At the core of some of the most important problems in plasma physics—from controlled nuclear fusion to the acceleration of cosmic rays—is the challenge to describe nonlinear, multiscale plasma dynamics. The development of reduced plasma models that balance between physical accuracy and computational complexity is critical to advancing theoretical comprehension and enabling holistic computational descriptions of these problems. In this talk, I will discuss the possibility of using data-driven techniques to develop accurate reduced plasma models (in the form of partial differential equations) directly from the data of first-principles particle-in-cell simulations. In particular, I will discuss 1) how data-driven model-discovery techniques based on sparse optimization can be used infer interpretable reduced plasma models, 2) an effective strategy for robust model identification in the presence of the high data noise that is intrinsic to first-principles particle-based simulations, and 3) strategies for embedding fundamental physics symmetries in the model-discovery methodology. I will demonstrate the potential of this approach by recovering the fundamental hierarchy of plasma physics models—from the Vlasov equation to single-fluid magnetohydrodynamics. I will end with an outlook on how this data-driven methodology offers a promising route to accelerate the development of reduced theoretical models of complex nonlinear plasma phenomena and to design computationally efficient algorithms for multiscale plasma simulations.