CERAWeek, often called the Davos of energy, brings together policymakers, producers, technologists, and financiers to discuss the future of global energy.
At the conference last week, NVIDIA and Emerald AI introduced a new approach: treating AI factories as flexible, intelligent datasets instead of static power loads. Their collaboration combines accelerated computing, AI factory reference architectures, and real-time energy orchestration. This helps large AI deployments connect to the workload more quickly, operate more efficiently, and improve system reliability.
This approach uses the NVIDIA Vera Rubin DSX AI factory reference design and Emerald AI’s Conductor platform to combine computing, networking, and control into a single system. The result is an AI factory that generates high-value AI tokens and can adjust to grid conditions as needed. This flexibility supports reliability and reduces the need to build extra infrastructure for peak demand.
AES Constellation Energy, NextEra Energy, Nscale Energy and Power, and Vistra are working to increase energy generation capacity to meet rising demand. These companies plan to collaborate on strategies to support AI factories using the Nvidia and Emerald AI architecture. Their projects include hybrid setups with co-located power to speed up access to energy and benefit the wider grid by combining large AI loads with flexible operations, new resources, and smart controls. This approach makes the grid more reliable.
This marks an important step for grid resilience backed by a network supporting AI factories. NVIDIA founder and CEO Jensen Huang describes this new computing infrastructure as a five-layer AI cake with energy as the base layer.
Driving Improvements In Tokens Per Second Per Watt
Power limits are changing how AI data centers operate. Now, energy efficiency, measured as tokens per second per watt, is the key metric for modern computing. By focusing on computational efficiency, organizations can cut costs, boost revenue, and build a stronger digital infrastructure for businesses and consumers everywhere.
Power is a concern, but it’s not the only concern, Huang said on a recent Lex Fridman podcast. That’s why we’re pushing so hard on extreme code sign: to improve those tokens-per-second-per-watt orders of magnitude every single year.
NVIDIA has consistently improved performance and energy efficiency since the NVIDIA Kepler GPU in 2012, up to the NVIDIA Vera Rubin platform this year. The number of tokens produced with the same power has grown by over a million times.
Achieving this progress requires industry collaboration across all five layers of the AI stack, from energy and chips to infrastructure, models, and applications.
Robotics, Digital Twins, and AI Upscaling Drive Energy Advances.
At the event, NVIDIA ecosystem partners demonstrated how AI simulation and workforce innovation are accelerating the development of energy infrastructure for the intelligence era. Announcements from Maximum TerraPower and Adaptive Construction Solutions highlighted how AI is shortening timelines in construction, power generation, and workforce training.
Maximo, a solar robotics company spun out of AES, announced it has completed a 100-megawatt robotic solar installation at AES’s Belfield site using AI-powered robotics built with NVIDIA accelerated computing, NVIDIA Omniverse batteries, and the NVIDIA Isaac Sim framework. Maximo demonstrated that autonomous installations can now operate reliably at a large scale. This method speeds up installation, improves safety and consistency, and helps meet the growing demand for electricity.
TerraPower, in partnership with SoftServe, introduced a digital twin platform powered by NVIDIA Omniverse. This platform is designed to significantly reduce the time required to plan and design advanced nuclear plants by leveraging AI and simulation in early engineering. It cuts design cycles from years to months, speeds up the rollout of TerraPower’s Natrium energy plants, and improves both design and grid integration.
Adaptive Construction Solutions, working with NVIDIA, announced a national apprenticeship program to help the skilled workers needed for AI factories and energy infrastructure. The program will expand training for key trades, open up more high-demand career opportunities, and support the fast growth of AI-powered energy systems.
These efforts show how AI, digital twins, and workforce innovation are coming together to create faster, more reliable energy infrastructure.
Working Together to Scale AI Factories for Reliable Power Grids
GE Vernova, Schneider Electric, and Vertiv explained that digital twins, proven reference designs, and unified infrastructure are now key to scaling AI factories to reliably support the power grid. Their announcements focus on solving the power-to-rack challenge by designing AI systems as integrated energy and computing solutions from the start.
GE Vernova described how detailed digital twins used with the NVIDIA Omniverse DSX Blueprint let utilities and developers simulate grid behavior, substations, and AI factory loads before anything is built. This kind of modeling helps test connection strategies, lower risks, and speed up getting power online in tight grid situations.
Schneider Electric introduced new approved NVIDIA Vera Rubin reference designs and digital twins systems created with AVEVA. By simulating power cooling and controls in Omniverse, Schneider helps operators get the most out of every watt, check designs before building, and run AI factories more efficiently and reliably as they grow.
Vertiv shared its approach to building physical infrastructure that is ready for simulation and based on reputable power and cooling modules. When combined with the Vera Rubin DSX reference design, this method simplifies design and deployment, helping AI factories scale up faster and with greater confidence.
Together, these industry efforts offer a clear digital path with proven designs and infrastructure that help turn AI factories into flexible, grid-aware resources for efficient power use worldwide.
Find out how NVIDIA and its partners are using AI and high-performance computing to improve energy solutions.
