Wuxi giving AI something new: a factory address

A phrase that recently gave me pause is "token factory", as Wuxi, one of China's most seasoned manufacturing cities, is preparing to give artificial intelligence something it has rarely had before: a factory address.

Not a factory that stamps metal parts, assembles circuit boards or loads containers onto trucks, but one designed to produce the measurable units consumed by large AI models — tokens.

On May 15, Honflex signed an agreement with the local government to build Jiangsu province's first Huawei Ascend 384 super-node computing cluster. On top of that infrastructure, Hongxin Electronics plans to establish a large-scale token factory in the city.

The first phase will deploy four Huawei Ascend 384 super-node servers. Each server carries 384 AI accelerator cards, and the four will be connected into one super cluster. In China's emerging market-facing token factory sector, that scale is being described as among the largest.

At first, the phrase sounds almost theatrical. But the more I followed the logic, the more industrial it became. A token is the basic unit processed by an AI model. Every chatbot reply, AI coding request, image-editing command or automated report consumes tokens.

A token factory is an industrialized production facility dedicated to producing the basic units of AI generation — tokens. Inside the token factory are computing-power infrastructure, models and algorithms, scheduling and distribution systems, and a MaaS — model-as-a-service — platform.

This is where Wuxi's story becomes larger than one data-center project. China's AI debate has long revolved around who has the smartest model. Wuxi is asking a more industrial question: can intelligence be supplied continuously, reliably and at scale?

Shen Yang, a professor of Tsinghua University's School of Journalism and Communication, said: "A token factory is the most important factory of the intelligent age. It can turn electricity and data into cognitive productive capacity."

The value of the concept, Shen said, is that it shifts the AI industry's narrative from "how smart the model is" to "whether intelligence can be continuously supplied."

That line captures the real significance of the Wuxi project. The city is not merely adding another computing cluster. It is trying to insert itself into China's next industrial chain — one built around domestic chips, domestic models and domestic applications.

Honflex said the project marks a key step in its domestic AI computing-power layout and an important breakthrough for Wuxi in implementing China's East Data, West Computing strategy and building a domestic computing ecosystem.

Shen added one more condition: global delivery.

"The token factory means it faces not a single market, but global division of labor," he said. "In the future, a country that truly qualifies as a world factory will certainly not be one that keeps models only at home, but one that can export intelligent services to the world."

Standing before this idea, I saw the old Chinese factory model being rewritten. The assembly line has not disappeared. It has become invisible. Its output moves not by truck or ship, but through an application programming interface call. The customer may never see Wuxi. But if the city succeeds, a piece of machine intelligence somewhere on a screen may begin here — inside a token factory.