CERAWeek, known as the Davos of energy, serves as a platform for policymakers, producers, technologists, and financiers to come together and discuss the future of powering the world. At the recent conference, NVIDIA and Emerald AI introduced a groundbreaking concept of treating AI factories as flexible and intelligent grid assets rather than static power loads. This collaboration combines accelerated computing, AI factory reference architectures, and real-time energy orchestration to enable large AI deployments to connect to the grid faster, operate more efficiently, and enhance system reliability.
The foundation of this innovative approach lies in the NVIDIA Vera Rubin DSX AI Factory reference design and Emerald AI’s Conductor platform. By integrating compute, power networking, and control into a single architecture, these AI factories can generate high-value AI tokens while dynamically adapting to grid conditions. This flexibility not only supports reliability but also reduces the need for overbuilding infrastructure for peak demand.
Leading energy companies such as AES, Constellation, Invenergy, NextEra Energy, Nscale Energy & Power, and Vistra are collaborating to meet the increasing power demand by building the necessary energy generation capacity. Their focus is on developing optimized generation strategies to support AI factories based on the NVIDIA and Emerald AI architecture. These hybrid projects leverage co-located power to accelerate time to power while delivering value to the broader grid, ultimately enhancing grid reliability.
This milestone in grid resilience is part of an ecosystem for advanced AI factories. NVIDIA’s founder and CEO, Jensen Huang, describes this new computing infrastructure paradigm as a five-layer AI cake, with energy serving as its foundational layer.
The drive for improvements in tokens per second per watt is reshaping AI data centers, with a focus on computational efficiency to lower operating costs, maximize revenue, and create a resilient digital infrastructure. Huang emphasizes the importance of extreme codesign to significantly enhance tokens per second per watt annually, citing NVIDIA’s history of driving performance and energy efficiency advancements over the years.
At the event, NVIDIA ecosystem partners showcased how AI, simulation, and workforce innovation are accelerating the development of energy infrastructure for the intelligence era. Companies like Maximo, TerraPower, and Adaptive Construction Solutions demonstrated how AI, robotics, digital twins, and workforce training are compressing timelines in construction, power generation, and talent development.
Maximo, a solar robotics company incubated at AES, showcased a 100-megawatt robotic solar installation at AES’ Bellefield site, highlighting the benefits of AI-driven robotics in utility-scale operations. TerraPower, in collaboration with SoftServe, previewed a digital twin platform powered by NVIDIA Omniverse to streamline advanced nuclear plant siting and design processes. Adaptive Construction Solutions introduced a national registered apprenticeship initiative, in partnership with NVIDIA, to train the workforce for AI factories and energy infrastructure projects.
Industry leaders such as GE Vernova, Schneider Electric, and Vertiv emphasized the importance of digital twins, validated reference designs, and converged infrastructure in scaling AI factories for grid reliability. By addressing the power-to-rack challenge and designing AI infrastructure as an integrated energy and compute system from the outset, these companies are paving the way for more efficient and reliable grid participation.
In conclusion, the collaboration between industry leaders and technology innovators is driving advancements in energy solutions through AI and high-performance computing. The future of AI factories as flexible, grid-aware assets holds promise for efficiently powering the world. To learn more about how NVIDIA and its partners are shaping the energy landscape with AI, visit their website.
For more Information, Refer to this article.
