Innovation ecosystems work through repeated organizing, not just static links: viewing ecosystems as socio-technical organizing processes reveals how day-to-day orchestration, evolving interdependencies, and institutional boundaries drive diffusion, market structure, and governance in AI and beyond.

Main Finding

Innovation ecosystems are best understood not primarily as static structural arrangements (networks, platforms, portfolios) but as ongoing socio-technical organizing processes. The paper defines innovation ecosystems as socio-technical organizing processes oriented to a shared value proposition under conditions of uncertainty and interdependence, and develops an integrated organizational vocabulary—centered on interdependence, orchestration, and institutional boundaries—to explain how coordination is practically accomplished and renegotiated across heterogeneous actors, technologies, and contexts.

Key Points

• Conceptual reframing: shifts the analytic focus from mapping “who is connected to whom” toward how actors repeatedly organize, coordinate, and renegotiate roles and resources in practice.
• Definition offered: innovation ecosystems = socio-technical organizing processes oriented to a shared value proposition under uncertainty and interdependence.
• Three complementary lenses for research:
  • Interdependence as relational and sociomaterial dynamics (how actors, technologies and artifacts co-shape dependencies).
  • Orchestration as governance-in-practice (how coordination, standards, incentives and power are exercised day-to-day).
  • Contextual boundaries as institutional embeddedness (how regulatory, normative and cultural contexts delimit and shape organizing).
• Integration formalized in two propositions (summarized):
• Ecosystem outcomes are jointly shaped by the sociomaterial patterning of interdependencies, the practices of orchestration, and the institutional boundaries that condition actors’ options.
• An organizing perspective yields additional explanatory power—beyond structural descriptions—when ecosystems face high uncertainty, strong complementarities, and heterogeneous institutional environments.
• Boundary conditions: the organizing perspective is most useful when (a) uncertainty is high, (b) interdependence is complex and changing, and (c) institutional contexts vary; less added value where interactions are routine, modular, and tightly codified.
• Provides a theory-informed research agenda emphasizing process-sensitive, comparative, and sociomaterial empirical work.

Data & Methods

• Paper type: conceptual/theoretical.
• Methodological approach: comparative and integrative literature review synthesizing innovation-ecosystem scholarship with organizational theories—particularly processual, sociomaterial, and institutional perspectives.
• Analytic moves: critique of prevailing structural framings; articulation of a processual definition; development of three lenses; formal integration into propositions; specification of boundary conditions; derivation of empirical research directions.
• Empirical recommendations (from the paper): longitudinal and process-tracing methods, comparative case studies, ethnography of organizing practices, sociomaterial analysis of artifacts and standards, and mixed-methods designs that capture governance-in-practice.

Implications for AI Economics

• Conceptual clarity for AI ecosystems: reframing AI ecosystems as organizing processes highlights dynamic coordination problems around data, models, compute, standards, and complementary services—moving beyond static maps of firms or platforms.
• Research directions and questions:
  • How do orchestration practices (by big tech platforms, standard bodies, public actors) shape diffusion, pricing, and market structure in AI? What governance-in-practice emerges for safety, model evaluation, and access to compute/data?
  • How do sociomaterial interdependencies (data pipelines, model APIs, tooling, hardware) create complementarities and bottlenecks that affect entry, competition, and specialization?
  • How do institutional boundaries (regulation, procurement rules, intellectual property regimes, cross-border data governance) influence AI investment, geographic specialization, and the global division of labor in AI development?
  • Under what conditions do organizing practices mitigate or exacerbate coordination failures, hold-up problems, and public-good/externality challenges (e.g., model safety, public datasets)?
• Methodological implications for empirical AI economics:
  • Use process-sensitive methods: longitudinal firm-level case studies, event studies around orchestration interventions (standards, regulations), network and temporal analyses of collaborations, ethnographies of platform governance, and process mining of development workflows (e.g., GitHub, model registries, API logs).
  • Combine administrative/transactional data (compute usage, API calls, licensing agreements), repo metadata (code and model evolution), and qualitative data (interviews, meeting traces) to capture governance-in-practice and sociomaterial coupling.
  • Leverage computational models (agent-based models, dynamic network models) that embed evolving interdependencies and governance practices to simulate policy interventions and market dynamics.
• Policy and managerial implications:
  • Policy design should attend to orchestration mechanisms (incentives, standards, intermediaries) that shape coordination in AI ecosystems, not only firm-level competition metrics.
  • Antitrust and industrial policy analyses should incorporate organizing processes—how dominant firms orchestrate ecosystems via vertical integration, platform governance, or standards—to assess market power and lock-in more accurately.
  • For innovation policy, interventions (public data, compute subsidies, standards-setting) need to consider institutional embeddedness and how local practices of organizing affect uptake and spillovers.
• Practical hypotheses for AI economics studies (examples):
  • Regions with stronger public orchestration institutions (procurement, standards agencies) will exhibit faster diffusion of safe AI practices and different specialization patterns than regions relying on market orchestration alone.
  • Firms that successfully perform orchestration (by setting de-facto standards, controlling key interfaces, or provisioning shared infrastructure) capture larger shares of downstream rents even when structural networks appear dispersed.
  • Sociomaterial bottlenecks (e.g., scarce fine-tuning compute, exclusive datasets) predict entry barriers and concentration more strongly than measures of firm count or connection density.

Overall, treating AI ecosystems as organizing processes redirects empirical AI economics toward studying the dynamics of coordination, governance-in-practice, and institutional context—yielding deeper explanations for diffusion, competition, and welfare outcomes than structural descriptions alone.