Google signs first deals to reduce AI energy use in high-demand hours

Google has signed its first agreements with U.S. power companies to cut down electricity use at its AI data centers when the power grid is under heavy load. This step is part of a larger effort to deal with growing energy challenges as AI systems continue to expand quickly.

The deals, made with Indiana Michigan Power and the Tennessee Valley Authority, let the power companies ask Google to temporarily slow down certain AI tasks, like machine learning, that use a lot of electricity. Google says this flexibility will make it faster to connect new data centers, lower the need for building extra power systems, and help keep the grid stable during heatwaves or times of high electricity use.

“Large electricity loads like data centers can now be interconnected more quickly,” said Google in a blog post, adding that the program benefits both grid operators and utility customers.

While cutting back on power use during peak times is common in industries like manufacturing and crypto mining, doing it specifically for AI tasks is something new. Google’s approach is based on earlier tests where it moved tasks to different places or times depending on when cleaner or cheaper energy was available.

The tech giant also said it is investing in workforce development and grid policy. In April 2025, Google announced a $10 million plan to train more electricians and backed efforts to speed up approvals for new types of energy like small modular reactors and geothermal power.

As AI energy use is expected to triple by the end of the decade, officials have warned that data centers might grow faster than the power supply in some parts of the U.S. Google’s move could set an example for other cloud providers, though similar deals haven’t been confirmed by Microsoft or Amazon.