Distributed edge intelligence (DEI) is a critical enabler of the manufacturing industry’s digital transformation and the intelligent advancement of the economy and society. It deeply empowers domains including intelligent transportation, industrial smart manufacturing, and embodied intelligence, while serving as the core foundation for achieving secure data circulation and efficient computing power collaboration. Currently, DEI faces four core challenges: the conflict between heterogeneous data distribution across edge devices and global model convergence; the mismatch between the hardware capabilities of resource-constrained terminals and the computational overhead of model training; the tension between frequent high-dimensional feature interaction and real-time performance under limited bandwidth; and the trade-off between token selective transmission and task performance. To address these challenges, this presentation adopts a paradigm that takes data as the foundation, computing as the core, and communication as the bridge, conducting research that extends from federated learning to split federated learning. We further construct a multi-dimensional collaborative DEI framework integrating data, computing, communication, and task dimensions, and design targeted technical solutions spanning data augmentation, adaptive model splitting, communication transmission optimization, and semantic-aware transmission. Finally, we outline future research directions and propose a development vision for the bidirectional integration of large language models (LLMs) and 6G networks.