This paper articulates the hypothesis that there is an optimal size of knowledge pools. Too little a density of innovation activities reduces the accessibility of external knowledge. Too large a density enhances congestion and reduces appropriability. Firms can benefit from actual increasing returns stemming from the indivisibility, replicability and non-exhaustibility of knowledge only when the size of innovation networks is comprised between the two extremes. The empirical evidence confirms that the output elasticity of knowledge, included in a typical Griliches production function, is itself a quadratic function of the size of innovation networks. Knowledge externalities do trigger increasing returns that are external to each firm, only within a well defined interval. Knowledge externalities are a property of the system into which firms are embedded. As such they are endogenous to the system and likely to exhibit specific properties related to the changing characteristics of the system itself. The quality of knowledge governance mechanisms in place plays a key role in assessing the actual size of the net positive effects of knowledge externalities.