Architectural hedging capacity: preserving strategic optionality in AI provider relationships

Purpose Firms increasingly depend on a small number of external providers for artificial intelligence (AI) models, yet that dependence is not only imposed by supplier concentration; it can also be reproduced by a firm’s own technology-integration decisions. This paper aims to examine how provider-specific integration progressively erodes a firm’s ability to switch providers, and what company leadership can do about it. Design/methodology/approach This paper is conceptual. Using resource dependence theory as its base, it draws on transaction cost economics, modularity and real options reasoning, organisational routines theory, and the attention-based view to develop a mechanism and an organisational capability, formalised in four core propositions, a typology matrix and a remedy audit. Findings The paper develops integration drift, a reinforcing process in which locally rational integration decisions cumulatively raise the cost of switching providers. It also introduces architectural hedging capacity (AHC), an organisational capability enacted through three recurring routines: governance-enforced modularity, rehearsed recalibration and sustained managerial attention. And lastly, the argument also identifies a boundary condition in resource dependence theory: where dependence is continuously reproduced through organisational integration, maintaining autonomy requires continuously reproduced exit capacity rather than a one-time remedy. Originality/value The paper extends resource dependence theory to settings where dependence is reproduced inside the firm through its own architecture, not only imposed externally. A typology and a remedy audit checklist give boards and managers a diagnostic for assessing and managing AI-provider dependence.