With blockchain technology getting wide adoption, digital platforms built on blockchains have also dramatically increased in scale. One crucial determinant of platforms’ performance is its governance structure, specifically, its degree of decentralization. Distinct from the most prevalent centralized governance in most corporations, blockchain platforms naturally carry certain elements of decentralization, which provides a unique opportunity to examine and compare the effectiveness of governance decentralization. In this study, using regression, machine learning, and causal inference, I show the inverted U-shaped relationship between blockchain platforms’ degree of decentralization and their performance, measured by several proxy variables such as developer engagement and market capitalization. Semi-decentralization is shown to be the ideal degree of decentralization. Besides, differences in the results for layer-1 and layer-2 applications indicate characteristic differences between the two blockchain layers.