Recent progress in computing, machine learning algorithms and networks, and the availability of large datasets for model training have created an enormous momentum in development and wide deployment of game-changing AI applications, with voice and vision being dominant ones. While AI computations historically been conducted in the cloud, the current trend is towards machine intelligence and analytics happening at the edge devices, such as smart phones and IoT devices, to meet privacy, power consumption, and latency requirements. Lower power consumption is an important factor for AI at the edge, yet it remains very challenging due power-hungry conventional hardware and high complexity of inference algorithms, let alone training on the device. In the extreme case of ultra-low power consumption, Qualcomm Technologies has pioneered an always-on machine vision technology combining innovations in the system architecture, ultra-low power designs, and dedicated hardware for computer vision algorithms running at the edge. This computer vision modele offers unique combination of benefits, such as less than 1 mW end-to-end power consumption, a tiny form factor, and low cost, that allow it to be integrated into a wide range of battery- and line-powered devices (IoT, mobile, VR/AR, automotive, etc.). It can perform object detection, feature recognition, change/motion detection, and other applications within the module’s processor itself and outputs only metadata.