The Phenomenon of Americans Flocking to Chinese Technology

China’s efforts to catch up with the United States in the artificial intelligence (AI) technology sector are finally bearing fruit. AI models built by China are increasingly being used by US-based companies. This phenomenon signals growing confidence in Chinese AI models, especially as their costs are far cheaper compared to US-made AI.

Veteran Chinese AI models such as DeepSeek and Z.ai are considered capable and competitive enough against leading US AI models from Anthropic and OpenAI. At the same time, the token prices for using US AI models are getting higher. Amid the current economic uncertainty, companies are considering cheaper alternatives rather than continuing to depend on expensive AI models offered by US giants.

According to data from OpenRouter, a platform that allows developers to access various AI models, there has been a significant increase in the use of tokens for Chinese AI models by US companies. Its market share has risen above 30% every week since 8 February 2026, reaching a peak of 46%, as reported by CNBC International on Tuesday (7/7/2026), based on OpenRouter data.

The rising popularity of Chinese open-source models coincides with the US government’s efforts to tighten regulations for its most advanced AI models, while continuing to curb the adoption of foreign-made alternatives. At the end of June, OpenAI said it would limit the launch of new models at the request of the Trump administration. Export control policies for Anthropic’s Mythos and Fable models were also implemented, although they were later revoked this month.

“Chinese AI models are very attractive to US companies right now, as the cost of AI [for US-made models] increases,” said Kyle Chan, a research fellow at the John L. Thornton China Center at the Brookings Institution think tank. “Previously, US companies prioritised AI adoption, now they are more conscious of the costs involved,” he added.

As companies begin to run AI models to create new products and improve internal efficiency, engineers are increasingly driven to experiment with open-source and open-weight models. For context, open-source models provide the entire ‘recipe’ of the AI system, from training parameters, training code, and datasets. Meanwhile, open-weight models only share the model parameters that can be downloaded, but ‘hide’ how the data was trained and used.

Open-source and open-weight models offered by Chinese AI producers allow developers to inspect, use, and sometimes modify them. This scheme differs from the closed-source models offered by US companies like OpenAI, Anthropic, and Google, where the code and ‘kitchen recipe’ become the company’s intellectual property.

In June, AI startup Lindy moved 100% of its service traffic from Anthropic’s Claude model to China’s DeepSeek model. “We saw costs drop dramatically,” CEO Flo Crivello told CNBC International. He said the decision saved the company billions of US dollars within just a few months.

During May and June, the token usage for the DeepSeek model surged on Vercel, a platform that allows developers to deploy and run applications and websites. The GLM 5.2 Z.ai model was launched in June and immediately saw the fastest adoption among AI models tracked by Vercel throughout 2026, according to Harpreet Arora, Vercel’s head of agentic infrastructure.

“In the first week after its launch, the daily token volume [for GLM 5.2] grew 27-fold and the number of consumers using it grew about 80-fold,” he added. According to Arora, price is the deciding factor for many companies switching to AI models from China. “When a task does not require the best model, company teams will start switching to the cheapest model that is good enough. The wave of new model releases from China is winning this momentum,” Arora explained.

Chinese open-source models can be around 60%-90% cheaper than leading models made by Anthropic and OpenAI, according to Justin Summerville, who analyses data at OpenRouter. Although Claude and ChatGPT still dominate usage on LaunchLemonade, an AI agent platform for regulated industries, GLM 5.2 has now entered the ‘top 5’ most used models on that platform, according to LaunchLemonade CEO Cien Solon.

“Chinese models like Z.ai and Qwen are becoming options for companies. These models are attractive because they combine performance and affordable pricing for specific tasks,” said Cien. The performance of Chinese AI models is also continuously improving. Although their costs are often cheaper than US competitors, these models operate “close to the capabilities of leading US models,” said Chan from Brookings, estimating that they are currently “six to nine months” behind the main US competitors.

“These new open-source models show good performance and are proven capable of handling almost all LLM tasks, except the most complex ones,” said Summerville. GLM 5.2 recorded results just one percentage point shy of Anthropic’s Opus 4.8 on a widely watched AI agent benchmark, at about one-fifth of the cost. Some researchers mention that GLM 5.2 is able to match the performance of leading US laboratories on several cybersecurity benchmarks.