IBM has introduced a more energy-efficient phase-change chip, designed for use in generative AI applications such as large language models, significantly reducing power requirements. This chip can integrate memory and processing functions through phase-change memory, achieving reasonably accurate voice recognition with lower energy consumption. The core component of the new chip is called a "tile," which is an array of crossbars composed of 512 units wide and 2048 units deep individual phase-change bits. The chip also includes high-speed communication hardware necessary for all bits, allowing communication across different tiles without the need for any analog-to-digital conversion. While this chip may not be particularly useful for training AI, it can significantly reduce energy consumption when matched with the right type of problems.