Flexible Loads, Resilient Grids

EPRI’s DCFlex initiative explores how large flexible loads can support grid reliability and AI-driven growth

By Chris Warren

Over his three-decade-plus career in the electric power industry, David Porter has seen just about everything. Before joining EPRI in 2009, Porter held a variety of positions in utility operations—from power marketing to transmission, distribution, energy efficiency, and customer programs for companies like Florida Power Corporation and Duke Power.

So, when an industry development catches Porter’s attention, it’s worth focusing on. These days, Porter is struck by the connection between grid reliability and the operation of large loads, especially data centers. “As someone who’s been in and around that business for a long time, grid operations, reliability, and resilience have not been so tied to large point loads as it is now,” said Porter, who is EPRI’s Vice President for Electrification & Sustainable Energy Strategy.

The link between grid reliability and resilience and data centers can only be expected to strengthen. A study released last year by EPRI found that data centers could consume as much as nine percent of U.S. electricity generation by 2030, more than double the amount currently used. EPRI isn’t alone in forecasting rapid data center load growth. According to an analysis conducted by the U.S. Department of Energy’s (DOE) Lawrence Berkeley National Laboratory (LBNL), data center electricity consumption tripled between 2014 and 2024 and is expected to double or triple again by 2028.

A Mismatch in Speed

The implications of data center load growth to support artificial intelligence (AI) applications and cloud computing are profound—both for the utility industry and society. “Every hour of every day, we really are interacting with the internet and with digital ecosystems in a way that relies on data center services. I like to think of data centers as the backbone of our modern economy,” said Briana Kobor, Head of Energy Market Innovation at Google.

For utilities, data center load growth poses opportunities and challenges. For one thing, meeting demand from data centers, manufacturing, electric transportation, and a growing population requires significant investment in the grid. The consulting firm Deloitte estimates utilities must spend about $1.4 trillion between 2025 and 2030—much of it on generation to serve new loads—and maintain similar levels of investment in the following decades.

While utilities are committed to serving new customers and loads, a significant challenge has emerged. “There’s a timing mismatch between how quickly tech companies can build a new data center facility – at a scale much larger than what the utilities are used to dealing with historically – and the utilities’ ability to provide that level of service in a similar time frame,” Porter said. “There’s a severe mismatch there today.”

Bridging the Gap

It’s a mismatch that must be closed for companies like Google to develop new products and serve customers. “The scale and speed with which the tech industry operates is on a different time spectrum than the electricity ecosystem,” Kobor said. “What we’re faced with is a growing pain between where our tech companies need to be to serve our customers and to serve this demand, and what the electric industry is capable of doing.”

But it’s also a mismatch that is driving innovation and expanded collaboration between utilities and data center developers and owners keen to access the electricity they need as quickly as possible. One promising area of innovation is data center demand flexibility. In the past, Porter says, large loads could be curtailed or interrupted when the grid was stressed. But data centers and the digital services they provide can’t just be turned off.

Demand flexibility, by contrast, offers an opportunity for data centers to connect to the grid faster and operate in ways that simultaneously serve their customers and the grid. “Having the ability to have large, flexible loads to improve the overall operation of the grid is key,” Porter said. “We believe that data centers can be that large, flexible point load. Not all the time, not for an entire year, but in times when the grid really needs some assistance to meet the loads. That’s when it’s really important.”

Launching DCFlex

EPRI launched the DCFlex initiative in 2024, a three-year effort to advance the understanding of data center demand flexibility and to begin testing how it can be improved in real-world conditions. DCFlex brings together hyperscalers like Google, Meta, Oracle, and Microsoft with technology companies like the chip maker NVIDIA, utilities, grid operators, data center developers, and equipment suppliers.

The initiative, which now has 49 participants, seeks innovative and collaborative strategies to empower AI growth while minimizing costs and enhancing grid reliability. While many of the companies participating in DCFlex are technology pioneers, the ability to transform data centers into dynamic AI and grid-supporting resources depends on both technological innovation and a shift in mindset around operations, business models, and regulations.

To pursue these goals, DCFlex is organized into four workstreams.

Real-World Living Laboratories

DCFlex will include five to 10 large-scale demonstration sites to serve as living laboratories where innovative tools to improve operational flexibility, grid integration, and backup power can be tested.

The first three demonstrations have already been announced. Two of the projects—one in Lenoir, North Carolina, and another in Phoenix, Arizona—will explore how computational flexibility can help support the grid. The third project is in Paris, France, and will examine data center response to power quality issues to develop best practices that will support grid reliability and resilience.

These demonstrations—which are taking place in production facilities, not laboratories—will provide instructive lessons about data center flexibility. Google and Duke Energy will work together at the demonstration in Lenoir, North Carolina. Kobor says collaboration has already begun to develop operational and communication protocols that will give Duke Energy confidence that it can rely on data center flexibility as a grid resource.

“Can we reduce the load and, most importantly, can Google and others continue to get their products out the door in the way they need to when we enact those load reductions?” Porter said. “The second piece looks at the ability to flex those loads for varying periods, depending on what the utility programs in place might look like.”

Similar to the Google site, the demonstration project in Phoenix, Arizona, will probe workload flexibility. In a preliminary field test at the Arizona site, Emerald AI successfully mapped which AI tasks can be slowed, paused, or relocated, allowing data centers to reduce power consumption during times of grid stress. This trial showed that an AI compute cluster could reduce its power consumption by 25% over a three-hour peak system event without compromising essential data center services.

The demonstration projects will also evaluate the ability of data centers to provide ancillary services to the grid and the possibility of using renewable fuels, like hydro-treated vegetable oil (HVO), instead of diesel, to run the backup generators at data center sites. Ultimately, the lessons from all the field tests will inform the framework EPRI is developing to guide the integration of flexibility into data center design and operation. The field testing could also speed data center grid connections by identifying and sharing the standards, frameworks, processes, and tools needed to overcome interconnection hurdles.

For his part, Porter is excited that EPRI is working with such a broad coalition of partners to advance AI in ways that benefit the grid and society. “The AI revolution is really our next industrial revolution, and it’s really changing our daily lives. When you couple that with the clean energy transition that’s underway in the utility industry, it’s an intersection of great opportunity and challenge,” Porter said. “I feel very fortunate that EPRI has a seat at the table with companies like Google, their peers, and others who are trying to drive change and build on the opportunities that are in front of us.”

---