Google has agreed to reduce electricity consumption at its artificial intelligence data centres during peak demand periods in the US, in a move aimed at easing strain on the country’s power grid as energy-intensive AI workloads expand.

The company has signed demand-response agreements with Indiana Michigan Power and the Tennessee Valley Authority, marking its first formal participation in such programmes, which have traditionally involved heavy industry and cryptocurrency mining.

Under the arrangements, Google will temporarily scale back machine learning workloads when requested by the utilities. The company said this flexibility could help prevent blackouts, defer costly investments in new power infrastructure, and support faster grid connections in regions where capacity is constrained.

In a blog post, Michael Terrell, Google’s head of advanced energy, said the new approach builds on earlier demand-response efforts that shifted lower-priority tasks such as video processing. Targeting AI workloads, he said, represents a “step change” in how large-scale computing can interact with power systems.

The agreements come as utilities face mounting pressure from the rapid growth of AI and data centre capacity. US data centres accounted for about 4.4 per cent of national electricity consumption in 2023, with analysts projecting that share could triple by 2028.

Google has also committed to a $3bn hydropower deal and $25bn in planned investments in the PJM Interconnection region, the largest US electric grid, to secure clean energy supplies for its operations.