was 11%, even just 4.5% in the first half of last year.

According to Ara Kharazian, chief economist at financial technology company Ramp, in the first five months of this year, the rate of choosing DeepSeek among the 70,000 US companies using Ramp's platform increased from 0.1% to 0.3%. Even in Silicon Valley, many leading AI startups also develop products at least partly thanks to the Chinese open source model.

Data from the Hugging Face platform shows that the Chinese model accounts for nearly half of all open source AI downloads from February 2025 to 2026. Cien Solon, CEO and founder of the LaunchLemonade agent platform, said Claude and ChatGPT still dominate in terms of usage, but GLM 5.2, launched in June, is also in the top 5 of this platform.

"The Chinese model is especially attractive to US businesses as AI costs skyrocket. Previously, they prioritized AI applications regardless of the model, but now they are starting to pay more attention to costs," Kyle Chan, an expert at the John L. Thornton China Center of the Brookings research organization, explained on CNBC.

Praveen Neppalli Naga, chief technology officer at Uber, told The Information in April that increased use of AI programming tools, especially Claude Code, caused the company to burn through its entire 2026 AI budget in a matter of months. "I had to start planning again from scratch, because the amount I thought was enough was now exhausted," he shared.

The Verge reported in May that Microsoft is canceling most licenses for direct use of Claude Code, a programming AI tool developed by Anthropic, and calling on employees to switch to GitHub Copilot CLI to prioritize internal products. However, some sources said that the real reason is that the cost of Claude Code increases as the number of users increases.

Tesla employees are also no longer able to freely use AI as the company seeks to contain skyrocketing costs. In early July, Tesla announced that employees would be limited to using AI tools at $200.

Many founders, software engineers, UX designers, and even vibe coders without a technical background are switching between models to save costs. They assign difficult, high-thinking tasks to advanced and expensive models, while assigning easy, repetitive tasks to older and cheaper models. This strategy helps make the most of the budget, using tokens effectively in the context of businesses increasing cost savings.

"I don't think they prefer the Chinese model, it's just that the best open source products come from China. They choose the best, most efficient and lowest cost models," Lan Xuezhao, a venture capitalist in San Francisco, told Reuters.

The open source and open weighting model allows programmers to inspect, use, and sometimes modify many different parts. They are different from closed models like the flagship products of OpenAI, Anthropic or Google, with the source code and internal working mechanism kept secret.

According to Justin Summerville, data and analytics expert at OpenRouter, the Chinese open source model can be 60-90% cheaper than the flagship products of Anthropic and OpenAI.

Chan said that, although often "a fraction of the cost", the performance of Chinese AI models is increasingly improving, even "close to leading American advanced models". He estimates they are only 6-9 months behind.

Several newly released models, including DeepSeek and GLM 5.2, are considered by many experts to be highly competitive compared to leading advanced systems from Anthropic or OpenAI. They achieved this performance advancement against a backdrop of sharp increases in token costs for the US model series, leaving companies struggling to manage budgets.

"New open source models are performing well and have demonstrated the ability to perform most large language model (LLM) tasks, except for the most complex tasks," Summerville said.

In a standard review, the GLM 5.2 is only about 1% behind Anthropic's Opus 4.8 in terms of performance but is 1/5 cheaper. Some researchers claim that GLM 5.2 can perform on par with leading US models in some cybersecurity assessments.

According to Reuters, a major barrier preventing GLM-5.2 from becoming popular is concerns about data security. This is also the reason why many American businesses limit the use of the Chinese model, especially in tightly regulated fields such as banking and cybersecurity. However, some experts believe that this concern is excessive, running the model on an American cloud service or on the company's own server will be enough to ensure security. While large corporations appear to be slow, many technology startups and small and medium-sized enterprises are changing their strategies using AI models much faster.

In the context of businesses increasing cost savings, Anthropic and OpenAI continue to launch products that are more expensive instead of cheaper. Claude Code and its influence in general started to explode at the end of last year. Over the past several months, Chinese AI companies, including DeepSeek and Moonshot AI, may be actively developing products and will soon release agents with comparable capabilities but at a much lower cost. Therefore, in the near future, America's leading names in artificial intelligence may face even fiercer competition.