Artificial Intelligence; Technological Trajectories; Industrial Concentration; Capital Deepening; Scaling Laws; Foundation Models; Evolutionary Economics; Infrastructural Control; Digital Oligopoly; Innovation Dynamics.

This paper analyzes the contemporary development of artificial intelligence (AI) as a concentrated technological regime characterized by scaling dynamics, cumulative learning processes, and infrastructural control. Departing from approaches that treat AI as an exogenous technological shock, the analysis adopts an evolutionary perspective in which technological change is path-dependent and co-evolves with industrial organization and institutional conditions. The paper argues that industrial concentration in AI is not a contingent outcome of market imperfections, but an endogenous feature of the underlying technological trajectory. The dominant search heuristic -scaling through compute-intensive architectures- privileges capital deepening, access to large datasets, and control over computational infrastructures. These features generate increasing returns, reinforce cumulative advantages, and produce persistent asymmetries across firms. At the same time, the paper challenges conventional interpretations of AI-driven productivity growth. Indicators such as revenue per employee are shown to reflect monetary measures of value appropriation rather than clearly defined increases in output. In this context, observed "hyperproductivity" is better understood as a function of pricing power, control over proprietary knowledge, and the ability to monetize access to AI infrastructures. The analysis further shows that the translation of task-level performance improvements into broader economic outcomes is structurally mediated. AI systems generate localized efficiency gains, but their effective deployment depends on organizational transformation, complementary capabilities, and position within the technological stack. As a result, diffusion remains uneven and contingent, and value capture is shaped by industrial structure. Finally, the paper highlights the material dimension of the AI trajectory, emphasizing the environmental externalities associated with large-scale computational infrastructures. These resource-intensive dynamics are not peripheral, but intrinsic to the current scaling paradigm and further reinforce barriers to entry and concentration. Taken together, the paper provides a unified framework linking technological trajectories, industrial concentration, and value appropriation, offering a structurally grounded interpretation of the contemporary AI regime