Abstract
Data processing in machine learning and Artificial Intelligence (AI) can be viewed as a conventional supply chain in which data are the primary flowing artefact: they are sourced, generated, collected, stored, and shared across upstream and downstream actors to produce valuable information products and business outcomes. Within AI-data supply chains, we map common training optimisation techniques to explicit extensions of conventional supply chain production technology, focusing on two classes of structural decisions: (i) merger (e.g., ensemble learning, distributed training, and knowledge distillation) and (ii) reconfiguration (e.g., algorithm hardware matching). Although such horizontal mergers and reconfigurations provide a mechanism for productivity growth and improved resilience under disruptions, their implications for overall supply chain productivity remain difficult to forecast. To address this gap, we develop a general framework for evaluating merger and reconfiguration decisions in complex supply chains, with an application to AI-data supply chains. Leveraging key features of the Free Disposal Hull (FDH) model and the free-arrangement assumption in network nonparametric production technologies, we propose a non-radial directional distance function model to support decision-making in merger and reconfiguration-oriented supply chain design. We demonstrate the applicability of the proposed approach using a dataset that represents an AI-data supply chain.