EPRI and Epoch AI Report Highlights Rising Power Needs for AI Model Training

The Electric Power Research Institute (EPRI) and Epoch AI have released a joint report indicating a significant increase in power demand due to the training of large-scale AI models. Announced in a press release, the report forecasts that by 2030, training a leading AI model could require more than 4 gigawatts (GW) of power, enough to supply millions of U.S. homes.

Despite advancements in computational efficiency, the power demands for training AI models have more than doubled annually over the past decade. The report suggests that the AI industry will continue to scale up its models, further increasing the need for electrical power. Currently, the total AI power capacity in the U.S. is estimated at around 5 GW, with projections indicating it could exceed 50 GW by 2030.

EPRI President and CEO Arshad Mansoor emphasized the importance of innovative solutions to meet these rising energy demands. The organization has launched the DCFlex collaborative to explore data center flexibility, aiming to transform data centers into grid assets to improve reliability and reduce costs. This initiative involves over 45 companies, including major tech firms like Google, Meta, and NVIDIA, and has begun field demonstrations in various locations.