The context of the Internet of Vehicles (IoV) and the growing vehicular deployment of sensors and smart devices have demanded alternative wireless communication technologies. Despite the prevalent works on radio-frequency (RF)-based vehicular communication technologies, visible-light communication (VLC) is envisaged to be a promising candidate due to its unlicensed spectrum and high transmission rates. However, the long-distance and high-speed transmission performance of vehicular VLC needs more investigation considering the massive data transmission needs for automotive applications. In this article, a vehicular multiple-input–multiple-output (MIMO) VLC system based on two commercial headlights and a self-designed PIN array is experimentally demonstrated as a proof of concept for the IoV. Considering the relative movement of vehicles, different communication areas are divided based on the position-dependent receiver receiving states. The boundary condition of each area is derived. For the first time, the selection of the strategy that best suits the MIMO demultiplexing scheme is discussed by analyzing the rank and type of the channel matrix. A modified pilot-aided phase recovery method based on polynomial curve fitting (PCF) is proposed to compensate for the phase noise caused by the sampling frequency offset (SFO). The proposed method maintains good robustness even when the VLC system suffers from the strong nonlinearity and bandwidth limitation of the headlights. Based on the proposed MIMO VLC system, we achieve a record-breaking data rate of 3.08 Gb/s at a 2-m indoor transmission link. We further extend the transmission distance to 100 m, and successfully achieve overall data rates of 336 and 362 Mb/s during the daytime and nighttime, respectively. To the best of our knowledge, these are the highest transmission data rates ever reported for a 100-m vehicular VLC system. Our experimental demonstration clearly verifies the feasibility of VLC in the application of IoV.