Skip to content
Advertisement
Google leans on demand response to keep AI data centers running at peak hours

Google leans on demand response to keep AI data centers running at peak hours

Google / Wikimedia Commons (Public domain)

The company has lined up 1 GW of flexible capacity with five US utilities and is scaling a virtual power plant deal with Voltus tenfold

Google wants its AI data centers to know when to take a breather.

The company is exploring demand-response programs to cover the power needs of its AI data centers during peak hours. Under these arrangements, Google’s facilities can dial down or reschedule work when the grid gets stretched thin.

AI workloads are driving electricity demand sharply higher. Utilities are being asked to connect very large new loads faster than they can build the generation and transmission to serve them. Demand response offers a way to squeeze more use out of the grid that already exists.

What Google has actually signed

For Google, that means shifting or curtailing certain machine-learning workloads when electricity is scarce.

By March 19, 2026, Google had integrated a total of 1 gigawatt (GW) of demand-response capacity. That capacity comes through long-term contracts with five US utilities:

Advertisement

Entergy Arkansas, Minnesota Power, DTE Energy, Indiana Michigan Power, and the Tennessee Valley Authority (TVA).

Then came the bigger move. On October 22, 2026, Google announced a tenfold expansion of its partnership with Voltus. The agreement grew from an initial 100 megawatts (MW) to approximately 1 GW of aggregated distributed capacity aimed at the PJM Interconnection region.

That Voltus contract runs through 2040 and will add 1,070 MW of virtual power plant capacity.

Virtual power plants, explained

A virtual power plant does not have smokestacks or turbines. It is a network of many small energy resources, managed together so they behave like one large one.

In Google’s programs, those resources come from residential, commercial, and industrial participants. Technologies such as batteries and smart thermostats let them trim their consumption on cue.

Google’s stated goals here stack up neatly. The programs are meant to improve grid reliability and help new large loads connect more quickly. They also aim to lower dependence on building fresh generation or transmission infrastructure.

How Google got here

Google’s first demonstrations of demand-response capabilities started in 2024 with the Omaha Public Power District.

Agreements with additional utilities followed in 2025, including Indiana Michigan Power and TVA. By early 2026, the patchwork had grown into the 1 GW portfolio spread across five utilities. The Voltus expansion later that year pushed the strategy into PJM territory with a much larger footprint.

Not everyone is applauding, though. The Michigan Attorney General has filed a petition challenging aspects of Google’s data center contracts with DTE Energy. The filing points to ongoing regulatory scrutiny of these arrangements and of what they could mean for costs borne by ordinary consumers.

What this means for utilities, regulators, and ratepayers

The Michigan Attorney General’s petition is a reminder that utility contracts with big tech firms are not purely private deals. When a utility signs special terms with a data center operator, regulators and consumer advocates want to know who ultimately pays.

What to watch next: how the Michigan challenge to the DTE Energy contracts plays out, whether the Voltus capacity in PJM comes online as planned, and whether Google adds more utilities to its roster.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

More in Technology

See all