AI object detection In 2021, the number of types increased to 80, and by 2024, it had grown to 21,000. Now, in 2025, AI can describe what people are doing and how they are doing things in an image. This is a dramatic advancement in just a few years. This means that the amount of computation required is also increasing at an incredible rate. On the other hand, there are embedded markets such as automotive and industrial, where cloud computing is not available and on-site operation is required. Resolving the contradiction between the limited resources of these edge devices and the increasing amount of computational power will be a major theme for edge AI in the future. Chairman Terada of airia Inc., who is working on this theme based on an inference framework developed in-house, will discuss solutions to these issues, including case studies, along with trends in the AI market.