Society & Economicsarticle2026-10-12

Governing critical AI capabilities through sovereign orchestration without full stack ownership

Open access0 citations

Abstract

Abstract AI increasingly depends on globally distributed infrastructures, including advanced semiconductors, cloud platforms, data centres, foundation models, software frameworks and specialised labour. This dependence has intensified policy interest in AI sovereignty, which is often associated with domestic ownership of technological assets and the development of a nationally controlled AI stack. This article argues that ownership is neither a sufficient nor universally necessary condition of sovereign control. Instead, AI sovereignty should be understood as the capacity of public authorities to govern the dependencies through which critical AI capabilities are produced, accessed and sustained. The article develops the concept of functional AI sovereignty and identifies six interdependent dimensions of effective sovereign control: visibility, authority, allocation, substitutability, continuity and accountability. It then introduces sovereign orchestration as an institutional model through which governments can coordinate heterogeneous public and private AI resources using regulation, strategic procurement, interoperability, multi-provider routing, capacity clearing and selective public ownership. This model seeks to preserve access to private investment, international supply chains and distributed expertise while preventing individual providers from acquiring an irreversible veto over critical public functions. The article also specifies the limits of functional sovereignty. Orchestration requires sufficient public-sector expertise, access to operational information and credible technological alternatives. Direct ownership remains necessary where interruption costs are high, substitution is infeasible and external legal or geopolitical exposure is substantial. Effective AI sovereignty therefore lies not in eliminating technological interdependence, but in ensuring that critical dependencies remain visible, reversible, resilient and publicly accountable.

// Source

View paper (DOI)Open access versionOpenAlexDiscover Artificial IntelligencePublished 2026-10-12

Authors: Yichen Hao