Efficiency at Scale: NVIDIA and Energy Leaders Accelerate Grid Reinforcement with Power-Flexible AI Factories

At CERAWeek, NVIDIA and Emerald AI unveiled a new approach that treats AI factories not as fixed power loads, but as intelligent assets that respond flexibly to grid conditions. The collaboration combines accelerated computing, an AI factory reference architecture, and real-time energy orchestration into a single framework, focused on speeding up grid interconnection, improving operational efficiency, and strengthening system reliability for large-scale AI deployments.

The foundation is the NVIDIA Vera Rubin DSX AI Factory reference design and Emerald AI's Conductor platform. Together, they integrate compute, power networking, and control so that AI factories can raise tokens per second per watt while also adjusting load to grid conditions when needed, reducing the burden of over-building infrastructure for peak demand.

AES, Constellation, Invenergy, NextEra Energy, Nscale Energy & Power, and Vistra are collaborating to expand generation capacity to meet surging power demand. They are jointly pursuing optimized generation strategies to support AI factories built on the NVIDIA and Emerald AI architecture, and have put forward hybrid projects leveraging on-site adjacent generation to shorten time to power while also adding value to the broader grid.

The core message is power efficiency. Jensen Huang recently emphasized in an interview that extreme codesign must improve tokens per second per watt by multiple orders of magnitude every year, and NVIDIA has increased the number of tokens generatable within the same power budget by more than 1 million times from the debut of the Kepler GPU in 2012 to this year's Vera Rubin platform.

In the energy sector as well, cases have emerged where AI, digital twins, and workforce development are accelerating construction, power generation, and deployment speed. Maximo completed a 100MW-scale robotic solar installation at AES's Bellefield site, and TerraPower, together with SoftServe, presented an Omniverse-based digital twin platform that shortens site selection and design periods for next-generation nuclear plants from years to months. Adaptive Construction Solutions, in partnership with NVIDIA, also announced a nationwide registered apprenticeship program to build the skilled workforce needed for AI factories and energy infrastructure.

At the same time, GE Vernova, Schneider Electric, and Vertiv stressed that digital twins, validated reference designs, and converged infrastructure are essential to making AI factories trustworthy participants in the grid.

• GE Vernova: High-precision digital twins aligned with the Omniverse DSX Blueprint jointly simulate the grid, substations, and AI factory loads before deployment
• Schneider Electric: Together with AVEVA, validated the Vera Rubin reference design and lifecycle digital twin architecture, optimizing power, cooling, and control in Omniverse
• Vertiv: Presented simulation-ready infrastructure based on repeatable power and cooling blocks to reduce design and deployment complexity

Ultimately, this trend redefines AI factories not simply as data centers, but as flexible infrastructure that interacts with the grid. NVIDIA's five-layer structure—energy, chips, infrastructure, models, and applications—is becoming the foundation of that transition.