NVIDIA and Emerald AI Want to Turn Data Centers Into Grid Assets — With 100 GW of Potential

NVIDIA and Emerald AI dropped one of the more consequential announcements at CERAWeek 2026 this week: a framework for turning AI data centers into flexible grid assets. Joined by AES, Constellation, Invenergy, NextEra Energy, Nscale Energy & Power, and Vistra, the initiative aims to solve one of the biggest tensions in the AI buildout — the conflict between insatiable compute demand and a power grid that wasn't designed for it.

The core technology is NVIDIA's Vera Rubin DSX AI Factory reference design, paired with a new software layer called DSX Flex. The idea is that AI factories — massive GPU clusters running inference and training workloads — can dynamically modulate their power consumption in response to grid signals. When the grid is stressed, AI workloads throttle back. When surplus power is available, they ramp up. The AI factory becomes a dispatchable load, not just a consumer.

The Numbers Are Staggering

A demonstration in Hillsboro, Oregon — run in collaboration with Portland General Electric and EPRI — showed that AI factories can respond precisely to utility signals while still maintaining priority workload performance. NVIDIA and Emerald AI claim that at scale, power-flexible AI factories could unlock up to 100 gigawatts of grid capacity on the existing U.S. power system. To put that in context, 100 GW is roughly the output of 100 nuclear reactors.

Emerald AI's Conductor platform sits at the center of the orchestration layer, coordinating compute workloads with onsite energy resources — batteries, generation assets, demand response programs. The goal is to maintain AI service quality while providing the kind of grid flexibility that utilities desperately need as electrification accelerates.

Why This Changes the Conversation

The narrative around AI and energy has been almost entirely negative: data centers are devouring power, straining grids, and threatening climate targets. U.S. data centers now consume approximately 176 terawatt-hours annually — 4.4 percent of the nation's total electricity — and that number is climbing fast. Goldman Sachs projects a 175 percent increase in data center power consumption by 2030.

This initiative flips the script. Instead of AI factories being a grid liability, they become a grid asset — one that can provide flexibility services, reduce congestion, and potentially defer billions in transmission upgrades. DSX Flex is expected to deploy at commercial scale later this year at NVIDIA's AI Factory Research Center in Virginia, which will be one of the world's first power-flexible AI factories running Vera Rubin infrastructure.

The strategic alignment between the compute industry and the energy industry has never been this explicit. Six of the largest power companies in the U.S. are at the table. NVIDIA is providing the hardware and software stack. And Emerald AI is building the coordination layer. If this works at scale, it doesn't just solve an AI problem — it creates a new category of grid infrastructure. That's worth watching closely.