Edge AI has emerged recently as the industry tries to overcome the limitations of cloud-based AI, such as network latency, data security, and related costs. Enabling intelligence through processing directly at the data source will provide faster and more efficient industrial applications. However, embedded devices at the edge have less computing power and energy sources than servers in the cloud. For this reason, hardware accelerators for edge AI have been developed to provide additional computing power with low energy consumption. This article compares the performance and energy consumption of two commercially available AI accelerators with the embedded GPU, and factors affecting their performance are discussed.