Kilpailu hallintapisteistä: Generatiivisen tekoälyn toimialarakenne ja arvonjako

Generative artificial intelligence (generative AI) is reshaping the digital economy, but the decisive contest is not only about building the best AI model. This thesis shows that value capture is shaped by control of three hard-to-bypass bottlenecks: scarce compute and serving capacity, the rules for who gets access, and the default interfaces users work through. Using an industry architecture lens, the thesis maps the generative AI value stack and identifies the control points through which participation conditions and rents are set across layers. Empirically, the study conducts a theory-driven, qualitative multiple-case analysis of four consequential actors (OpenAI, Google, Meta, and Hugging Face) across the post-ChatGPT acceleration period (November 1, 2022–November 30, 2025). Public framing artifacts are compared against high-salience strategic actions (including product launches, partnerships, acquisitions, and investments) to make frame–action alignment and divergence visible. To reduce interpretive drift in public-source research, the analysis applies an a priori codebook, an explicit evidence hierarchy (salience weighting), and traceable matrix-based comparison across cases and themes. Across cases, the evidence converges on three recurring dynamics: (1) compute supply and specialized infrastructure act as primary conditioning constraints for frontier participation, (2) access is increasingly governed via enforceable boundary resources, safety and compliance regimes, and marketplaces that define participation rules, and (3) interface and workflow capture emerges as a central lever for demand routing and switching costs. Decentralization mechanisms (modularity, open access, and extensibility) broaden diffusion and increase substitutability, but in practice relocate rather than eliminate platform power, shifting it toward hubs, standards, and orchestration layers. The thesis contributes (1) a synthesized framework for contested industry architecture under discontinuity, (2) an operationalization of centralization and decentralization postures for analyzing strategic moves across layers, and (3) empirically grounded implications for where control points are likely to stabilize. The results support researchers studying emerging platform structures and inform policymakers and practitioners evaluating competition, dependency, and governance in the generative AI ecosystem.