Nvidia and Emerald AI Jointly Announce Development of Configurable Power AI Factory

NVIDIA and Emerald AI jointly announce the development of configurable power AI factories. This collaboration aims to transform large-scale data centers into flexible power grids through innovative hardware and software integration, increasing grid connection speed and enhancing the reliability of power systems.

DSX Reference Architecture Optimizes Energy Allocation Efficiency

The next-generation AI factory will incorporate NVIDIA Vera Rubin’s DSX AI Factory reference design, integrated with the DSX Flex software library developed specifically for grid services. To overcome the traditional lengthy grid connection process, this architecture allows factories to initially deploy hybrid power sources using on-site generation and energy storage facilities, which can later be flexibly fed back into the grid to support the overall power system. This model not only accelerates AI computing capacity deployment but also creates multiple values for customers and local communities.

Additionally, the DSX reference architecture supports flexible configurations without auxiliary energy sources, enabling more efficient grid interconnection. NVIDIA CEO Jensen Huang stated that AI factories should be viewed as integrated engines driving progress, requiring the energy, computing, networking, and cooling systems to be designed collaboratively within the overall architecture.

Emerald AI Conductor Platform Provides Flexible Power Dispatch

Within this framework, Emerald AI’s Conductor platform plays a key coordinating role. It integrates computational flexibility with on-site generation, battery storage, and other behind-the-meter resources, ensuring high-quality service (QoS) for AI tenants while precisely responding to grid dispatch needs.

Emerald AI CEO Varun Sivaram said AI factories should not be seen as isolated power loads. Through the DSX Flex system, data centers can generate high-value data models and tokens while also providing relief during peak grid stress. This flexible dispatch mechanism effectively reduces the need for large-scale capacity expansion during peak demand, easing overall infrastructure costs.

Building AI Factories to Strengthen the Power Grid

Today’s power systems are designed to meet peak demand but are underutilized most of the time. Flexible power distribution AI factories can optimize infrastructure design, efficiently utilize existing assets, and build new generation capacity when needed, unlocking up to 100 gigawatts of capacity in the U.S. power system. During periods of limited grid stress, AI factories can also flexibly adjust operations to reduce large-scale grid expansions required for reliability.

Many gigawatt-scale AI projects are shifting toward integrated power generation and storage because traditional grid construction timelines cannot keep pace with AI investments. However, permanently isolating generation and storage facilities from the grid can lead to underutilized capacity, increase long-term costs per AI token, and hinder energy resources from supporting grid reliability.

To address the lengthy grid connection application process, the team proposes a solution that co-builds power generation, storage, and AI factories. In the initial phase, on-site energy can serve as transitional power to accelerate factory deployment; once connected to the grid, these assets become flexible resources capable of supplying power. This hybrid approach can overcome the slow interconnection process and unlock potential power capacity.

Vistara CEO Jim Burke stated that the U.S. grid is designed to handle peak demand, which accounts for a very small portion of the year. AI factories that can flexibly adjust their power consumption based on grid conditions offer a faster solution, especially when equipped with combined heat and power (CHP) facilities, making better use of existing infrastructure. This approach helps accelerate deployment while continuing to develop long-term infrastructure.

Over the past year, Emerald AI and NVIDIA have completed flexible power tests at five commercial data centers worldwide. According to the plan, DSX Flex will be commercially deployed at NVIDIA’s AI research center in Virginia by the end of 2026, which will also be one of the first sites to adopt Vera Rubin infrastructure. Both parties plan to continue promoting this reference design to help accelerate large-scale AI infrastructure development, open new pathways for next-generation power facilities, boost regional economic benefits, and strengthen technological leadership in the energy sector.

This article about NVIDIA and Emerald AI jointly announcing the development of configurable power AI factories first appeared on Chain News ABMedia.