Google Partners With 2 Utilities On 'Demand Response,' A Key Solution To Data Center Power Crisis

Google has signed deals with a pair of U.S. utilities, agreeing to reduce power consumption at certain artificial intelligence data centers when high electricity demand is putting the regional power grid under strain.

The agreements represent an early test case for Big Tech’s adoption of “demand response,” an approach that many across the data center and utility sectors believe is a critical part of the solution to the emerging data center power crisis. 

On Monday, Google announced in a blog post that it has signed demand response agreements with Indiana Michigan Power and Tennessee Valley Authority, major utilities that together serve more than 10 million customers across nine states.

Rather than simply consuming hundreds of megawatts of electricity, some Google data centers will now become “flexible loads” capable of curbing their power consumption on command when utilities determine that excessive demand is pushing the grid toward failure. 

The hyperscale cloud provider will cut power when needed by curtailing machine learning workloads, a type of AI computing. Google has provided few details regarding the financial terms of the deal or the participating data centers, although Monday’s announcement specifically mentions the company’s $2B Project Zodiac development underway in Fort Wayne, Indiana.

Executives at Google and the utilities touted the demand response efforts as a critical tool that will expedite access to massive blocks of power for data center firms amid an AI-fueled building boom while helping mitigate increasingly dire reliability risks and rising costs for utilities. 

“It allows large electricity loads like data centers to be interconnected more quickly, helps reduce the need to build new transmission and power plants, and helps grid operators more effectively and efficiently manage power grids,” the company wrote in Monday's blog post. 

Demand response isn't a new concept. Many regional power systems have long offered some form of demand response program, providing financial incentives for large industrial power customers that agree to on-demand curtailment to help utilities avoid blackouts and other catastrophic events when energy demand is high but supply is low.

This is particularly important in markets where a high percentage of electricity comes from renewable sources during periods when solar and wind farms aren't producing energy. Demand response allows other grid users to access more renewables by giving grid operators demand-side levers to help them balance the more volatile energy supply.

For a data center or other large load energy user, participation in demand response often allows power to be acquired faster, as utilities can approve grid connections for blocks of power that would have created grid reliability problems if the load were not flexible.

Demand response programs have seen widespread adoption by manufacturers and other industrial users in certain markets, as well as operators of data centers used exclusively for mining cryptocurrencies. But other than a few scattered exceptions, mainstream data center developers and operators have largely stayed on the sidelines in markets where such programs are offered.

“It’s a proven model. Data centers have just been a little slow to adopt it,” Allan Schurr, chief commercial officer of distributed energy firm Enchanted Rock, told Bisnow last year.

Part of the problem is that many data centers are mission-critical facilities that need to remain fully operational at all times. Unlike a bitcoin mine, they can’t simply shut down and stop using power. 

Data center operators and tenants are inherently risk-averse, and participation in a program that would hand certain power management decisions to a third party and periodically remove a layer of redundancy for the facility’s energy supply would run afoul of most tenant contracts, experts have said. 

Yet as surging data center power demand gives rise to a looming grid reliability crisis, spikes prices and slows the supply of new data center capacity for hyperscalers, there have been growing calls for data centers to address these issues through demand response. 

“Data centers have a choice: They can either create the problem or they can solve their own problem and help the grid,” Schurr said.

There have been gradual steps toward greater adoption of demand response throughout the data center ecosystem. 

In October, energy think tank Electric Power Research Institute and a group of the largest data center developers, tech giants and utilities — a consortium that includes Google — launched DCFlex, an initiative to develop flexible-load data center test projects in different energy markets across the country. The program aims to develop models and proofs of concept for the technologies, policies and operating practices needed across these sectors to make such grid-responsive data centers viable at a large scale

Monday’s announcement was also not the first time Google has engaged in its own demand response initiatives, publishing blog posts outlining limited efforts as early as 2023. 

“The first data center demand response capabilities we developed involve shifting non-urgent compute tasks — like processing a YouTube video — during specific periods when the grid is strained,” the company wrote in Monday's blog post, pointing to specific initiatives with utilities in Belgium and Taiwan.  “[W]e've leveraged this capability to help grid operators maintain reliability during those periods of the year when demand is the highest.”

Still, Google has indicated its focus on demand response through curtailing machine learning workloads will be much more expansive than these earlier efforts. The firm framed its latest efforts as a key element of its energy strategy.

Google expects to reach similar agreements with utilities across the company’s expanding data center portfolio. 

Machine learning workloads consume massive amounts of power, can be easily segregated and aren't mission-critical in the same way as traditional cloud workloads, making these deployments well-suited for demand response. According to Google, the company’s new initiative follows a successful demonstration project with a Nebraska utility through which Google successfully reduced power demand from machine learning workloads during three grid events last year.

“As AI adoption accelerates, we see a significant opportunity to expand our demand response toolkit, develop capabilities specifically for ML workloads, and leverage them to manage large new energy loads,” the company said in a statement. “By including load flexibility in our overall energy plan, we can manage AI-driven growth even where power generation and transmission are constrained.”