Making Inference Infrastructure Public: A Three-Layer Production Model for Small and Mid-Sized States

Abstract

As AI evolves from an applied technology into the foundational substrate of economic coordination and public governance, the central strategic imperative for states has shifted: from ownership of frontier models to reliable, governable access to large-scale inference capacity.

This paper extends the AI as Public Infrastructure framework by introducing Inference Infrastructure (I&I)—the nationally embedded capacity to generate and operationalize machine-mediated reasoning at scale—and argues it emerges only through synchronized co-development of three interdependent layers: physical compute and data infrastructure, operational AI systems capability, and institutional demand architecture. The absence of any single layer produces systemic underperformance, formalized through the multiplicative capacity function I = C × S × D.

Integrating this I&I model with the Infrastructure Status Index, the paper introduces the asynchrony penalty — the systematic loss of inference capacity that follows from uneven development across the C, S, D layers — as the structural explanation for the documented gap between national AI strategy ambition and observable AI integration across the developing and middle-income world. The paper offers a diagnostic and policy framework for small and mid-sized states navigating the structural tension between digital sovereignty and integration with the global AI ecosystem.

Drawing on evidence from the global shift toward inference-dominant AI workloads and national AI strategies—including Azerbaijan’s AI Strategy 2025–2028 and Digital Economy Development Strategy 2026–2029, we demonstrate that the strategic objective for non-frontier states is not technological prestige but institutionalized access to machine reasoning capacity under conditions of governed interdependence.