AsianFin -- Taiwan Semiconductor Manufacturing Co. (TSMC), the world’s largest contract chipmaker, outlined a new strategy on Wednesday to make AI computing chips up to 10 times more energy efficient, leveraging artificial intelligence to design the semiconductors themselves.

Speaking at a conference in Silicon Valley, TSMC showcased advances that could help reduce the massive energy demands of AI data centers. Current Nvidia flagship AI servers can consume as much as 1,200 watts during intensive tasks — equivalent to the continuous electricity use of roughly 1,000 U.S. homes.

To address this, TSMC is focusing on a new generation of chip architectures built around “chiplets” — smaller components that can be fabricated with different technologies and then packaged together into one computing system. This modular approach is designed to improve both performance and efficiency compared with traditional monolithic chip designs.