Google is taking steps to make its data center fleet more flexible. Today, the company announced two new demand response agreements with the utilities Indiana Michigan Power and Tennessee Valley Authority. During times of grid stress, certain Google data centers will reduce the power use associated with machine learning workloads.
Michael Terrell, Google’s head of advanced energy, noted in a blog post accompanying the announcement that this is the first time that Google is “delivering data center demand response by targeting machine learning workloads.” This follows a successful demonstration with the Omaha Public Power District, which reduced power demand during three emergency grid events in 2024.
Google has been using demand response tools to enhance the flexibility of its data centers since 2023, when it announced it had developed a way to reduce a data center’s energy use at times of grid stress by shifting some of its compute tasks to other times and locations. Until recently, however, the tool only targeted non-urgent tasks, such as the processing of YouTube videos.
“Incorporating ML workloads is an important step to enable larger-scale demand flexibility,” Terrell wrote.
And that demand flexibility is increasingly important as load growth fundamentally reshapes the grid. The artificial intelligence boom is inflating the data center sector’s energy demand, flooding utilities with load requests and straining an already congested grid, so much so that in some areas, new data centers have to wait nearly a decade for power.
To alleviate the problem, industry players have been experimenting with flexibility solutions, which could unlock over 100 gigawatts of load in the existing U.S. power system, according to recent studies. For instance, Google is a member of DCFlex, a coalition led by the Electric Power Research Institute bringing together hyperscalers, utilities, RTOs and ISOs to explore ways to turn data centers into flexible grid assets, rather than just drains on power resources.
Google’s load-shifting tool is only one potential flexibility solution; others include batchable workloads and load curtailment supported by on-site generation. Last month, DCFlex shared its first set of results from a flexibility “hub” in Phoenix, Arizona, at an Oracle data center. The pilot project used the startup Emerald AI’s platform to reduce power use by 25% at peak demand, by choreographing clusters of Nvidia GPUs in real time.
One of the key benefits of these flexibility opinions is that they can reduce the need for new infrastructure and can be deployed quickly, according to Terrell, who also said that they can help “bridge the gap between short-term load growth and long-term clean energy solutions.”
‘Holistic approach’
These solutions, however, are far from being deployed widely. As Terrell put it, “data center demand flexibility is still in the early stages and will only be available at certain locations,” given that critical services such as Google Search, Maps, and Cloud for essential industries like healthcare need “high levels of reliability.”
The Indiana Michigan Power project will be at a new data center in Fort Wayne, Indiana; president and COO Steve Baker said in a statement that “Google’s ability to leverage load flexibility as part of the strategy to serve their load will be a highly valuable tool to meet their future energy needs.”
And grid flexibility has the potential to benefit Google as well, as the company balances its ambitious net-zero commitments with the emissions that are already increasing as a result of the AI boom. The company said in its 2024 environmental report — its first since the AI race really ramped up — that its emissions increased by nearly 50% in just five years, and its data centers devoured 17% more electricity in 2023 than in 2022. Terrell described a “holistic approach” to achieving net zero by 2030, which also includes procuring a range of clean firm energy.
In late June, Google announced it had made a second capital investment in nuclear fusion company Commonwealth Fusion Systems, while entering into a direct corporate offtake agreement for 200 megawatts of energy from their first commercial plant. In mid-July, it announced what’s believed to be the world’s biggest hydropower deal: a 20-year power purchase agreement in which it agreed to pay Brookfield Asset Management $3 billion for up to 3 gigawatts of electricity from two hydropower plants in Pennsylvania. And a week later, it entered into its first commercial long-duration storage deal with Energy Dome, which develops a carbon dioxide-based battery.
