Emerald AI Joins EPRI’s DCFlex for Grid-Flexible Data Centres

Emerald AI has joined the Electric Power Research Institute's (EPRI) DCFlex Initiative as a full member, expanding its collaboration to develop AI data centres capable of adjusting electricity consumption in response to grid conditions. Announced on 2 November 2026, the partnership builds on six field demonstrations conducted with NVIDIA, utilities, grid operators and cloud computing providers. Previous trials demonstrated power reductions of 25% over three hours in Arizona and up to 40% in London, while maintaining specified computing performance requirements. Emerald AI estimates that widespread adoption of flexible AI infrastructure could potentially unlock up to 100 gigawatts (GW) of capacity within the existing US electricity system.

 

Six Demonstrations Validate Flexible AI Computing

 

Emerald AI and EPRI have worked together since 2024 to evaluate how data centres can modify their electricity demand during periods of grid congestion without disrupting critical computing workloads. Their demonstrations have involved organisations including NVIDIA, Oracle, Salt River Project, PJM Interconnection, ComEd, Constellation, Dominion Energy, Portland General Electric, National Grid and Nebius. Emerald AI's software platform, Emerald Conductor, coordinates AI workloads according to electricity system requirements, allowing computing facilities to respond to utility requests while maintaining service commitments.

A demonstration in Phoenix, Arizona, published in the journal Nature Energy, showed that an AI computing cluster could reduce electricity consumption by 25% for three hours during peak demand without breaching workload service commitments. Subsequent testing in London recorded reductions of up to 40% while meeting more than 200 power targets. Trials in Virginia and Chicago demonstrated transferring AI inference workloads between locations to relieve local electricity constraints while maintaining low latency. In Oregon, the technology was integrated with NVIDIA's DSX platform to respond to utility requests under scenarios including freezing weather and prolonged heatwaves.

 

Read more: State Grid and CPFL Energia Advance Smart Grid Technology Cooperation in Brazil

 

Flexible Data Centres Could Address Grid Connection Constraints

 

Rapid growth in AI computing infrastructure has increased pressure on electricity networks, with grid connection capacity emerging as a significant constraint on new data centre development. Emerald AI's approach focuses on making electricity demand more responsive rather than requiring computing facilities to operate continuously at maximum capacity. By temporarily adjusting workloads during periods of grid stress, data centres could provide additional operational flexibility to utilities and potentially reduce the need for immediate network expansion.

The company estimates that broader deployment of this approach could unlock up to 100 GW of capacity on the existing US electricity system, although this remains a potential system-wide opportunity rather than demonstrated available capacity. Achieving such benefits would depend on the operational reliability of flexible computing, utility requirements and the ability of grid operators to incorporate verified demand reductions into their planning processes. Arshad Mansoor, President and CEO of EPRI, said the demonstrations were helping establish a framework for integrating data centre flexibility into electricity system operations while supporting grid stability and AI infrastructure development.

 

NVIDIA Collaboration Integrates Power Flexibility into AI Factories

 

Emerald AI is also working with NVIDIA to integrate flexible electricity consumption into the DSX reference architecture for AI Factories. The collaboration includes commercial-scale testing at NVIDIA's AI Factory Research Center, where computing performance, facility operations and electricity demand can be assessed together. Incorporating these capabilities into data centre design could allow developers to demonstrate potential grid support before facilities begin operating, potentially influencing connection assessments and infrastructure planning.

The expanded DCFlex collaboration will support independent research into the technical characteristics utilities need to evaluate flexible data centres. These include the magnitude of power reductions, response speed, duration and availability under different electricity system conditions. Establishing consistent assessment methods could help utilities determine whether AI computing facilities can provide dependable flexibility during periods of high demand or network congestion.

 

Explore OneStop ESG Marketplace: Grid, utilities & demand management

 

Industry Collaboration Targets Faster Grid Connections

 

Emerald AI's participation in DCFlex complements its work with Google and NVIDIA as founding members of the AI Energy Management Alliance (AEMA). While DCFlex focuses on technical testing and operational frameworks, AEMA advocates for policies that recognise verified electricity demand flexibility and support faster grid connections for qualifying data centres. The two initiatives address the technical and regulatory conditions needed to integrate rapidly expanding AI infrastructure into existing electricity networks.

Dr Varun Sivaram, Founder and CEO of Emerald AI, said, "We are working toward a model in which AI data centers can secure power sooner by providing dependable flexibility to the grid." He added that the company's objective was to support AI infrastructure growth while strengthening electricity system reliability and protecting affordability for surrounding communities.

 

Demonstrated Grid Flexibility Will Shape Future AI Infrastructure

 

As Emerald AI moves toward larger deployments, its collaboration with EPRI will focus on building evidence utilities and grid operators can use to assess flexible electricity demand. The demonstrations have shown that AI workloads can respond to changing power conditions across several computing environments, but wider adoption will require consistent performance standards and operational confidence.

The ability to verify and incorporate these capabilities into grid connection decisions could become increasingly important as AI electricity demand expands. Further commercial deployments will help determine whether flexible computing can deliver meaningful improvements in grid utilisation, connection timelines and electricity system reliability.

 

Source: Emerald AI