Model predictive control (MPC) is essential to optimal decision making in a broad range of applications like building energy management and autonomous racing. MPC provides significant energy cost savings in building operations in the form of energy-efficient control with better occupant comfort, lower peak demand charges, and risk-free participation in demand response. In autonomous racing, MPC computes a safe minimum-time trajectory while driving at the limit of a vehicle’s handling capability. However, the ease in controller design depends upon the modeling complexity of the underlying physical system. For example, the identification of physics-based models of buildings is considered to be the biggest bottleneck in making MPC scalable to real buildings due to massive engineering effort. Thus, the traditional modeling approaches like the white-box and the grey-box techniques, although detailed, are considered cost and time prohibitive. In the case of autonomous racing, one of the fundamental challenges lies in predicting the vehicle’s future states like position, orientation, and speed with high accuracy because it is inevitably hard to identify vehicle model parameters that capture its real nonlinear dynamics in the presence of lateral tire slip. To this end, we present methods for data-driven MPC that combine predictive control and tools from machine learning such as Gaussian processes, neural networks, and random forests to reduce the cost of model identification and controller design in these applications. First, we introduce learning and control algorithms for building energy management based on black-box modeling that require minimum external intervention and solve some of the fundamental practical challenges ranging from experiment design to predictive control to online model update. We learn dynamical models of energy consumption and zone temperatures with high accuracy, and demonstrate load curtailment during demand response, energy savings during regular operations, and better occupant comfort compared to the default system controller. We validate our methods on several buildings in different case studies, including a real house in Italy. Next, we present a model-based planning and control framework for autonomous racing based on discrepancy error modeling that significantly reduces the effort required in system identification of the vehicle model. We start with an easy-to-tune but inaccurate physics-based model of the vehicle dynamics and thereafter correct the model predictions by learning from prior experience. Our approach bridges the gap between the design in a simulation and the real world by learning from on-board sensor measurements. We demonstrate its efficacy on a 1/43 scale autonomous racing simulation platform.