The exponential growth of Artificial Intelligence (AI) and the Internet of Things (IoT) demands a paradigm shift in hardware design to ensure energy-efficient, sustainable computing. This talk will explore emerging trends in low-power AI hardware for Edge applications, for billions of autonomous systems operating under stringent energy constraints. These systems will drive advancements in Industry 4.0, autonomous robotics, personalized healthcare, and smart environmental monitoring. A key challenge lies in designing cognitive AI chips that integrate recent breakthroughs in 2D semiconductors, functional oxides, and neuromorphic computing architectures, seamlessly coupled with CMOS circuits. We will discuss novel energy-efficient computing paradigms inspired by the brain, leveraging neuromorphic computing, end-to-end spiking IoT nodes, memristive devices and novel adaptive materials, to push the boundaries of intelligence at the Edge. Some of targeted figures of merit include, but are not limited to: energy efficiency (e.g. <1 pJ per synaptic operation), computing throughput and density (e.g >10 TOPS/W), latency (e.g. sub-ms per inference step) and scalability (e.g. sub-10 nm devices with 3D integration potential). We cover critical aspects of sensing, processing, communication, and energy management for self-sustained AI hardware, aiming to bridge the gap between fundamental nanoscience and real-world AI deployment.