Why Does AI Platform Competition Diverge? A Dynamic Analysis of Adoption, Recommendation Mechanisms, and Price Feedback

Generative AI and agentic systems are reshaping platform competition and its long-run welfare consequences. Existing research, however, does not explain why similar AI interventions can produce concentration, costly quality races, cyclical repricing, and different welfare outcomes. A dynamic model distinguishes traditional consumers from AI-mediated consumers and characterizes adoption thresholds, recommendation-driven quality boundaries, price-feedback bifurcations, and welfare across long-run regimes. AI adoption interacts with network effects to narrow the range of stable symmetric competition. Recommendation amplification lowers the threshold for asymmetric quality states and causes profit-eroding quality competition to arise earlier. Price feedback can generate stable local cycles around the symmetric state and resilience thresholds around asymmetric leadership states. When quality and price feedback coexist, more dispersed or more active competition need not improve the model-implied welfare measure. Subject to the focal-episode assumptions, the model identifies adoption, recommendation amplification, and price feedback as distinct sources of regime divergence. It thereby explains why otherwise comparable AI-platform environments can develop different market structures and welfare rankings.