Organisations are urged to place AI model training in regions powered by renewable energy. For instance, training in Canada rather than Poland could reduce carbon emissions by approximately 85 %. Strategic location choices are becoming vital for greener AI.
Selecting appropriate hardware also plays a pivotal role. Research shows that a high-end NVIDIA H100 GPU carries three times the manufacturing carbon footprint of a more energy-efficient NVIDIA L4. Opting for the proper GPU can deliver performance without undue environmental cost.
Efficiency should be embedded at every stage of the AI process. From hardware procurement and algorithm design to operational deployment, even fractional improvements across the supply chain can significantly reduce overall carbon output, ensuring that today's progress doesn't harm tomorrow's planet.
