Software Startups’ Patenting And Venture Capital Responses During The Early Years Of Machine Learning

Abstract

This paper investigates how investors value patents in a changing technological and legal environment in the IT software ecosystem. We propose that appropriability theory and signaling theory yield divergent predictions about investor preferences, creating a core tension in the evaluation of upstream versus downstream startups in the industry. Drawing on a sample of 3,175 U.S. IT software startups founded between 2005 and 2017 during the development of machine learning, one of the AI precursors, the study examines the "patent-funding link". To address potential endogeneity, we use a two-stage least squares (2SLS) identification strategy, instrumenting patent approval with quasi-random examiner leniency. The findings reveal that a first patent grant increases the likelihood of securing external funding by approximately 16% for upstream startups. This contradicts traditional appropriability logic, which suggests that value capture shifts downstream toward application holders, and instead aligns with signaling and market-for-technology theories, which predict that investors would fund upstream startups on the basis of their tangible signals and contractible assets, such as patents. The effect increases to 27% when the patent specifically involves computational claims (e.g., core algorithms and architecture). These results suggest that in the unsettled early stages of a GPT, patents function less as barriers to entry and more as high-fidelity signals of technical. This study contributes to the literature on startup financing by demonstrating that upstream innovation remains a primary driver of venture-capital interest.