The term "token" has dominated headlines this year, largely driven by a staggering statistic: in March, China's daily token usage surpassed 140 trillion, surging over a thousandfold in just two years. This explosive growth fuels an equally massive demand for computing power. Recent reports have highlighted Wuxi's pioneering efforts in AI computing operations with its "token supermarket," a platform that makes accessing large language models both affordable and convenient. On August 7th, Wuxi secured a prestigious national designation—the National AI Industry Innovation Application Pilot Zone. As the only prefecture-level city among the selected zones, Wuxi has earned a pivotal vote of confidence in the race to dominate AI. From the micro-level innovation of the "token supermarket" to the macro-level national pilot zone, Wuxi is charting a clear path, one that may reveal China's shifting AI strategy from a focus on raw technology to real-world implementation.

What concrete problem does the "token supermarket" solve?

Consider a real-world scenario involving Jiangsu Xiwesi Security, a company developing intelligent security software. Software department head Li Tao observed that since last year, 85% of the company's software development has been handled by AI. Initially, this worked well. However, as computing needs surged, problems emerged. The company had to purchase tokens from multiple AI model platforms, each with different performance and use cases. This led to rising costs and the administrative burden of managing several accounts and switching between models. When the Wuxi AI Industry Service Platform—the "token supermarket"—launched in May, Li Tao was an early customer. The "shelves" displayed major models like DeepSeek, Tongyi Qianwen, and Kimi side-by-side. Prices were the bulk-discounted rates from the platform's centralized procurement, and a single account allowed access to various models at any time. He calculated the savings: a monthly top-up of 20,000 yuan gave his team of 30-plus engineers access to 4 to 5 billion tokens. A bundled "concurrency" package allowed 30 people to work simultaneously. Costs dropped by nearly 30%, and software development cycles were cut from two weeks to less than one. The operator of the supermarket, Wuxi Digital New Infrastructure Co., Ltd., explained that it functions like a one-stop shop, where engineers need only one account to browse and purchase model resources, slashing R&D costs by 30% and boosting efficiency by 50%. It has already served over 50 companies. The platform's added value is its integration with financial services, embedding financial products into the payment channel to lower the capital barrier for companies pursuing AI innovation. This is not just a tool for small and medium-sized enterprises (SMEs) to reduce costs and enhance efficiency; it is a breeding ground for a local AI ecosystem. This reveals a key logic: lowering the barrier to AI adoption can activate an industrial ecosystem more effectively than simply amassing computing power.

Of course, Wuxi's ambitions are grander. The "token supermarket" is just an entry point. What Wuxi is truly building is a "token factory" system, structured as a "public platform plus market-oriented benchmarks plus edge nodes." This system can be best understood through three pillars:

The first pillar is the inclusive foundation—the "token supermarket." As a public platform, it addresses the core pain points for SMEs: the inability to afford or effectively use AI. It bundles dozens of large models, offering centralized procurement, unified accounts, and pay-as-you-go pricing, making computing power as accessible as water or electricity. The second pillar is the capacity benchmark—the "super factory." Building on the inclusive platform, Wuxi is developing hardcore production capacity. This includes plans for a 1.4 trillion daily token industrial base by iSoftStone and the nation's first "East-to-West Computing, Computing-Electricity Integration" mega token factory by Suihong Huachuang. The goal is to provide large-scale, high-quality computing supply. The third pillar is the edge node—the "workshop micro-factory." Leveraging its strong manufacturing base, Wuxi is encouraging leading enterprises like Xingcheng Special Steel, Heilan Group, and Far East to build "mini token factories" right on their production lines. This "produce-for-own-use" model precisely meets the stringent requirements for low latency and data security in industrial settings, effectively bringing computing power down to the factory floor. A city official from the Municipal Bureau of Industry and Information Technology stated that building this token factory system will meet growing computing demands while accelerating a positive feedback loop of "token consumption, industrial upgrading, and tax revenue growth," securing Wuxi's early advantage in the smart economy.

How substantial is Wuxi's computing power foundation?

The confidence in this system stems from Wuxi's solid computing power base. Currently, Wuxi has 19 intelligent computing centers with a total intelligent computing scale of 15,619 P. The city's computing power scheduling and operations platform is live, connected to a 10,000-card heterogeneous computing resource pool, enabling cross-architecture, cross-regional unified scheduling across the Yangtze River Delta. Concrete examples illustrate this capability. In May, the nation's first computing equipment industrial park broke ground in the Wuxi High-tech Zone. Spanning about 1,000 mu, the park focuses on a "modular AIDC" model, compressing the typical 18 to 24-month construction period for a traditional data center into just months, rapidly assembling intelligent computing centers like building blocks. On June 11th, the first phase of the Wuxi (Huishan) Domestic Intelligent Computing Center, powered by domestic GPU leader Moore Threads, was officially launched. Moore Threads also joined 16 other companies in Wuxi to launch an Industrial Embodied Intelligence Innovation Center and is planning a domestic intelligent computing cluster supporting a 100,000-card network. Another domestic GPU company, Muxi, has partnered with Cloud Factory Technology to put the first phase of a 10,000-card intelligent computing cluster into operation. Equipped with Muxi Xiyun C550 computing devices, it supports large model training and inference scenarios. Perhaps most impressive is a cross-ocean demonstration. In July, at AMD's Global Developer Conference in San Francisco, nearly 1,000 developers conducted simultaneous AI application development, all powered by a single intelligent computing room in the Wuxi High-tech Zone, over 10,000 kilometers away—the Xinwei Zhi Computing Thousand-Card Cluster. This global first AMD ROCm open-source ecosystem intelligent computing center completed 1,159 development environment launches over two days, covering over 900 active developers with 100% core service availability, zero faults, and zero interruptions. The depth of computing power directly determines the density of industrial clustering. The influx of major projects confirms this "magnetic effect." Mega-projects like Shenghejingwei and the third phase of Huahong Semiconductor have landed. Numerous hard-tech companies, including Xinje Electric, JD Intelligent Robots, Gaode Electronics, Lianmao Electronics, Yaohong Electronics, Muxi, and Zhipingfang, have signed agreements to establish bases. In the first half of this year alone, Wuxi attracted nearly 200 AI projects, with signed investments reaching 63 billion yuan, accelerating the formation of a closed-loop chain encompassing "Wuxi chips, Wuxi models, Wuxi tokens, Wuxi terminals, and Wuxi applications."

How to monetize the shift from computing power to industry?

The ultimate test for the token factory is its user base, application, and subscriber retention. According to Yang Yi, a senior architect for public services at Volcano Engine, Wuxi's strategy is to deeply embed tokens into the production processes of thousands of industries by leveraging real-world industrial scenarios across the city, creating a "super experimental field." In Wuxi, AI finds its purpose everywhere. In manufacturing, AI acts as the "brain," with thousands of factories lining up for digital upgrades. Massive markets are opening up for industrial vertical large models, intelligent inspection, digital twins, and industrial robots. In the city, AI brings "warmth," with government service models, smart transportation, smart healthcare, and city lifelines being fully deployed. Over 100 urban AI agents are on the job, making city life more intuitive from healthcare and education to transportation. For entrepreneurs, there is "fertile ground," with communities like OPC, AI drama spaces, and Alibaba empowerment centers setting up. Here, good technology will not lack a scenario and can quickly commercialize. Li Wei, head of Huazhi AI, noted that to make robots truly "understand physics," they need to "see" and "experience" enough real-world scenarios, not just follow a few instructions. This year, a key change was made to the national pilot zone's name: the word "industry" was added before "innovation application." This subtle shift reflects a top-level strategic pivot—the second half of AI competition is about integration into industry and practical application, not just pure technology. Wuxi's fundamental character has always been built on industry. In 2025, its GDP reached 1.68 trillion yuan, with a total industrial output value above a designated size of 2.71 trillion yuan. Eight industrial clusters each exceed 200 billion yuan. In the first half of the year, the core AI industry scale reached 43.838 billion yuan, growing 33.4%. What likely convinced the judges, however, is not these numbers alone, but the capability to embed AI into genuine industrial scenarios. Wuxi boasts 18 typical AI application scenarios for SMEs selected by the Ministry of Industry and Information Technology, the highest number in the province. Wuxi has now set its "small goal": by 2027, it aims to have over 500 AI enterprises above a designated size, launch 200 vertical models, and implement 300 "AI + Manufacturing" scenarios, striving to reach the international first tier. From the micro-innovation of the "token supermarket" to the macro-layout of the national pilot zone, Wuxi's logic remains clear: not to chase concepts, but to root AI in industry and empower the real economy. In this race from algorithms to the physical world, a city built on a manufacturing foundation is using persistence and ingenuity to explore a path to the "AI+ " high ground.