Control of centralized data architectures, not mere data generation, is emerging as the decisive source of competitive advantage in AI-enabled platforms; SMEs face value capture through architecture-mediated data extractivism absent stronger portability or transparent governance.

Main Finding

Control over centralized aggregation architectures — not mere localized data generation — is becoming the primary source of competitive advantage in AI-enabled platform ecosystems. Platforms that aggregate cross-SME data capture compounding “learning rents,” producing disproportionate value accrual to architecture controllers and undermining the resource-based assumption that ownership of local resources guarantees rent appropriation.

Key Points

• SMEs generate substantial transactional and behavioral data through platform participation, but platform sponsors typically control data aggregation and model-training architectures.
• Structural decoupling: data generation (by SMEs) is separated from aggregation/learning control (by platforms), weakening SMEs’ ability to appropriate value from their own data.
• Data extractivism is characterized as an architecture-mediated mechanism of value capture: architectures determine who benefits from pooled data and learned models.
• Cross-SME data pooling creates compounding learning rents (increasing returns to aggregated data) that accrue to the actors who control centralized infrastructures.
• The effect is amplified under particular conditions:
  • High switching costs for SMEs and users
  • Limited data portability or interoperability
  • Opacity in governance and model-training practices
• The framework reframes competitive advantage as architecture-dependent and positions SME performance as conditioned by ecosystem governance and technical design rather than by local resource ownership alone.

Data & Methods

• The paper is conceptual/theoretical, integrating multiple literatures rather than reporting new empirical data:
  • Resource-Based View (strategic management)
  • Value appropriation logic
  • Data-enabled learning and machine learning economics
  • Platform governance and digital ecosystems scholarship
• Methods consist of literature synthesis and development of a conceptual framework/model that explains mechanisms (data extractivism, architecture-mediated value capture) and boundary conditions (switching costs, portability, governance opacity).
• The model likely outlines propositions about how architecture control maps to emergent rents and SME outcomes; no primary quantitative or experimental data are reported in the abstract.

Implications for AI Economics

• Market structure and rents
  • Expect increased concentration of market power and rents around platform sponsors who control aggregation and model-training architectures.
  • Traditional measures of resource ownership misestimate who captures value; need to measure architecture control and aggregated-data advantages.
• Measurement & theory
  • Economic models should incorporate architecture-dependent learning returns and compounding effects from pooled datasets.
  • New empirical constructs: learning rents, architecture-control, data portability friction, governance opacity indices.
• Policy & regulation
  • Antitrust and competition policy may need to focus on architecture control, data portability, interoperability, and transparency of model training/monetization practices.
  • Interventions (e.g., mandated portability, open APIs, standardized governance disclosures) could reduce extractivism and rebalance value appropriation.
• SME strategy
  • SMEs should consider collective strategies (data pooling under shared governance, federated learning, co-ops), contractual safeguards, or alternative architectures to retain more value from their data.
  • Negotiating for portability and transparency clauses, or participating in interoperable ecosystems, can mitigate capture risk.
• Research directions
  • Empirical testing of compounding learning rents: quantify how pooled data increases model performance and firm-level rents.
  • Causal identification of architecture control on SME outcomes.
  • Comparative analysis of governance regimes and technical architectures (centralized vs. federated) on value distribution and welfare.