The modular assembly of organic molecules commonly guides the design of metal–organic frameworks (MOFs), yet the systematic exploration of inorganic building blocks remains limited. Given the promising applications in catalytic reactions and topological materials, the study of metal–inorganic frameworks (MIFs)—which fundamentally differ from the well‐established MOFs—has become urgent. This study introduces a strategy for designing MIFs using host atom‐centered building blocks, applied to the platinum–Phosphorus (Pt–P) system. By combining experimental observations with theoretical calculations, a Pt18P18 framework utilizing PtP3 building blocks, demonstrating significant energetic benefits is proposed. PtP3 units are employed to characterize Pt3P6 clusters within the pores, leading to a “compass” model that aligns with experimental findings. To address computational challenges associated with the large periodicity of the superstructure, a robust machine learning force field is developed. The analysis, combining surface nucleation studies, first‐principles calculations, and machine learning techniques, provides a comprehensive understanding of MIF structure. This validated Pt–P MIF exhibits exceptional catalytic properties with diverse reaction sites, significantly outperforming Pt(111) in the hydrogen evolution reaction. The findings not only present additional candidates for practical applications of metal phosphides but also highlight the vast potential of MIFs, paving the way for the discovery of numerous promising materials.