Scaling AI Sustainably: An Uncharted Territory

Explore the environmental implications and sustainability challenges of scaling AI technologies in this joint keynote address from USENIX ATC '24 and OSDI '24. Delve into the dramatic increase in compute power over the past 50 years and its impact on AI development. Examine the operational carbon footprint of AI computing throughout the model development cycle, including data, algorithms, and system hardware. Investigate the life cycle of system hardware, focusing on architectures and manufacturing technologies. Discover key efficiency optimization opportunities for cutting-edge AI technologies, from deep learning recommendation models to multi-modal generative AI tasks. Learn about the importance of making AI and computing more efficient and flexible to scale sustainably. Gain insights into optimizing across the entire life cycle of computing infrastructures, from hardware manufacturing to datacenter operation and end-of-life processing. Conclude with important development and research directions for advancing the field of computing in an environmentally responsible and sustainable manner, based on industry experience and lessons learned.