Educational systems traditionally emphasize the acquisition of knowledge and procedural skills through the development of long-term memory. However, cognitive science suggests that learning depends on earlier stages of the cognitive process — including sensory attention, working memory, and metacognitive regulation — that receive comparatively little systematic attention in formal education. This paper proposes the Cognitive Infrastructure Theory of Learning, arguing that attention and working memory constitute foundational processes through which knowledge becomes possible, and that the infrastructure metaphor generates novel, testable predictions about educational design that predecessor models — including cognitive load theory, Global Workspace Theory, ACT-R, and SOAR — do not individually produce. Drawing on research in cognitive psychology and learning sciences, the paper introduces a three-layer architecture of learning consisting of sensory attention, working memory integration, and long-term knowledge formation, with metacognitive regulation operating across all three layers. The framework highlights a structural imbalance in contemporary education and provides concrete examples of cognitive infrastructure training at primary, secondary, and tertiary educational levels. Five theoretical propositions are developed to guide future empirical research, and practical implications for instructional design are discussed.