As foundation models become rentable utilities, AI ceases to be a rare capability and instead becomes table stakes; durable advantage for startups now lies in proprietary data, tacit domain expertise, customer relationships, and the organizational skill to orchestrate AI into distinctive offerings.

Main Finding

When foundation generative-AI models become broadly and cheaply accessible, the direct source of competitive advantage shifts away from the commoditized model and toward co-specialized complements — proprietary and hard-to-replicate data (especially feedback loops), domain judgment and tacit expertise, customer relationships/trust, and the firm-level orchestration and absorptive capacities that turn a common model into a distinctive value proposition. Advantage thus "relocates" rather than disappearing, but AI-derived rents erode faster and depend more on dynamic reconfiguration than one-time investments.

Key Points

• Commoditization is of capability, not just infrastructure: unlike earlier IT, generative AI can replicate cognitive work and so compresses differences in capabilities that once conferred advantage.
• Resource-based logic: a commoditized model fails the RBV rarity and inimitability conditions; the model itself is unlikely to sustain durable advantage.
• Advantage-relocation mechanism: value capture migrates to complementary assets that are scarce, co-specialized, and hard to trade (e.g., proprietary feedback-loop data, specialized human judgment, distribution channels).
• Absorptive capacity matters: firms with prior related knowledge and managerial competence (to recognize, assimilate, and apply model outputs) extract more value from identical models — orchestration capability becomes a central differentiator.
• Dynamic capabilities are critical: because model performance and access evolve rapidly, firms need sensing/seizing/reconfiguring capabilities to sustain advantage; AI-derived advantage is more rapidly eroded than advantage based on durable assets.
• Layered market structure: end-user access to models is largely commoditized, but training and frontier-model provision remain concentrated (large providers), producing asymmetric rents across layers.
• Empirical nuance on data: raw data alone is a weak moat; data embedded in proprietary feedback loops or tightly integrated with business processes is a stronger, co-specialized complement.
• Countervailing differentiation (fine-tuning, retrieval augmentation, vertical stacks) still depends on scarce complements (e.g., proprietary data, integration skill) — these do not overturn the relocation thesis but illustrate it.

Data & Methods

• Paper type: conceptual, theory-building contribution (no primary empirical data).
• Methodological approach: purposive integrative literature review synthesizing four theoretical strands — resource-based view (RBV), complementary-assets theory, absorptive capacity, and dynamic capabilities — and engaging the information-technology commoditization debate as precedent.
• Sources: peer-reviewed and authoritative work (2023–2026) on generative AI, strategy, and IS literature; working papers and practitioner material used to characterize emerging phenomena.
• Analytical design: deductive derivation of falsifiable propositions from integrated theory; emphasis on startups as a sharp boundary condition (resource-constrained firms most affected by capability commoditization).

Implications for AI Economics

• Distribution of rents: returns shift from model-building capital (compute, frontier-model IP) to complements — specialized data-feedback loops, customer access, and orchestration capabilities — altering where firms and investors should expect long-term rents.
• Entry and competition: generative AI lowers technological barriers to entry (reducing incumbents' advantage from in-house models) but raises the importance of non-imitable complements, potentially increasing winner-take-most dynamics around companies that secure co-specialized data or distribution.
• Factor returns: wages/premia may shift from large pools of model engineering toward roles and skills tied to domain expertise, product orchestration, customer relationship management, and continuous learning (dynamic capabilities).
• Investment strategy: firms (esp. startups) should prioritize building feedback-rich data-generation processes, industry/domain expertise, customer lock-in mechanisms, and absorptive/orchestration capabilities rather than attempting to compete on base-model ownership.
• Market structure and policy: concentration risk at the frontier-model provision layer suggests regulatory and antitrust considerations distinct from user-level commoditization; data governance and portability rules could materially affect where advantage locates.
• Empirical agenda for AI economics:
  • Measure and operationalize absorptive and orchestration capacities to test their effect on value capture from identical model access.
  • Quantify the value of feedback-loop vs. static proprietary data across industries and applications.
  • Longitudinal studies to estimate erosion rates of AI-derived rents and to observe dynamic-capability-driven renewals of advantage.
  • Natural experiments (API outages, price shocks, model provider changes) to identify the causal role of model access vs. complements.
  • Industry heterogeneity analysis: which sectors exhibit stronger relocation to complements (e.g., regulated, tacit-knowledge-intensive, or customer-interaction-heavy industries).
• For macro and policy modelers: incorporate two-tier dynamics (concentrated model provision + commoditized user access) and differential capital returns to complements when projecting market concentration, innovation incentives, and labor-market impacts.

Suggested falsifiable propositions (from the paper) - Competitive advantage will be positively associated with ownership or control of co-specialized complementary assets (proprietary feedback-loop data, domain expertise, customer channels) when foundation-model access is equivalent across firms. - Among firms with equivalent model access, greater absorptive capacity and orchestration capability predict higher value extraction from generative AI. - AI-derived competitive advantages erode faster than advantages rooted in durable, non-commoditized assets absent continual reconfiguration (dynamic capabilities).

If you want, I can convert this into a one-page infographic-style brief for investors or a short checklist founders can use to assess whether their startup is positioned to capture relocated AI value.