140 trillion 'Ciyuan': China overtakes the US in AI model adoption as 87% of the population trusts the technology

While Silicon Valley focuses on flashy demonstrations, Beijing has officially named a new unit for the digital economy: “Ciyuan.” With processing speed of 140 trillion tokens per day and a wave of brands listing on stock exchanges, China’s AI push is shifting from follower status to building a distinct ecosystem.

Beijing names “Ciyuan” as an official token unit

For the first time, China has an official term for “tokens,” the data unit used in large language models. Liu Li-hong, head of the National Data Administration of China, introduced the term at a State Council press conference in March, describing Ciyuan as a “payment unit connecting technology supply with commercial demand.”

National Data Administration figures show China processes about 140 trillion Ciyuan per day, up from around 100 billion at the start of 2024.

Market momentum: AI models, IPOs, and investor sentiment

On OpenRouter, a model marketplace, Chinese-origin code is reported to be outperforming U.S. rivals in both efficiency and popularity. The article also links China’s AI momentum to a partial easing of pressure from extended regulatory scrutiny, with AI acting as a “lifeline” for tech giants.

In Hong Kong, IPO activity is at its highest in five years, supported by unicorn startups including MiniMax, Zhipu AI, and chip designer Biren.

Mohit Kumar, global macro strategist at Jefferies, said in mid-March that confidence in China’s position is driven by “attractive valuations, wide AI adoption, and advantages in energy production and supply.”

Adoption and consumer trust

Consumer openness is presented as a key accelerant. An Edelman survey from October found 87% of Chinese respondents trust AI, compared with about 32% in the United States.

AI’s practical impact: short-film production

A cited example of real-world application is the short-film sector. The article says about 470 new films are released daily using AI tools, cutting production costs from 1 million yuan to 100,000 yuan (about $14,600) and reducing production time from about a month to under five days.

Different strategies among major tech firms

Alibaba is described as betting on open source. Its Qwen model allows developers to download and customize freely, attracting users across Southeast Asia and the Middle East. The article also notes that Meta’s Muse Spark was partly trained on Qwen.

Separately, the article says Alibaba CEO Eddie Wu reorganized all AI activities into an “Alibaba Token Hub” (ATH), with a stated mission to “create, distribute, and apply tokens.”

ByteDance is characterized as pursuing a closed approach centered on user experience. Its Doubao chatbot is described as the most-used AI application in China, with 100 million daily active users during the Lunar New Year.

Tencent, while described as trailing, is said to have embedded ClawBot directly into WeChat, enabling more than one billion monthly users to access OpenClaw through the chat interface.

“Physical AI” and profitability: robotics and hardware

The article highlights hardware AI startups leveraging supply chains. Unitree Robotics filed for an IPO valued at 4.2 billion yuan (about $610 million) on the STAR Market in Shanghai. It is described as one of the few robotics names with profits, posting roughly 600 million yuan in net income (about $87 million).

Financial performance: AI labs still face losses

Despite the boom, the article says pure AI labs continue to show losses. MiniMax revenue reached $79 million in 2025 (up 159%), but net loss was $250 million. Zhipu AI revenue exceeded $104 million, yet it posted a loss of $680 million, attributed to a 45% increase in R&D costs.

Market sentiment remains strong: Zhipu’s stock is reported to have surged more than 570% from its IPO price. Moonshot AI—backed by Alibaba and HongShan—is described as reaching an $18 billion valuation, even as Western investors remain cautious due to U.S. chip export controls.

Constraints from U.S. chip export controls

The article notes that U.S. chip export controls still restrict Chinese firms. To respond, Alibaba launched a data center operated entirely with its own Zhenwu-designed chips, though it says performance and output remain uncertain compared with U.S. supply chains.

Sector frictions and international moves

Capital and technology pressures are also described as creating internal frictions. Manus AI is cited as an example: the startup reportedly relocated quietly to Singapore and was acquired by Meta for $2 billion. The article adds that the founders now face travel bans from Beijing authorities.

Outlook: a “token economy” built around efficiency

Overall, the article frames China’s strategy as building a “token economy” focused on efficiency and real-world applicability. It says capex by major players such as Alibaba ($17 billion) and ByteDance (roughly $23 billion) remains below levels seen at Alphabet or Meta, while China benefits from energy infrastructure—an anticipated surplus of 400 gigawatts by 2030—and strong domestic user support.