EventNet is a neural network architecture that can efficiently process asynchronous event streams from event cameras in real-time. It uses a temporal coding function to recursively update the network's state as new events arrive, avoiding redundant computation. Experimental results show it can perform tasks like target motion estimation and ego-motion estimation at rates over 1000 Hz using only the new event data. Compared to frame-based and PointNet approaches, EventNet significantly reduces computation time by recursively updating representations rather than reprocessing all prior events.