In early 2024, China processed roughly 100 billion AI tokens per day. As of March 2026, that number is 140 trillion. That is a 1,400x increase in two years.
To put this in perspective: the entire global conversation about AI in the West revolves around which model scores highest on benchmarks, who raised the biggest round, and whether we need to slow down. China skipped that conversation entirely. They built an economy around it.
They literally coined a word for it
In March 2026, Liu Liehong, the administrator of China’s National Data Administration, unveiled the term “ciyuan” at a State Council press conference. It is the Chinese word for token. He defined it as “the settlement unit linking technological supply with commercial demand”.
Read that again. A government official, at a national press conference, declared the AI token to be an economic unit. Not a technical artifact. Not a developer tool. An economic primitive, like the dollar or the kilowatt-hour. A unit of exchange in a new kind of economy.
No Western government has done anything remotely comparable. The closest equivalent would be if the US Secretary of Commerce held a press conference to announce that the API call is now a unit of GDP measurement. It sounds absurd in English. In China, it is policy.
The IPO numbers are hard to believe
Hong Kong raised almost $14 billion in AI and tech IPOs in Q1 2026 alone. That is a 490% increase year-on-year. Five-year high. The pipeline includes MiniMax, Zhipu AI, Biren, Moonshot AI, and Unitree Robotics.
The stock performance of the companies that already listed is staggering. Zhipu AI is up 570% from its IPO price. MiniMax is up 470%. These are not meme stocks. Zhipu generated 724 million yuan ($104.8M) in revenue last year, a 132% year-on-year increase. MiniMax hit $79 million in revenue with 159% growth, and 70% of that revenue comes from overseas markets.
They are losing money. Zhipu has accumulated 4.7 billion yuan ($680M) in total losses. MiniMax posted a $250 million adjusted net loss. But the market is pricing them on trajectory, not profitability. The same way it priced Amazon in 2002.
The spending is real
Alibaba spent 123 billion yuan ($17B) in capex in 2025. Tencent spent 79 billion yuan ($11.6B). ByteDance is expected to spend $23 billion on AI infrastructure. For context, Alphabet spent $94 billion globally and Meta spent $75 billion.
Chinese tech giants are spending at the same order of magnitude as American ones. With cheaper labor, cheaper construction, and a government that actively subsidizes data center buildout.
Goldman Sachs estimates that China will have approximately 400 gigawatts of spare power capacity by 2030. That is roughly three times the projected global data center demand. The infrastructure bottleneck that constrains American AI companies does not exist in China in the same way.
The trust gap is the real story
An Edelman survey from October 2025 found that 87% of Chinese respondents trust AI. In the United States, that number is 32%.
This is not a technology gap. It is a cultural and institutional gap. When your population trusts AI at nearly 3x the rate of your competitor’s population, adoption is faster, regulation is lighter, and the feedback loop between deployment and improvement spins faster.
Chinese consumers are already using AI-generated short dramas at scale. Roughly 470 new AI-assisted dramas launch every day. Production cost: about 100,000 yuan ($14,600), roughly 10% of conventional production. Timeline: under 5 days, down from 15 to 30. This is not a lab demo. This is a consumer market.
Pony AI launched Europe’s first commercial robotaxi service, in Zagreb. WeRide is partnering with Uber for fully driverless robotaxis in Dubai. The Chinese robotics company Unitree is filing a 4.2 billion yuan ($610M) IPO with an adjusted net profit of 600 million yuan. Profitable. In robotics.
What this actually means
There are two ways to read this situation.
The optimistic reading: competition accelerates innovation. Chinese AI companies pushing hard forces American companies to move faster. The global pie grows. Everyone benefits.
The realistic reading: two AI ecosystems are emerging with fundamentally different structures. The American ecosystem is privately funded, safety-conscious, and increasingly regulated. The Chinese ecosystem is state-supported, deployment-first, and operating at a scale that compounds daily.
Chinese AI models have already surpassed US models on OpenRouter, a popular marketplace for AI model access. Whether those rankings hold is less important than what they signal: the performance gap that US companies relied on as a moat is closing, and in some categories it has closed.
US export controls on advanced AI chips remain in place. But as Fortune reports, Meta’s own Muse Spark model was trained partly on Alibaba’s Qwen. The lines between the two ecosystems are blurrier than the policy framework assumes.
The question is not whether China is a serious competitor in AI. That question was answered somewhere between 100 billion and 140 trillion daily tokens. The question is whether the West’s approach to AI development, slower, more cautious, more regulated, produces better long-term outcomes, or whether it produces better arguments for why the outcomes were not as good.
That is an honest question. It does not have an easy answer. And 140 trillion tokens a day are not waiting for one.
Sources: Fortune: China’s AI Boom, Fortune: Hong Kong AI IPOs