en.Wedoany.com Reported - Recently, the "Data Factory White Paper," jointly compiled by Beijing Jiaotong University and Beijing University of Chemical Technology, leading over 30 organizations including Huawei, Beijing Data Group, Tsinghua University, and Peking University, was officially released. This white paper is China's first systematic study on the policies, theories, and practices of data factories. It clearly defines the concept, classification, and key characteristics of data factories for the first time, providing a systematic reference for the construction of China's data element industry and the implementation of artificial intelligence scenarios.
The white paper consists of six chapters. The first part, starting from the current state of AI development, the formation of the digital-intelligent industry chain, and its weak links, elaborates on the necessity and urgency of data factories as key facilities for the large-scale production of high-quality datasets. The second part reviews the exploration and practices of typical data factories both domestically and internationally. International cases include Scale AI, the U.S. Stargate Project, European Data Centers, European Data Lab, Parse Biosciences, Bioptimus, and Descartes Labs. Domestic cases cover the "Fan | Future Agriculture Intelligent Hub" super data factory at the Yazhou Bay National Laboratory, the Guizhou Main Hub Storage Capacity Center and Data Element Guarantee Base, the Kupasi "Corpus Super Factory," the AI Data Factory in the Beijing-Tianjin-Hebei Data Element Industrial Park, the Guangdong Advanced Storage Capacity Center, the Pasini Embodied Intelligence Super Data Factory, the "Model Speed Space" in Shanghai's Xuhui District, the "Model Data World" in Beijing's Economic-Technological Development Area, the "Model Data Space" in Zhuhai, and the "Large Model Super Factory" in Beijing's Shijingshan District. It also summarizes the common characteristics of domestic data factories. The third part proposes that data factories are large-scale production facilities and new business forms for high-quality datasets, and are an important component of national data infrastructure. They are classified into centralized, semi-centralized, and distributed types, featuring five major characteristics: diversification, scale, systematization, standardization, and AI integration. The fourth part elaborates on the composition, positioning, and functions of data factories from both broad and narrow perspectives. Broadly, they include the national data base, national data factory, and national AI training ground. Narrowly, they include the reserve workshop, production workshop, and pilot workshop, corresponding to the data raw material supply base, high-quality dataset assembly line, and dataset testing ground, respectively. The fifth part discusses four construction models (iteration by data annotation enterprises, upgrade of data storage bases, extension by AI enterprises, and innovation and creation by technology enterprises), four operational mechanisms (guarantee, customization, pairing, and e-commerce), and three deployment strategies (deploying general knowledge dataset data factories on the national infrastructure base, deploying industry general knowledge dataset data factories in important functional facilities, and deploying industry-specific knowledge dataset data factories at various business nodes). The sixth part proposes development suggestions from four dimensions: policy, standards, technology, and platform, including accelerating the introduction of guiding opinions, initiating standard development, and breaking through key technology bottlenecks.
