Market Games for Generative Models: Equilibria, Welfare, and Strategic Entry
Xiukun Wei, Min Shi, Xueru Zhang
TL;DR
This paper models competitive multi-platform markets for generative models, analyzing equilibria, welfare, and strategic entry, revealing how market structure and model diversity impact social welfare and platform strategies.
Contribution
It formalizes a three-layer market game, identifies conditions for pure Nash equilibria, and introduces best-response training schemes for strategic model entry.
Findings
Market structure depends on model performance and user attraction.
Expanding model pools does not always improve welfare or diversity.
New training schemes enable strategic model introduction.
Abstract
Generative model ecosystems increasingly operate as competitive multi-platform markets, where platforms strategically select models from a shared pool and users with heterogeneous preferences choose among them. Understanding how platforms interact, when market equilibria exist, how outcomes are shaped by model-providers, platforms, and user behavior, and how social welfare is affected is critical for fostering a beneficial market environment. In this paper, we formalize a three-layer model-platform-user market game and identify conditions for the existence of pure Nash equilibrium. Our analysis shows that market structure, whether platforms converge on similar models or differentiate by selecting distinct ones, depends not only on models' global average performance but also on their localized attraction to user groups. We further examine welfare outcomes and show that expanding the…
Peer Reviews
Decision·ICLR 2026 Poster
The topic is timely and relevant, theoretical analysis is nice and solid. Presentation is easy to follow. Despite the weaknesses I pointed out below, I really like the angle from which the paper formulates the problem and the style of presentation.
As a theoretical-oriented paper with the claimed strength lying on the proposed game-theoretic model, my main concern is that the model structure and theoretical finds are not sufficiently interesting for providing new insights. Here is some of my thoughts: 1. The model seem to me is too stylistic. The hardmax user choice model is overly simplified, it would be more interesting to consider softmax or other alternative stochastic choice model and see if similar observation holds. If hardmax is a
The main strengths are: 1) Originality in model formulation: The paper introduces a three-layer model, platform, user framework to study competition in generative AI markets. Prior work, typically focuses on two-layer settings involving only users and platforms/models. Their framework captures the distinct incentives at each layer and the competitive interactions (among platforms and among model providers). 2) Section 3 is a highlight of the paper. It provides a clear and rigorous characterizat
1. Platforms Limited to a Single Model: The modelling assumes that each platform selects one model provider. This does not reflect the papers motivation where platforms like Azure and Bedrock host multiple foundation models simultaneously. As a result, the framework cannot capture strategies such as model bundling, which are important for platform differentiation and for covering diverse user needs. This is not a major weakness, but it would be useful if the authors could discuss how multiple-m
Originality: The three-layer model formulation is a timely and non-trivial extension of prior work, better capturing the structure of modern generative AI ecosystems. Theoretical Rigor: The analysis of equilibrium conditions, linking market structure to the balance between average performance and deviation advantage, is a solid theoretical contribution. Holistic Approach: The paper cohesively analyzes the market from platform, user, and (to a limited extent) model provider perspectives, pr
Strong and Potentially Unrealistic Assumptions: The model relies on several strong assumptions that limit the practical applicability of its conclusions. Complete Information: The model assumes that platforms and model providers have perfect knowledge of the user type distribution π_θ and their reward functions r_θ(x). This ignores the significant challenge of preference learning and the strategic implications of information asymmetry in real-world markets. Deterministic User Choice: The har
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsDigital Platforms and Economics · Auction Theory and Applications · Game Theory and Applications