Abstract: In recent years, with the large-scale deployment of sensors, acquisition devices, and perception terminals, as well as the integration of new technologies such as Artificial Intelligence, 5G, and BeiDou, there are many intelligent applications in Power Internet of Things (PIoT), such as intelligent inspection, online monitoring, and demand response. These applications have generated massive amounts of sensory data. Uploading the data to cloud servers consumes a significant amount of communication bandwidth, placing immense pressure on network channels and cloud resources. Moreover, the data process cannot meet the real-time requirements of these applications. To address these issues, the concept of edge intelligence to PIoT has emerged. By combining edge computing with artificial intelligence, edge intelligence to PIoT uses AI algorithms to perform preprocessing, local computation, and inference near the data source. This approach reduces the communication bandwidth, and decreases transmission latency and power consumption. Edge Intelligence to PIoT provides an effective solution to the above problems. This article first explains the concept and evolution of edge intelligence to PIoT. Then, the architecture of edge intelligence in smart grid IoT is proposed, and the hardware and software foundations of edge intelligence to PIoT are analyzed from three levels: edge-side AI chips, edge computing operating systems, and edge computing frameworks. Next, key technologies in edge intelligence to PIoT are discussed from five aspects: cloud-edge collaboration, model compression, model acceleration, swarm intelligence, and federated learning. Furthermore, application scenarios for edge intelligence to PIoT are explored in the five aspects of "generation, transmission, transformation, distribution, and utilization". Finally, the opportunities and challenges of edge intelligence to PIoT are analyzed.