We develop a model with multiple sectors connected through both knowledge and physical input-output networks, in which network structures determine economic dynamics and constitute systematic risk in asset pricing. The model predicts that the sparsity of knowledge network increases aggregate consumption and leads to positive risk premium. Our model implications receive empirical support as follows: (i) consumption growth increases with changes in knowledge network sparsity; (ii) firms with higher exposures to knowledge network sparsity carry higher expected stock returns; and (iii) knowledge network sparsity exists in the stochastic discount factor and helps price the cross section of stock returns.