This paper proposes a layered framework for defining the architecture of the computational unit of an Open Scholarly Infrastructure (OSI) and presents a concrete implementation possibility through the introduction of OpenCitations, an OSI dedicated to publishing citation data and bibliographic metadata. Grounded in the Principles of Open Scholarly Infrastructure (POSI), the study focuses on the technical dimensions of openness, sustainability, interoperability, and reproducibility that are required for a robust OSI. The proposed approach first identifies the core principles and technical features that a computational block in an OSI should guarantee, including separation of operational domains, orchestration, scalability, observability, automation, and workflow portability. It then maps these requirements onto a layered architectural model composed of Hardware, Virtualization, Orchestration, and Application layers, complemented by a transversal Meta layer. The OpenCitations case study shows how this framework can be instantiated in practice through an on-premises infrastructure. The case is presented with details regarding the technicalities and actual implementations of the proposed methodology. By combining a conceptual definition with a real-world implementation, this work offers both a practical reference and a methodological basis for designing the computational block of an OSI.