Equilibrium asset pricing models, especially with limited liquidity, can usually be characterized by systems of forward-backward stochastic differential equations (FBSDEs). Although global equilibrium is achieved under specific market dynamics, the nonlinear system of fully coupled FBSDEs falls outside the scope of any known well-posedness results. In this talk, we show how to leverage deep-learning techniques to obtain numerical solutions with calibrated parameters to market prices and trading volumes. In particular, we propose the novel usage of generative adversarial networks (GANs) as a numerical algorithm for equilibrium models, where GANs not only overcome the curse-of-dimensionality and but also show their great scalability.