U.S. Firms Expand Use of Chinese AI on Low Cost, Openness

2026-07-06

U.S. companies are widening their use of Chinese AI, drawn by lower prices and open access. A temporary halt to use of the advanced AI model 'Mythos' under a U.S. government directive also pushed users to switch providers.

Low prices drive wider use

Brian Armstrong, chief executive of major crypto exchange Coinbase Global, posted on social media at the end of June that the company had cut its AI usage costs by half. In response to rising fees, it increased use of Chinese AI models 'GLM' and 'Kimi'.

GLM was developed by Beijing Zhipu Huazhang Technology (Z.AI), while Kimi was developed by Moonshot AI. Both are strong at coding and long-duration tasks, and their per-token pricing is only about one-twentieth of Anthropic's latest model. The 'GLM-5.2' released in mid-June also drew attention for performance, ranking fifth among roughly 500 entries in a practical skills test covering about 9,000 questions, and some in the U.S. AI industry have compared it to DeepSeek.

Since OpenAI's 'ChatGPT' appeared in 2022, U.S. companies have led in AI performance while Chinese players have trailed. However, while many U.S. firms use closed models that do not disclose technical details, many Chinese AI systems are open models that can be more easily replicated by anyone.

Use of Chinese AI surged in June

As AI has become more advanced, rising usage fees have become a burden for companies. As more tasks involving long and complex work are handed over to AI, token consumption, which directly affects costs, has also increased. More U.S. companies, including Airbnb and Uber Technologies, have disclosed internal use of Chinese AI.

In June, Mythos and the similarly capable 'Fable' were temporarily unavailable under a U.S. government directive, accelerating the shift toward Chinese AI. The U.S. administration, which had been reluctant to regulate AI, moved toward tighter control, and awareness of the risk that government intervention could suddenly cut off access also helped boost usage.

U.S. startup OpenRouter offers a service that lets users switch among more than 400 AI models. Based on usage data from more than 8 million users, it compared token consumption between U.S. and Chinese AI. According to the data, Chinese AI, which had been less than half as heavily used as U.S. AI at the start of the year, began to surpass U.S. models in some periods from February onward, and growth accelerated in June. In the week at the end of June, Chinese AI usage totaled 25 trillion tokens, doubling from the end of May and coming in 78% above the U.S. level.

By company in June, DeepSeek led with a 19% share. Z.AI and Xiaomi also increased usage, while the shares of U.S. Google and OpenAI declined. OpenRouter said users have become more price-sensitive and that Chinese AI is likely to remain in the lead.

U.S. wary of technology leakage

As Chinese AI gains prominence, concern is growing in the United States. Alongside trade and defense, the two countries are intensifying rivalry in AI, a technology that will shape the next era of hegemony. U.S. companies are pouring huge sums into development and infrastructure, and the U.S. government is also providing support.

Anthropic CEO Dario Amodei has said AI could be used in China to strengthen authoritarian rule, including through surveillance of citizens, and weaken democracy. The U.S. ban on advanced Nvidia semiconductors for China remains in place, with the aim of curbing China's technological rise.

One issue U.S. companies are especially concerned about is 'distillation' by Chinese firms. The method uses the output of high-performing models to help develop another AI system, making it easier to improve performance efficiently. But many U.S. companies prohibit unauthorized use as misconduct. In mid-June, Anthropic sent a letter to U.S. senators seeking countermeasures, saying China's Alibaba Group was carrying out distillation on an organized basis.

Open models also face challenges

Open models also carry other risks. There is no guarantee that Chinese companies will continue to make them available externally, and if a high-performance model comparable to Mythos were to emerge, the risk of misuse for cyberattacks or biological weapons development could rise sharply.

Shinichi Takayanagi, an AI specialist at Boston Consulting Group, said open models can be copied, making measures such as service suspension and pre-clearance difficult. He also said price competition could put downward pressure on U.S. AI as well, and if that makes it harder to sustain sufficient investment, safety measures could fall behind.

Enjoyed this article? Share it with your network!