中圖分類號：F49文獻標志碼：ADOI:10.19358/j.issn.2097-1788.2026.04.002
中文引用格式：涂群，耿貴寧，張茜茜. 數據工廠的構成、建設模式和運營機制研究［J］.網絡安全與數據治理，2026，45（4）：9-16.
英文引用格式：Tu Qun, Geng Guining, Zhang Qianqian. Research on the composition, construction models and operation mechanisms of data factories［J］.Cyber Security and Data Governance，2026，45（4）：9-16.

Research on the composition, construction models and operation mechanisms of data factories

Abstract： High-quality datasets are the core fuel for training large AI models. Currently, the construction of high-quality datasets is mainly carried out by AI enterprises themselves, which presents the characteristics of fragmentation, workshopstyle operation and non-standardization, making it difficult to meet the rapid development needs of large AI models. Drawing on the development patterns of resource-based infrastructure such as water and power plants, and combining domestic and international best practices in facility-based production, this paper proposes the concept of "data factory"，defining it as a production facility specifically designed for the application of large AI models and for the facility-based, largescale construction of highquality datasets. The paper systematically expounds the threelevel architecture system of the data factory, which consists of storage workshop, production workshop, and pilot workshop. Four construction models and four operation mechanisms are proposed, providing theoretical support and practical references for promoting the facilitybased and largescale supply of highquality datasets.

Key words : data factory; high-quality dataset; data infrastructure; data element