The potential growth of both the electric and information technology sectors due to artificial intelligence (AI) could enhance people’s daily lives and drive economic growth. However, realizing these benefits requires significantly more electricity and a potential shift in how data centers are built and operated. To address rising energy demand and maintain reliable power, stakeholders are exploring a range of flexibility solutions.
For example, building new infrastructure, such as transmission lines, to add capacity to the system can take up to 10 years. But deploying data center operational flexibility may help unlock capacity in a much shorter timeframe. This highlights the timeliness of DCFlex, EPRI’s three-year flexible load initiative to balance the needs of power systems with the energy requirements of large loads, such as data centers.
Since its launch last year, DCFlex has grown from 14 to 45 members. The global initiative brings together hyperscalers – like Google, Microsoft, Meta – with data center developers, technology companies, utilities, grid operators, and equipment suppliers to identify innovative strategies to enable AI growth while minimizing costs and enhancing electric system reliability.
Today the collaborative is announcing the first three real-world demonstration sites to explore the opportunity and feasibility for load flexibility:
Field demonstrations are a unique element of DCFlex and are designed to accelerate the adoption of solutions. Each demonstration brings together critical parties across multiple industry sectors to conduct these living laboratories. This research will help determine data center and power supplier strategies to enable operational flexibility and streamline the integration of data centers with the electric system.
These demonstrations will test a variety of use cases, such as load flexibility, power quality, and backup energy solutions, among others. Sites one and two will be demonstrating how computational flexibility can address grid needs. Site three explores data center response to power quality events to help identify best practices for energy system reliability and resiliency. The criteria for determining the test sites included diverse geographies and market structures, data center size, ownership models, and on-site generation technologies, among others.
In a preliminary test of the depth of computational flexibility possible in an AI data center, the Arizona demonstration site experienced some early success. It showcased the potential for an AI data center to provide grid relief during a peak system event - such as a hot summer day with high power demand - by temporarily and precisely ramping down its electricity consumption without compromising data center performance.
Early results from some of the demonstration projects announced today are anticipated by this fall. Simultaneously, the selection process is underway for additional demonstration sites, which are expected to be announced in the same timeframe. Future demonstrations may expand to include how data centers’ on-site backup generators powered by low-carbon fuels -- such as biodiesel, renewable natural gas, hydrogen plus storage, or thermal energy and storage -- could support a more reliable grid.
All the field testing, in aggregate, could inform the ultimate framework for operationalizing flexibility and help establish how data centers can incorporate flexibility into design and operations to support AI-driven load growth. Results could also determine how data centers can be more quickly connected to the power grid by equipping the industry with the necessary standards, frameworks, processes, and tools to ease adoption barriers.