Digital sovereignty and AI governance: from claim to instrument in a comparison of the Brazilian AI Plan and America's AI Action Plan

<jats:p>Digital sovereignty has become a common vocabulary of national artificial intelligence (AI) strategies, yet the literature describes sovereignty claims without explaining why the same vocabulary yields opposite policy configurations. A theoretical gap remains: no framework connects the layer of the claim to the layer of the instrument, and none accommodates sovereignty exercised over the access of third parties. This article compares the Brazilian Artificial Intelligence Plan (PBIA 2024-2028) and America's AI Action Plan (AAP, 2025) in order to answer how each plan translates sovereignty claims into policy instruments, and what that divergence reveals about the relation between structural position in AI production and the sovereignty strategy available to a state. The study adopts critical realism and retroductive inference, with comparative documentary analysis of two maximally different cases coded along five dimensions and three layers: declared, instrumented and exercised sovereignty. The findings show that the PBIA declares sovereignty across all of its axes yet instruments it through subsidised credit, allocating 0.45 per cent of the envelope to the governance axis and around 2 per cent of structural investment to the electricity enabler, whereas the AAP instruments sovereignty through deregulation, permitting acceleration, export controls and the export of an integrated technology stack. The theoretical contribution lies in distinguishing the three layers, in extending the typology of state roles to incorporate control over the access of others, and in three propositions with stated boundary conditions. Documentary analysis captures institutional design rather than implementation dynamics.</jats:p>