Google has begun throttling data center power consumption using AI to support the US power grid during peak hours

The rise of AI data centers comes with a huge energy consumption. Powering these data centers has become a major challenge for companies. In some cases, this has led to public backlash in the regions where these facilities operate. As a result, Google has begun exploring solutions to reduce the load on the U.S. power grid during periods of peak demand.

The company's demand response program was launched in 2023 as a way to reduce the load on the U.S. power grid when needed. Google says the initiative allows large energy-consuming facilities, such as data centers, to come online faster and helps reduce the need to build new transmission lines and power plants. It also allows operators to manage their power grids more efficiently and effectively.

The agreements with the two utilities mark the first known instance of energy consumption being reduced by reducing the amount of computation associated with machine learning.

“This builds on our successful collaboration with the Omaha Public Power District, where we reduced power consumption associated with machine learning workloads during three grid congestion events last year, paving the way for us to implement similar initiatives in other locations,” Google explained.

This spring, the International Energy Agency predicted that energy demand from AI-powered data centers would quadruple by 2030. By the end of the decade, the technology will consume almost as much electricity as Japan does today.