A coalition of leaders from the artificial intelligence and energy sectors -- including NVIDIA, Emerald AI, EPRI, Digital Realty, and the PJM Interconnection –- recently announced the world's first power-flexible AI Factory.
The facility, the 96-MW Aurora AI Factory in Virginia, will serve as the implementation of a new reference design and certification standard for power-flexible AI infrastructure, enabling utilities to unlock existing grid capacity and rapidly connect new AI data centers, Emerald AI said.
The new power-flexible reference design, if adopted nationwide, could unlock an estimated 100 GW of capacity on the existing electricity system to power AI's growth, equivalent to 20% of total U.S electricity consumption in a year, the company said. “Flexible AI Factories could more swiftly connect to power grids, accelerating the buildout of AI infrastructure and accelerating AI innovation while advancing power grid reliability and affordability.”
The Aurora AI Factory, under construction by Digital Realty in Manassas, Va., and slated to open in the first half of 2026, will be the first facility built to this new standard and will serve as a live innovation hub.
"It aims to demonstrate how AI compute can align with grid needs to improve the utilization of existing power networks, relieving grid stress during peak demand, and ultimately contribute to a more stable and affordable energy future," Emerald AI said.
As the demand for generative AI drives unprecedented growth in compute, Emerald AI is working with NVIDIA's development of AI factory infrastructure, that not only advances AI, but also actively participates in stabilizing the grid. At the heart of this architecture is close integration between technology from Emerald AI, an NVIDIA NVentures portfolio company and NVIDIA Inception member, and the NVIDIA AI software stack, Emerald AI said.
Emerald AI’s GridLink and Conductor products leverage NVIDIA AI Enterprise components, including NVIDIA NIM microservices, in coordination with the NVIDIA Mission Control, "providing a crucial link between the grid's real-time needs and the data center's operational controls, allowing for intelligent workload scheduling and power management to meet both grid needs and stringent AI workload performance standards."
To validate the architecture's effectiveness, the Aurora facility will undergo demonstration testing in collaboration with the EPRI DCFlex Initiative. DCFlex explores how data centers can provide grid services, benefiting utilities, operators, and consumers alike.
The effort will test precise, real-time responses to simulated grid stress events. Scenarios will include mimicking demand spikes during summer heatwaves or sudden drops in renewable energy generation. The overall aim of the demonstration is to see whether the data center can intelligently ramp power consumption down to support the grid while using sophisticated workload orchestration to assure acceptable Quality of Service and performance for mission-critical AI training and inference workloads.
NVIDIA and Emerald AI previously conducted a demonstration of AI data center flexibility through EPRI’s DCFlex Initiative in Phoenix, Ariz., along with Oracle and Salt River Project, an Arizona public power utility.
In addition to U.S. trials, Emerald AI will conduct an upcoming demonstration in the United Kingdom with its strategic partners and investors, National Grid and NVIDIA, which was announced as a part of the U.S.-U.K. Tech Partnership in London last month.
