Exploring the Heterogeneous Knowledge Spillover Effects from University AI Research on the Creation and Performance of AI Start-ups

This paper investigates whether and which type of AI academic research at universities can create regional knowledge spillover effects that improve the quantity and quality of AI start-ups. Using data of AI start-ups from Crunchbase.com and AI conference publications from CSRankings.com, we find that knowledge spillovers from university AI research indeed contribute to the creation and VC financing performance of local AI start-ups at the MSA level in the United States. Moreover, we find significant heterogeneous effects of knowledge spillovers in different AI subfields. The knowledge spillovers from research published in machine learning conferences, including only ICML and NIPS, have the strongest effects on the creation and performance of AI start-ups. In addition, we find evidence that impactful conferences exhibit stronger spillover effect. At the same time, surprisingly, our results show that knowledge spillovers from theoretical AI research have stronger effects on the creation of start-ups.