China’s top data authority has unveiled a sweeping nationwide plan to build high-quality datasets for artificial intelligence training, as developers worldwide grapple with a looming shortage of human-generated text that has powered the current generation of large language models. The initiative, published by the National Data Administration (NDA) as a draft plan on Monday, positions data as a core strategic asset in Beijing’s push to secure a leading position in the global AI race, according to the South China Morning Post.
The urgency of the plan is underscored by a warning from the US-based research institute Epoch AI, which has cautioned that publicly available human-generated text data could be depleted between 2026 and 2032. As AI developers worldwide scramble to secure new sources of training material, China is moving to build a domestic supply that is both abundant and strategically controlled.
A 2028 Roadmap for Industry-Specific Data
The NDA’s draft plan outlines a broad roadmap to expand the supply, circulation, and commercialization of industry-specific datasets. By 2028, the agency aims to field a validated ecosystem of datasets covering bedrock sectors including scientific research, manufacturing, agriculture, energy, transport, finance, healthcare, education, and e-commerce. Cutting-edge frontiers such as embodied AI, autonomous driving, low-altitude aviation, and biomanufacturing are also explicitly included.
The plan calls for an expansion into multimodal data, spanning text, code, images, audio, and video, to train advanced systems capable of complex reasoning, agentic behavior, and controlling intelligent robots. This multimodal emphasis reflects the direction of frontier AI development, where the most capable models are no longer limited to text but must process and generate across multiple data types simultaneously.
The initiative is designed to anchor Beijing’s “AI Plus” strategy, a top-down mandate to weave artificial intelligence into the industrial fabric of the world’s second-largest economy. The NDA, established in late 2023 to spearhead China’s digital infrastructure agenda, said it would collaborate with other ministries to build a national dataset management system and tighten governance over data collection, processing, and use.
Synthetic Data and Automated Annotation
Facing the global data crunch head-on, Beijing is urging the industry to embrace simulation and synthetic data generation as alternatives to human-produced content. The strategy is already producing results. Last week, researchers from the Chinese University of Hong Kong and SenseTime-backed startup Ace Robotics unveiled Kairos-HomeWorld, a framework that generates synthetic residential scenes to train humanoid robots, directly tackling the long-standing shortage of real-world data in robotics training.
On the annotation front, the NDA is calling for a transition away from labor-intensive manual labeling toward automated, AI-assisted tagging. This shift acknowledges both the scale of the data challenge and the efficiency gains that AI-assisted pipelines can deliver compared to traditional human annotation workflows.
Financializing Data as a Strategic Asset
Perhaps the most ambitious element of the NDA’s plan is its proposal to financialize China’s data resources. The agency revealed plans to explore sophisticated commercial models, including subscription services, marketplace-based distribution, and token-based pricing for datasets. It also encouraged experiments in data-backed financing, data trusts, and asset securitization, effectively transforming raw data into bankable financial instruments.
This financialization agenda reflects a broader shift in how Beijing conceptualizes data: not merely as an input to AI development, but as a form of capital that can be owned, traded, and leveraged. By creating liquid markets for high-quality datasets, the government hopes to incentivize both public and private actors to contribute to the national data supply.
A Structural Advantage in the Data Race
The NDA’s initiative highlights a structural advantage that China’s centralized governance model may offer in the coming phase of the AI race. While Western AI developers must navigate fragmented data markets, complex privacy regulations, and competitive hoarding of proprietary datasets, Beijing can coordinate data sharing across government agencies, state-owned enterprises, and private companies through administrative mandate.
The plan’s explicit inclusion of sectors like healthcare, agriculture, and manufacturing, areas where China possesses enormous volumes of operational data, suggests that the government intends to unlock reserves of structured, domain-specific information that could give Chinese AI models a significant edge in applied, real-world settings. As the global supply of general-purpose internet text dwindles, the ability to access deep, sector-specific datasets may prove decisive in the next era of AI development.
Whether the NDA’s ambitious roadmap translates into practice will depend on the government’s ability to overcome persistent barriers to data sharing, including institutional reluctance, privacy concerns, and the technical challenges of standardizing data from heterogeneous sources. But the plan itself signals that Beijing views the data bottleneck as a solvable problem, and one it intends to solve at national scale.