The model predictive path integral (MPPI) control algorithm is applied to fixed-wing aircraft. Because MPPI imposes no restriction on the form of the cost function, arbitrarily complex maneuvers can be crafted through cost function design. MPPI works by propagating (thousands of) trajectories forward in time using random control inputs. This procedure is done in parallel using a graphics processing unit (GPU). The optimal control is computed by weight-averaging the controls from these trajectories (with each weight corresponds to the respective trajectory cost). A nonlinear wind-axes aircraft model is used in the MPPI propagation step. This choice of model offers two benefits. First, the model remains nonlinear which is appropriate for aggressive flights. Second, the model does not require parameters. Prediction error due to parametric uncertainty can be eliminated. We demonstrate our method in a simulated air racing scenario. Despite the simulated result, the real-time feasibility is validated because 1) the implementation is done in a flight-capable software framework, and 2) the simulation is done in a high fidelity flight dynamics simulator. We show that MPPI offers superior performances compared to a typical waypoint guidance. These include a smaller altitude drop in tight turns, a better ability to stay on course, and a lower maximum load factor.