The Evolution of Competition in the Market for AI Models: From Data-Enabled Learning to Platform Ecosystems — A Theoretical Analysis

While the current AI market shows signs of a fairly competitive environment, with a large number of foundational models and apps developed over the last few years, researchers still question the sustainability of these trends. The objective of this paper is to analyze how platformization might affect the competitive dynamics and efficiency outcomes in the market for AI foundational models. Previous research shows that when firms are one-sided and data-enabled learning does not create a consumer coordination problem or biased belief formation, the equilibrium outcome is efficient. We extend this model by introducing a platform strategy in which firms open their foundational models to complementary application developers, creating indirect network effects between users and developers. The theoretical analysis identifies a "leapfrogging" scenario where a firm with an inferior learning function can overturn a more efficient rival by platformizing first and implementing a seeding strategy (in-house apps) to lock in both sides. This suggests that while data plays an important role in the competitiveness of AI foundational models, a correctly timed platform strategy might lead to inefficient lock-in.