Cheap and capable! American companies are increasingly fond of using Chinese AI, with OpenRouter's share once reaching 46%.

Chinese AI models are rapidly penetrating the U.S. enterprise market with prices significantly lower than their American counterparts. The cost-performance advantage, combined with continuously narrowing performance gaps, is driving more U.S. developers and enterprises to switch their workloads to Chinese models.

On the developer platform OpenRouter, the token share of Chinese AI models used by U.S. enterprises has remained above 30% each week since February 8, peaking at 46%. In contrast, the average over the previous 12 months was only 11%, and it was as low as 4.5% in the first half of 2025. This sudden shift comes as top U.S. AI labs like OpenAI and Anthropic continue to raise prices for their flagship models, accelerating the move toward cost-sensitive enterprises.

Cost Pressure Drives the Switch

Price is the core driver of this migration wave. According to Justin Summerville of OpenRouter's data and analysis team, the usage cost of Chinese open-source models is "60% to 90% cheaper" than the leading models from Anthropic and OpenAI.

AI startup Lindy is a typical case of this trend. The company switched all its traffic from Anthropic's Claude model to DeepSeek in June 2026. "You can see that cost curve just drop straight to the floor," Lindy CEO Flo Crivello told CNBC. He expects this decision to save the company millions of dollars over the coming months.

Kyle Chan, a researcher at the John L. Thornton China Center of the Brookings Institution, noted: "As AI costs soar, Chinese AI models have become significantly more attractive to U.S. enterprises. Previously, U.S. companies didn't care about the source of AI models when adopting them, but now they are becoming increasingly cost-conscious."

Performance Rapidly Catching Up to the Frontier

Beyond the price advantage, the performance of Chinese models is also drawing market attention.

GLM 5.2, released by Zhipu in June, scored less than one percentage point behind Anthropic's Opus 4.8 on a widely watched agent benchmark, while costing only about one-fifth as much. Some researchers say GLM 5.2's performance on certain cybersecurity benchmarks is already comparable to that of top U.S. labs.

Harpreet Arora, head of agent infrastructure at Vercel, said GLM 5.2 achieved the fastest model adoption rate Vercel has tracked in 2026—"in the first full week after launch, daily token volume grew about 27 times, and the number of using customers increased about 80 times."

According to Vercel data, DeepSeek's gateway token share also continued to rise between May and June.

Summerville said the new batch of open-source models "performs well, proving practical utility in all but the most complex large language model tasks." Chan estimates that China's top models currently lag behind U.S. frontier products by "about six to nine months," but are already competitive enough for many mainstream use cases.

Enterprise Adoption Expands

The penetration of Chinese models is expanding from startups to a broader range of enterprises.

On the AI agent platform LaunchLemonade, which targets regulated industries, Claude and ChatGPT still top usage, but GLM 5.2 has entered the platform's top five. LaunchLemonade founder and CEO Cien Solon said: "Chinese models like Zhipu's GLM and Alibaba's Qwen are becoming viable options for enterprises, offering an ideal combination of performance and cost for specific workloads. Enterprises with more mature AI strategies are increasingly willing to use them where the technology or business logic makes sense."

Yacine Jernite, head of machine learning at Hugging Face, pointed to a deeper structural trend: "We are seeing enterprises increasingly want to shift to a lower-cost AI stack that they can control and customize themselves, and given the current state of open-source and open-weight models, this often means leveraging Chinese models."

Risk Disclaimer and Terms

        Market risk: invest with caution. This article does not constitute personal investment advice, nor does it take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular circumstances. Investment based on this content is at your own risk.