China's AI industry is inventing a new unit of measurement for itself.
In March of this year, Liu Liehong, director of China's National Data Administration, publicly introduced the term token. The word is positioned as the official settlement unit connecting "technology supply" and "commercial demand"—how much information AI processes is measured in tokens.
According to a Fortune report, as of early 2026, the number of tokens processed daily in China reached 140 billion. That figure is 1,000 times higher than in early 2024.
Hong Kong IPO wave
At the same time, Chinese AI companies are proving themselves in another way: lining up to list in Hong Kong.
The number of IPOs on the Hong Kong Stock Exchange this year is at its highest in five years, driven primarily by AI companies:
| Company | Listing status | Stock performance |
|---|---|---|
| Zhipu AI | Listed | Up 570% from issue price |
| MiniMax | Listed | Up 470% from issue price |
| Biren Technology (chip) | Listed | — |
| Unitree Technology | Applied for 4.2 billion yuan IPO, materials submitted | Pending |
Moonshot AI (parent of Kimi) is valued between $10 billion and $18 billion and is reportedly also considering a Hong Kong listing.
Unitree Technology is a special case: the robotics company has an adjusted net profit of about 600 million yuan, making it almost unique in profitability among China's AI IPO candidates. The vast majority of peers are losing money.
Losses remain the main theme
Speaking of losses, the numbers are not pretty:
- Zhipu AI: Cumulative losses of 4.7 billion yuan, R&D spending up 45% in 2025
- Alibaba: Capital expenditure of 123 billion yuan in 2025, net profit down 66% year-on-year
- ByteDance: Expected AI infrastructure spending of $23 billion
Compared with US peers, Chinese companies are still smaller in scale—Alphabet's 2025 capital expenditure was $94 billion, and Meta's was $75 billion. This gap cannot be closed quickly through policy alone.
Agent adoption: more people trust AI
One comparison in the Fortune report is worth noting: 87% of users in China say they trust AI, compared with 32% in the US.
This gap is directly reflected in product data. ByteDance's Doubao surpassed 100 million daily active users during the Spring Festival this year. Tencent embedded its AI Agent ClawBot into WeChat, allowing users to perform various tasks without leaving the app. Some local governments have begun subsidizing "one-person companies"—entrepreneurs who run their businesses using AI agents, cutting production costs to a fraction of previous levels.
Short dramas are a concrete example: producing a short drama used to cost 1 million yuan; using AI generation, it costs about 100,000 yuan, dropping to 10% of the previous cost.
Structural constraints remain
Several issues remain unresolved:
Chips: US export controls lock Chinese companies into domestic options like Huawei. They can run, but at higher cost and lower efficiency in some scenarios.
Capital: The main source of funding for China's AI is domestic capital; international capital is generally cautious about Chinese AI. Financing costs are relatively higher.
Talent flow: Manus AI moved its corporate headquarters to Singapore and was later acquired by Meta for about $2 billion; its two co-founders reportedly face exit bans. Such cases are not isolated in the industry.
One structural advantage is real: Goldman Sachs estimates that China will have about 400 gigawatts of idle power capacity by 2030, three times the projected global demand for data centers. Computing power does not lack electricity, which is more important than many realize.
Signals from the open-source market
Finally, one data point: downloads of Chinese large models on OpenRouter surpassed those of US models for the first time.
This is not a policy number; it is the result of developers' independent choices. Alibaba's Qwen, under the Apache license, has attracted a large number of global developers. Meta's Muse Spark was also partially trained based on Qwen—something almost unimaginable two years ago.
Whether the concept of a "token economy" gains international acceptance remains to be seen, but China's AI, in terms of model capability and ecosystem coverage, is becoming an option that must be taken seriously.
Sources: CocoLoop, China's AI boom: blazing IPOs, an AI agent craze, and a new 'token economy' (Fortune)