China's AI Data Surge Signals Economic Pivot to Token Economy

China's artificial intelligence infrastructure is undergoing a structural transformation that extends far beyond algorithmic improvements. Daily token usage — the smallest measurable units processed by large language models — has climbed from just over 1 trillion at the start of 2025 to 140 trillion by March 2026. The numbers, disclosed at the 9th Digital China Summit in Fuzhou, Fujian province, represent more than a growth curve. They signal a fundamental reordering of how economic value is measured and traded.

Liu Liehong, head of the National Data Administration, presented the figures during the summit opening on Wednesday, April 30, 2026. According to China Daily's Global Edition, the trajectory shows "exponential growth" with daily usage reaching 100 trillion by year-end 2025 before climbing another 40 percent in three months. The acceleration is not merely quantitative. It reflects what officials describe as the emergence of a token economy, where computing activity becomes measurable, tradable, and increasingly central to economic output.

Independent reporting from Yicai Global corroborates the timeline and scope. Liu told reporters in Beijing that average daily token usage exceeded 140 trillion in March, representing over 1,000 times the 100 billion recorded at the start of 2024. The scale is difficult to visualize without grounding it in physical reality. Consider the servers humming in data centers across Fujian, the cooling systems working overtime, the electricity bills mounting as inference workloads consume power around the clock.

The data generation figures reveal an even more significant inflection point. China produced 52.26 zettabytes of data in 2025, up 27.28 percent from the prior year. AI and system software accounted for 26.92 zettabytes — overtaking internet-of-things sensing data for the first time. This is not a marginal shift. It marks a structural pivot from passive data collection to AI-driven data creation. The distinction matters because it changes who controls the pipeline and how value flows through the system.

More critically, inference data — generated when models are actively used rather than trained — exceeded training data for the first time in 2025. Of the 199.48 exabytes deployed for AI training and inference, inference reached 101.34 exabytes. The crossover suggests China's AI industry is moving beyond model building toward large-scale commercialization. (This is the moment when theoretical capability meets actual billing cycles.)

Industry experts note the rapid scaling of token usage carries far-reaching implications. Pricing models for AI services must adapt. Global competition over data resources intensifies. As AI systems shift from generating content to executing decisions — and from digital environments into the physical world — the volume, variety, and velocity of data are expected to accelerate further. Data becomes a central pillar of economic growth in the AI era, not an afterthought.

Behind the aggregate numbers, a new layer of industrial infrastructure is taking shape. Fujian-based Joyful Embodied is building what it describes as a large-scale robotic data collection facility. The operation runs around the clock, capturing high-precision industrial data. Arrays of cameras and sensors record robots replicating complex physical tasks, converting them into structured datasets for training embodied AI systems. The physical reality involves metal arms moving through programmed sequences, sensors logging millimeter-level movements, and storage arrays filling with terabytes of motion data.

Chen Yishi, president of Joyful Embodied, said the company's automated pipelines — combining perception, computation, and execution — achieve data collection efficiency of up to 95 percent. That figure sits significantly above industry norms. The goal is to standardize data generation and improve the usability and transferability of datasets across applications. Chen also noted the company is expanding overseas, partnering with firms in Indonesia and eyeing broader Southeast Asian markets. This reflects a wider push by Chinese AI companies to export both technology and data infrastructure.

Such efforts point to a growing race not only for computing power and algorithms, but also for high-quality data. Data is increasingly seen as the bottleneck in advancing next-generation AI, particularly in robotics and embodied intelligence. The Government Work Report delivered to the legislature earlier this month stated that efforts will be made to create a new form of intelligent economy and to further expand the "AI +" initiative. It also called for further development and utilization of data resources and high-quality datasets.

2026 marks the start of China's 15th Five-Year Plan and has been designated as the "Year of Data Element Value Release" by the National Data Administration. By the end of last year, more than 100,000 high-quality datasets had been established nationwide, with a total volume exceeding 890 petabytes. That is equivalent to about 310 times the digital resources of the National Library of China. The next step involves advancing data-enabled innovation in AI development and implementing a new action plan for high-quality dataset construction.

The plan involves strengthening foundations and expanding capacity, tackling annotation challenges, improving quality and efficiency, empowering applications, managing services, and releasing value. The aim is to create technically feasible, practical, and quality-assured AI-ready high-quality datasets. At the same time, efforts will accelerate the establishment of a unified national data property registration system and issue policies for building a national integrated data market.

Whether this infrastructure investment translates into sustained economic advantage remains an open question. The numbers are impressive, but infrastructure alone does not guarantee commercial success. Companies still need to build applications users actually want to pay for. The token economy may be measurable and tradable, but whether it delivers real value beyond accounting entries is what matters. Time will tell if the data becomes a durable asset or just another commodity in a crowded market.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn