Level of Detail 3 (LOD3) has been part of the CityGML standard for nearly two decades, yet it has never become a routine product in operational 3D city model production. While LOD2 building models are now widely generated and maintained by mapping agencies, LOD3 remains largely limited to research prototypes and pilot projects. This paper examines why this gap persists despite sustained progress in photogrammetry, computer vision, and AI-assisted facade reconstruction. We examine LOD3 as a production ecosystem problem involving application demand, industrial workflows, quality assurance, maintenance, and production effort, not as another reconstruction algorithm. We argue that semantically enriched LOD2 remains sufficient for many urban applications, whereas explicit facade geometry can add analytical value for applications such as material inventories, BIM–GIS integration, entrance-aware navigation, flood analysis, and radio propagation simulation in the Next Generation City Networking (NGCN) context. Building on this distinction, we present an industrial pilot in which AI-assisted window and door detection is integrated into an existing photogrammetric LOD2 production workflow and validated by experienced operators. The final pilot dataset contains 505 LOD3 buildings, produced at approximately 50–75% lower operator effort than fully manual photogrammetric restitution according to indicative production estimates rather than controlled benchmarks. We hypothesize that advances in reconstruction technology alone are insufficient to explain or overcome the limited adoption of LOD3, and that ecosystem-level constraints have become an important barrier alongside the reconstruction challenges that remain. The pilot demonstrates workflow integration and operational feasibility rather than city-, regional-, or national-scale scalability.