China has a new unit of economic ambition: the token. Cities including Shenzhen, Hefei and Chongqing are building what officials call “token factories” — large-scale computing operations designed to churn out the fundamental unit of large-language-model processing — as Beijing reframes computing power as core national infrastructure, People’s Daily Online reported on October 13.
What a token factory actually is
In AI, a token is the smallest unit of text a model processes — roughly a word fragment. Every chatbot answer, every line of generated code, is billed and built in tokens. A “token factory” is the industrial metaphor applied to that process: standardised, high-throughput computing centres whose output can be measured, priced and traded like electricity from a power plant. The framing is deliberate. It turns an abstract digital service into something planners can set targets for.
The numbers are already enormous. China’s average daily token calls exceeded 30 trillion in 2024; by March 2026, the figure had climbed past 140 trillion per day. Chinese computing power now accounts for about 42% of the global total, per figures from the China Academy of Information and Communications Technology (CAICT) cited in the report.
An ecosystem, not just data centres
The token push spans the full stack: basic research, foundational models, computing infrastructure, industry applications and commercialisation. It is also a jobs programme — the build-out has reportedly created more than 1,000 new specialised roles, from model engineers to data-centre operators. The strategy connects directly to the hardware race we covered in NVIDIA and Cisco’s optical interconnect launch: whoever supplies the plumbing for AI compute profits from every token that flows through it.
There is a catch, and Chinese analysts state it openly: domestic model capabilities still trail the American frontier by an estimated nine to twelve months, and model prices keep falling — which squeezes the economics of selling tokens even as volumes explode.
The bubble question
The report does not shy away from the risks. Experts quoted warn of a “token bubble”, irrational local-government investment and smart computing centres built faster than demand can fill them, leaving utilisation rates low. It is the classic industrial-policy dilemma, now denominated in tokens: build ahead of demand and risk waste, or build behind it and risk dependence.
The demand side, at least, is not hypothetical. AI usage is compounding across the economy — even as it reshapes consumer markets, driving the memory-chip squeeze behind smartphone sales falling to a 14-year low. Every one of those interactions is tokens, and tokens need factories.
Why it matters beyond China
Treating tokens as a measurable national output — like steel tonnage or electricity generation — gives Beijing a dashboard for the AI race that Washington does not formally maintain. Whether the token-factory model produces an efficient market or a capacity glut will help decide the cost curve of AI worldwide: if Chinese tokens get very cheap, every AI product built on price-sensitive inference feels it. The factories are humming either way. The open question is who buys the output — and at what price.
