Emerald AI is joining EPRI’s DCFlex Initiative as a full member, taking a larger role in efforts to make AI data centers more flexible users of electricity. The company announced the move on November 2, 2026, after six demonstrations involving NVIDIA, utilities, grid operators and data-center partners.
The goal is simple to describe, even if the engineering is not: data centers should be able to adjust their electricity use during limited periods of grid stress without disrupting critical computing work. Emerald AI says that approach could unlock up to 100 GW of capacity on the existing U.S. power system, helping new AI infrastructure connect sooner while preserving reliable and affordable electricity.
Emerald AI’s relationship with EPRI dates to the company’s founding in November 2024, the same fall that DCFlex launched. The demonstrations have included Salt River Project, Oracle, ComEd, PJM Interconnection, Constellation, Dominion Energy, Portland General Electric, National Grid and Nebius. NVIDIA participated in each demonstration.
What the demonstrations showed
Emerald Conductor, the company’s system for managing AI workloads, coordinated computing activity in response to grid needs. The results offered several different examples of how that flexibility might work in practice, from reducing power use at one site to moving workloads between regions.
In Arizona, an AI computing cluster reduced its power draw by 25% for three hours during peak demand while meeting all workload service commitments, according to Emerald AI. Nature Energy published the results. In a later London demonstration, Emerald Conductor met more than 200 power targets, with reductions reaching 40%.
Another test shifted inference workloads between Virginia and Chicago to relieve a local power constraint while maintaining low latency. In Oregon, Emerald AI integrated with NVIDIA’s full DSX stack and responded to utility requests during scenarios that included freezing weather and a prolonged heat wave.
These tests address a question utilities need answered before they can count on a data center as a grid resource: what response can the facility provide, how large will it be, how fast can it arrive, how long can it last and how often will it be available under specific conditions?
Turning flexibility into a planning tool
EPRI President and CEO Arshad Mansoor welcomed Emerald AI as a full member. He said the field demonstrations were intended to inform a framework for putting data-center energy-use flexibility into operation while supporting grid stability and AI-driven growth.
Through full DCFlex membership, Emerald AI will collaborate with EPRI on independent research. EPRI’s demonstrations and shared technical frameworks provide a consistent basis for evaluating the capabilities of flexible data centers, while deeper participation could help utilities include those capabilities in planning and operating decisions.
Founder and CEO Varun Sivaram said Emerald AI is working toward a model in which AI data centers can secure power sooner by providing dependable flexibility to the grid. He described the company’s ambition as supporting AI growth while strengthening the power system and protecting electricity affordability in the communities hosting that infrastructure.
Emerald AI and NVIDIA are also working to integrate power flexibility into the DSX reference architecture for AI Factories. That effort connects the grid question to the physical buildout now taking place across the United States.
The pressure behind the new approach
AI infrastructure is moving toward a manufacturing mode, with standardized systems and faster construction timelines. Giga Energy’s GigaBase system, for example, combines standardized infrastructure, parallel site development and off-site manufacturing. The company says it can take projects from shovel-ready to rack-ready in nine months.
A Brookings Institution paper modeled the physical investment tied to AI infrastructure under a central scenario. Between 2025 and 2032, investment in data centers, specialized computing equipment, power infrastructure and related systems could total about $10.3 trillion, averaging 3.63% of U.S. GDP per year.
The paper estimates that level of investment would exceed major U.S. railroad, highway and electrification buildouts. It also estimates that the projected buildout would double the electricity consumption of the U.S. residential sector.
That scale creates strain as well as opportunity. Critics say the surge in connections is testing America’s aging grid infrastructure, with concerns that the short-term energy crunch could drive up electrical bills and that longer-term pressure could lead to outages or other technical problems.
At the same time, the new demand is pushing grid innovation. WeaveGrid founder and CEO Apoorv Bhargava described a power system with new variables, including renewable energy, electric vehicles, data centers and heat pumps, and said utilities now need to manage and coordinate those resources.
Veir CEO Tim Heidel said the scale of AI infrastructure investment is driving new technologies toward commercial use. Veir sells superconductive power lines that promise more efficient energy transfer. The company has said its cables can carry power five times farther than traditional cables while requiring 20 times less space, and it raised $110 million in a Series C funding round earlier this month.
Kristin Eberhard, vice president of policy at Rewiring America, described the current period as an “improv era,” where grid planners need to combine new transmission lines and centralized power plants with resources that had previously gone unused.
A wider push for practical grid innovation
The 2026 Fortnightly Top Innovators Awards reflected the same focus on practical solutions. ComEd, National Grid, Southern Company, TS Conductor and Veolia were named winners for projects addressing grid modernization, resilience, AI and cybersecurity, resource stewardship and load growth.
For the first time in the award program’s history, ceremony attendees selected one of the five category winners as Top Innovator of the Year through a live vote. National Grid won the honor with Viridi Parente and Good Carbon Co.
The award results and Emerald AI’s DCFlex work point to the same challenge: AI data centers need power, but their demand does not have to remain completely rigid. If operators can provide utilities with measurable and dependable responses, new computing capacity may become part of the answer to the grid’s constraints rather than only another source of pressure.
Based on
- Emerald AI Becomes Full Member of EPRI’s DCFlex Initiative — unite.ai
- The Center Will Not Hold | The Verge — theverge.com
- 2026 Fortnightly Top Innovators Awards Recognize Breakthroughs in Grid Modernization, Resilience, and Load Growth Solutions | Markets Insider — markets.businessinsider.com
- AI data-center development is moving toward a manufacturing mode | TechCrunch — techcrunch.com
- The data center boom has an upside: innovation for the US grid | TechCrunch — techcrunch.com
