This work presents an integrated computational framework that addresses key challenges in large-scale three-dimensional explicit topology optimization: unreliable component connectivity, convergence instability, and prohibitive computational cost. We propose a nodal joint-driven moving morphable component (JMMC) method based on the moving morphable component (MMC) framework to address the key challenges in large-scale three-dimensional explicit topology optimization. This method constructs the joint component and achieves the topology optimization goal by driving the geometric parameter changes of the common component and the joint component, which effectively avoids the boundary fusion issues in the traditional MMC method. To enable this framework at scale, two synergistic algorithmic advancements are developed. First, a novel Probability Density Condensation strategy is proposed to replace conventional KS aggregation for multi-component field composition. Its density-weighted mechanism preserves critical physical signals, ensuring stable sensitivity updates while reducing memory overhead. Second, a Problem-Independent Machine Learning (PIML) surrogate model is integrated. By learning an implicit mapping from material distribution to mechanical response offline, it delivers instantaneous stiffness predictions during online optimization, decoupling computational cost from finite element mesh resolution. Numerical experiments on benchmark 3D structures (e.g., cantilever and torsion beams) demonstrate the framework's efficacy. The results consistently yield optimized topologies with guaranteed connectivity, smooth transitions, and high performance, while achieving computational speedups of one to two orders of magnitude compared to conventional explicit methods. This work establishes a practical and efficient new paradigm for high-resolution, design-ready 3D topology optimization.