The following draft Chinese government policy is designed to increase the quantity and quality of datasets available to train AI models for various industries. The draft addresses data as a growing bottleneck for AI development in China, particularly as regards datasets that can be used to train and test AI systems that interact with the physical world, such as AI-enabled robots. The authors invited public feedback on the draft from April 15 to 22, 2026. China subsequently released the final version of this policy on June 8, 2026; a CSET English translation of the final version is underway.
An archived version of the Chinese source text is available online at: https://perma.cc/V4DA-DY39
Implementation Plan for the Initiative to Promote the Construction of High-Quality Industrial Datasets
(Draft for Feedback)
High-quality industrial datasets are collections of industry data that have undergone data handling, including collection, processing, and other operations; can be used directly to develop and train artificial intelligence (AI) models; and can effectively improve the effectiveness of applications such as models, intelligent agents, and smart terminal devices. They include industrial general knowledge datasets and industrial specialized knowledge datasets. High-quality industrial datasets are foundational and key resources for promoting “AI+” to empower all industries and achieve industrial implementation. This plan is formulated in order to implement the Outline of the 15th Five-Year Plan for National Economic and Social Development, thoroughly implement the “AI+” initiative, promote the construction and adoption of high-quality industrial datasets in step with, and in mutual reinforcement of, “AI+,” and strengthen the role of data in empowering the innovative development of AI.
1. Overall Requirements
Guided by Xi Jinping Thought on Socialism with Chinese Characteristics for a New Era, we will thoroughly implement the spirit of the 20th Party Congress and all subsequent plenums, comprehensively implement the “AI+” initiative, proactively adapt to the paradigm shift in AI development, and follow the principles of “demand-driven, urgent needs first, application validation, and security assurance.” Focusing on key industries for national economic development and strategic emerging industries, and centering on key links such as the supply, circulation, and application of high-quality industrial datasets, we will launch six special initiatives: (1) Strengthening foundations and expanding capacity, (2) tackling the challenges of data labeling, (3) improving quality and efficiency, (4) empowering applications, (5) management services, and (6) creating value. This will form a “data flywheel” in which “scenarios lead the way for data, data drives models, models empower applications, and applications create value,” and accelerate the building of a symbiotic ecosystem in which the data factor of production (数据要素) and AI evolve in coordination.
By the end of 2028, a number of high-quality industrial datasets that cover key fields and have been validated through application will have been established; a number of typical application scenarios in which data drives the innovative development of AI will have been created; a number of innovative data enterprises and professional personnel with leading advantages will have been cultivated; and a number of standards and tools for building high-quality industrial datasets will have been developed. A virtuous cycle from data supply to value creation will have basically taken shape, the role of data in empowering the innovative development of AI will become more pronounced, the data industry and AI will be deeply integrated, and new growth points in the intelligent economy will continue to emerge.
2. Implement the Initiative to Strengthen Foundations and Expand Capacity
In response to the trend of AI’s accelerated penetration into industries and its paradigm shift from dialogue toward multimodal generation, decision-making and execution, embodied AI, physical interaction, and other areas, broaden data supply channels, enrich the types of data supplied, and accelerate the building of high-quality industrial datasets to provide sufficient “fuel” for AI development and application.
To view the rest of this translation, download the pdf below.