Google is taking steps to make its data center fleet more flexible. Today, the company announced two new demand response agreements with the Indiana Michigan Department of Electricity and Tennessee Valley. At the time of grid stress, certain Google data centers reduce the power usage associated with machine learning workloads.
Michael Terrell, head of Advanced Energy at Google, pointed out in a blog post following the announcement that this is the first time Google has provided a data center demand response by targeting machine learning workloads, following a successful demonstration in the Omaha Public Power District, which reduced power demand at three emergency grid events in 2024.
Since 2023, Google has been using demand response tools to increase data center flexibility and has announced that it has developed a method to reduce data center energy use during grid stress by shifting some of its computing tasks to other times and locations. Until recently, however, the tool was targeted at urgent tasks, such as processing YouTube videos.
“Incorporating ML workloads is an important step that allows for flexibility in demand at scale,” writes Terrell.
And as load growth fundamentally reshaping the grid, the flexibility of its demand is becoming increasingly important. In some areas, new data centers will have to wait nearly a decade as the artificial intelligence boom has inflated energy demand in the data center sector, flooding utilities with load requests, and straining grids.
To mitigate the problem, recent research shows that industry players are experimenting with flexibility solutions, allowing them to unlock loads of over 100 gigawatts of existing US power systems. For example, Google is a member of DCFLEX and a coalition led by the Power Institute, which will bring together hyperscalers, utilities, RTOs and ISOs to explore how to turn data centers into flexible grid assets rather than exhausting electricity resources.
Google's load shifting tool is just one of the potential flexibility solutions. Others include batchable workloads and load reductions supported by on-site generations. Last month, DCFLEX shared its first set of results from its Phoenix, Arizona flexibility “hub” at its Oracle Data Center. The pilot project reduced power usage by 25% in peak demand by choreographing clusters of NVIDIA GPUs in real time using the startup Emerald AI platform.
One important advantage of these flexibility opinions is that they can reduce the need for new infrastructure and can be deployed quickly. It also states that it can “fill the gap between short-term load growth and long-term clean energy solutions.”
“Holistic Approach”
However, these solutions are far from being widely deployed. As Terrell said, “The demand for data centers is still in its early stages and will only be available in certain locations,” and critical services such as Google Search, Maps and the cloud require a “high level of reliability” for key industries such as healthcare.
The Indiana Michigan Power Project is located in a new data center in Fort Wayne, Indiana. “The ability to leverage load flexibility as part of Google's strategy to provide loads will be a very valuable tool to meet future energy needs,” President and COO Steve Baker said in a statement.
The flexibility of the grid could also benefit Google as it balances the already growing emissions as a result of the AI boom with ambitious net-zero commitments. In its 2024 Environmental Report, the company said that for the first time since AI races actually increased – emissions have increased by nearly 50% in just five years, with 17% more electricity than 2023 being greedy.
In late June, Google announced it had made its second capital investment in fusion company Commonwealth Fusion Systems, and signed a direct offtake contract for 200 megawatts of energy from its first commercial plant. In mid-July, we announced what is considered to be the world's largest hydroelectric transaction. This is a 20-year power purchase agreement that has agreed to pay $3 billion to $3 billion for up to 3 gigawatts of electricity from two Pennsylvania hydroelectric plants. A week later, we entered our first commercial long-term storage agreement with Energy Dome, which develops carbon-based batteries.
