How Do Commercial Interests and Resource Accessibility Shape Open-Source LLMs: Evidence from Hugging Face

Abstract

The emergence of open-source large language models (LLMs) has introduced new challenges to the open-innovation paradigm. Unlike earlier open-source software (OSS), developing LLMs requires substantial computing resources and involves tasks that are not easily divisible. Drawing on data from Hugging Face, we show that open-source LLM development is heavily influenced by commercial interests and disparities in resource accessibility. In particular, improvements in model performance are primarily driven by external contributors affiliated with commercial entities that possess significant computational capacity, and only foundation models released under licenses permitting downstream commercialization benefit from meaningful outside contributions. Interestingly, while OSS contributors can accelerate model adoption, they often channel users disproportionately toward their own derivative models rather than the foundation model, thereby creating a more fragmented ecosystem. This fragmentation—and the resulting erosion of foundation-model market share—is especially pronounced when the OSS contributor has strong commercial interests or access to advanced computing resources. Taken together, these findings highlight a distinct cooperation-competition dynamic between LLM organizations and OSS contributors.