端侧AI行业｜AI终端爆发前夜，再看硬件升级逻辑

Core Viewpoint - The focus of AI competition has shifted towards terminal application scenarios, with a significant price reduction in large models driving innovation and expanding the user base and usage scenarios [3]. Group 1: AI Application Development - Since 2025, the competition in large models has transitioned into a price-competitive phase, stimulating innovation at the application level and attracting developers to AI application development [3]. - AI is empowering hardware in two main directions: enhancing traditional consumer electronics like smartphones and PCs to facilitate AI applications, and creating new hardware products such as the AI Pin by Humane, which interacts through voice and gestures [4]. Group 2: AIoT Ecosystem - AI technology is crucial for the realization of AIoT, transforming the early IoT focus on simple device connectivity into intelligent connections that enable devices to make autonomous decisions [4]. - AI algorithms, including deep learning and natural language processing, are essential for extracting digital value from the vast amounts of data generated by terminal devices, advancing IoT from mere connectivity to intelligent interconnectivity [4]. Group 3: Future Directions in AI Hardware - The main directions for the evolution of edge AI devices include more efficient model architectures, integrated storage and computing, edge-cloud collaborative inference, and privacy enhancement technologies [5]. - The performance of AI semiconductor hardware, including power consumption, low latency, and environmental resilience, along with semiconductor architecture design, are core elements for hardware evolution [5]. Group 4: Competitive Landscape in AI Semiconductor - The edge AI semiconductor market is becoming competitive, with domestic manufacturers gaining traction in wearable and smart home sectors, leveraging their advantages in audio and video processing [6]. - The next generation of memory technologies, such as MRAM and ReRAM, is nearing practical application, and domestic manufacturers are actively pursuing customized storage solutions to meet the specific needs of edge AI [6].