Google to Reduce AI Data Center Power Use During Peak Demand

Google has announced new partnerships with two US power utilities, Indiana Michigan Power (I&M) and the Tennessee Valley Authority (TVA), aimed at cutting back electricity use from its artificial intelligence data centers during periods of high demand.

The move marks a pivotal moment in Big Tech’s relationship with the US power grid, as concerns grow over the ballooning energy needs of AI infrastructure.

The company made the announcement on Monday through a blog post by Michael Terrell, Google’s Head of Energy Development, in which he described the strategy as part of a broader effort to integrate “demand response” — a method of temporarily reducing electricity usage during peak hours — into its operations.

AI power demands are outpacing supply

AI models are growing in complexity, and training or running them can consume tens or even hundreds of megawatts of power, sometimes continuously for hours or days. Meanwhile, US grid operators have been sounding the alarm, as electricity requests from tech companies surge past local supply capacity in several states.

In the agreements with I&M and TVA, Google has committed to rescheduling or pausing non-urgent AI workloads, such as video processing or machine learning model training, when the grid is strained due to weather events or other high-demand conditions.

“This builds on our successful demonstration with Omaha Public Power District (OPPD), where we reduced the power demand associated with ML workloads during three grid events last year,” said Terrell.

Google emphasized that this flexibility “helps reduce the need to build new transmission and power plants, and helps grid operators more effectively and efficiently manage power grids.”

Steve Baker, president and chief operating officer of Indiana Michigan Power, said in the same announcement, “As we add new large loads to our system, we must partner with our customers to effectively manage the generation and transmission resources necessary to serve them.” He added that Google’s ability to shift workloads “will be a highly valuable tool” in addressing growing energy needs.

Not every task can be paused

While this power-saving strategy is innovative, it does come with limits.

Some critical workloads, such as Google Search, Maps, and services used by healthcare or cloud customers, can’t afford to be interrupted; as such, Google acknowledged that demand flexibility will only be implemented at certain locations, and only where service reliability won’t be compromised.

Still, the company believes it can scale the model. Google said it already uses demand shifting for non-AI tasks, like processing YouTube videos, and moves these to data centers in locations with more power availability.

Google emphasized this is just one part of its broader energy strategy; the company is also investing in carbon-free energy sources to meet its 24/7 clean energy goals. In May, Google inked a deal with Elementl Power to support the development of small modular nuclear reactors in the US.

As tech giants race to secure more power, learn how researchers cut data center energy use by 30% with just 30 lines of code.