Pharmaceutical research increasingly depends on the combined interpretation of chemical, biological, preclinical, clinical, safety, manufacturing, and real-world evidence. These data are commonly distributed across scientific functions, technology platforms, legal jurisdictions, and stages of the product lifecycle. Centralized architectures attempt to overcome this fragmentation by consolidating data into shared repositories, but consolidation alone does not preserve scientific context, resolve semantic incompatibility, establish fitness for reuse, or reconcile competing ownership and access obligations. This original pharmaceutical data architecture article develops a proposed ownership-preserving pharmaceutical data mesh in which scientific domains remain accountable for versioned data products while participating in federated semantic, lineage, access, computational, and governance services. The approach integrates evidence from pharmaceutical data science, FAIR data practice, biomedical interoperability, distributed computation, provenance engineering, model governance, and translational implementation. Its principal contribution is a conceptual architecture that distinguishes physical data location from semantic interoperability and separates data availability from scientific suitability. Domain-owned data products are connected through explicit semantic contracts, distributed evidence-lineage records, federated stewardship, policy-aware access control, governed compute-to-data services, and qualified model-reuse mechanisms. The architecture is evaluated through multiple dimensions rather than a single performance measure, including semantic validity, lineage completeness, policy correctness, reproducibility, model transportability, evidence-reuse fidelity, and organizational sustainability. The proposal remains conceptual: it does not establish privacy guarantees, empirical superiority, regulatory acceptability, clinical utility, or deployment readiness. Its value lies in organizing the scientific and governance conditions under which pharmaceutical evidence may become reusable without being detached from its ownership, provenance, uncertainty, and intended context.