This article shows how innovation in chip architectures and hardware accelerators is enabling AI at the edge. While each architecture has its merits, it’s critical to consider how these platforms impact the compute performance, power consumption, and latency of neural network operations and systems as a whole.