The Economics of AI Foundation Models: Openness, Competition, and Governance

Posted: 20 Feb 2024 Last revised: 30 May 2024

See all articles by Wei Chen

Wei Chen

University of Connecticut - Department of Operations & Information Management

Xiaoyu Wang

Washington University in St. Louis - John M. Olin Business School

Karen Xie

University of Connecticut - Department of Operations & Information Management

Fasheng Xu

University of Connecticut - Department of Operations & Information Management

Date Written: February 14, 2024

Abstract

AI is undergoing a paradigm shift with the rise of foundation models (e.g., GPT-4, Claude 3, Gemini, Llama 3, Stable Diffusion) trained on broad data using self-supervision at immense scale, which can then be adapted to myriad downstream tasks. This paper offers an economic theory of foundation model ecosystems that consist of upstream model developers and downstream model deployers. We model how fine-tuning games arise as downstream deployers competing in foundation model adoption and fine-tuning effort, and how model openness affects the fine-tuning games. Our theory provides insights into the economic implications and trade-offs for multiple stakeholders throughout the AI value chain (including developer, deployers, and consumers) and sheds light on how to harness the full potential of the foundation model value chain interactions and avoid pitfalls. An important implication of our findings is that the policymakers should explicitly consider potential unintended economic consequences of AI regulation on the ecosystem around open foundation models, in particular, the so-called "openness trap" (i.e., a range of medium openness levels that should be avoided). Furthermore, we explore the welfare implications of prevalent market strategies employed by upstream developers, such as vertical integration and offering free trials. Our findings reveal that vertical integration proves effective when model openness is relatively limited. The developer's strategy of providing free trials can negatively affect the leading deployer within a moderate range of model openness, while benefiting all other stakeholders.

Keywords: Foundation models, large language models (LLMs), GPTs, generative AI, model openness, fine-tuning, competition, AI governance, AI value chain

Suggested Citation

Chen, Wei and Wang, Xiaoyu and Xie, Karen and Xu, Fasheng, The Economics of AI Foundation Models: Openness, Competition, and Governance (February 14, 2024). Available at SSRN: https://ssrn.com/abstract=4727903

Wei Chen

University of Connecticut - Department of Operations & Information Management ( email )

1 University Pl
Stamford, CT 06902
United States

Xiaoyu Wang

Washington University in St. Louis - John M. Olin Business School ( email )

One Brookings Drive
Campus Box 1133
St. Louis, MO 63130-4899
United States

Karen Xie (Contact Author)

University of Connecticut - Department of Operations & Information Management ( email )

1 University Pl
Stamford, CT 06901
United States

Fasheng Xu

University of Connecticut - Department of Operations & Information Management ( email )

1 University Place
Stamford, CT 06901
United States

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Abstract Views
582
PlumX Metrics