Emerald AI is joining EPRI’s DCFlex Initiative, deepening a collaboration that has been central to our work since our founding in November 2024. DCFlex launched that same fall, and our work together has flourished. We are proud to have collaborated on six DCFlex demonstrations with NVIDIA, utilities, grid operators, and data center partners.

Access to electricity has become a critical constraint on AI infrastructure growth. Emerald AI’s approach is to enable data centers to adjust their power consumption during limited periods of grid stress while protecting the performance of critical computing workloads. Applied broadly, this flexibility could help unlock up to 100 GW of capacity on the existing U.S. power system, supporting faster, larger connections for AI infrastructure while helping maintain reliable, affordable electricity for the communities that host it.

Our work with EPRI has brought together DCFlex members and demonstration partners including Salt River Project, Oracle, ComEd, PJM Interconnection, Constellation, Dominion Energy, Portland General Electric, National Grid, and Nebius. NVIDIA has collaborated in every demonstration, with Emerald Conductor coordinating AI workloads in response to grid needs.

“We welcome Emerald AI as a full member of DCFlex,” said Arshad Mansoor, EPRI’s President & CEO. “We’ve made great strides testing AI power flexibility in operating data centers. The team at Emerald AI has been an important collaborator in many of these field demonstrations, designed to inform a framework for operationalizing energy-use flexibility in data centers, supporting both grid stability and AI-driven growth.”

Our DCFlex demonstrations have proven an increasingly broad range of capabilities. Results from our Arizona demonstration, published in Nature Energy, showed that an AI computing cluster could reduce its power draw by 25% for three hours during peak demand while meeting all workload service commitments. Subsequent demonstrations have extended that work: in London, Emerald Conductor met more than 200 power targets, with reductions reaching 40%; between Virginia and Chicago, we shifted inference workloads to relieve a local power constraint while maintaining low latency; and in Oregon, we integrated with NVIDIA’s full DSX stack to respond to utility requests under scenarios ranging from freezing weather to a prolonged heat wave. Together, these projects have tested flexibility across different grids, computing systems, and operating conditions.

“EPRI has been a close partner since Emerald AI’s earliest days, helping us demonstrate our technology alongside the utilities and grid operators that need to be confident in its performance,” said Dr. Varun Sivaram, founder and CEO of Emerald AI. “We are working toward a model in which AI data centers can secure power sooner by providing dependable flexibility to the grid. Our ambition is to enable AI growth in a way that strengthens the power system and protects electricity affordability for the communities that host it.”

Emerald AI and NVIDIA are collaborating to integrate power flexibility into the DSX reference architecture for AI Factories, including at commercial scale at NVIDIA’s AI Factory Research Center. Incorporating flexibility into an AI factory’s design creates an opportunity to evaluate the full relationship between computing performance, facility operations, and grid response. The findings can help inform how future facilities are designed and how their flexibility is assessed during interconnection and operation.

Joining DCFlex enables Emerald AI and EPRI to collaborate on independent research that advances the state of the art for AI flexibility. Utilities need a clear understanding of the response a facility can provide, including its magnitude, speed, duration, and availability under specific grid conditions. EPRI’s demonstrations and shared technical frameworks help establish a consistent basis for evaluating those capabilities. For Emerald AI, participating more deeply in that work is an important part of making flexibility a dependable resource that utilities can incorporate into planning and operating decisions.

Our participation in DCFlex also complements our work with Google and NVIDIA as founding members of the AI Energy Management Alliance (AEMA). AEMA advocates for policies that recognize verified flexibility and support faster interconnection for data centers that contribute to grid reliability and energy affordability. The technical evidence developed through DCFlex and the policy work undertaken through AEMA address different requirements for wider adoption, both of which are important to Emerald’s mission of making AI data centers good grid citizens.

Joining DCFlex reflects our commitment to continue this work with EPRI and the partners who have helped develop and test it. As we move into larger deployments, our focus remains on enabling AI infrastructure to grow in ways that utilities can confidently accommodate and that benefit the communities served by the same power system.