The blueprint for scaling data center power infrastructure

Enabling speed-to-power for AI factories The rapid evolution of AI infrastructure is transforming data center power strategy, making it imperative to design power systems alongside compute architecture. An on-demand web event on Factor This, Speed-to-Power for AI Factories: The Blueprint for Scaling Data Center Power Infrastructure, explores the DSX Design Blueprint, a technical framework for planning and optimizing gigawatt-scale AI factories, integrating grid interconnection strategy, substation design, energy storage, power distribution, load flexibility, and utility coordination into a unified approach. Nicole Ware, senior director of software and controls product management at Fluence Energy, examines the new AI infrastructure equation and dives into why power strategy has become increasingly important to data center development. “AI isn’t just using more power; it’s really changing the shape of the demand itself,” she said. “Workloads are concentrating power within individual racks and creating changes in demand as computational workloads start to stop, start, and shift. Inside the facility, that means the power system has to support dense and dynamic computing equipment without compromising performance, while outside the facility, utilities and grid operators need the confidence that these large loads are going to behave and recover from disturbances in a predictable way.” Sai Somayajula, a principal electrical design engineer at NVIDIA, joins the conversation to share how next-generation AI workloads are changing the realities of rack density and facility design. Ella Zhou, senior director of data center technologies and partnership at Siemens, details the electrical backbone needed to support such facilities, including power distribution, automation, and controls. Ware also covers the role of battery energy storage in day-one power strategies that support faster deployment, operational resilience, load management, and more predictable grid integration. Watch the on-demand event to learn how leading developers are using parallel power and compute infrastructure planning to accelerate deployment timelines, mitigate power constraints, improve operational resilience, and support the reliable scaling of next-generation AI workloads. Also on tap: the engineering and infrastructure considerations shaping the future of AI-ready data centers, from interconnection and energization to long-term operations and capacity expansion. How The DSX Design Blueprint Answers Common Data Center Questions Zhou says every conversation she’s had with data center owners recently comes back to the same three questions. How fast can I get power to my racks? How do I keep this running reliably at scale? And how do I build today without locking myself out of what’s coming next from NVIDIA? To answer them, Siemens Energy relies on the DSX Design Blueprint. “This reference design addresses all three, not as a finished engineering package for a specific site, but as a validated architecture with redundancy philosophy, equipment selection, topology, protection coordination, and automation design developed jointly with NVIDIA and reviewed against these DSX AI factory requirements,” Zhou explained. “Your site conditions, your utility, your layout still shape the detailed engineering, but instead of starting from scratch, you start from a proven baseline, and that eliminates months of design iterations.” What Zhou finds most exciting is that for the first time, we have access to reference architecture where the power system, the cooling system, the energy storage, and the compute layer are designed as one integrated system from day one. “They’re not stitched together after the fact. And that’s what the Siemens partnership ecosystem, including NVIDIA and Fluence, enables. And as GPU architectures continue to evolve, this foundation is built to evolve with them,” she added. Designing the Future NVIDIA is designing future chips to deliver greater AI performance while helping data centers manage the higher power density and dynamic demand of next-generation workloads. “There are two things here. The first is power dynamics,” articulated Somayajula in the live event on Factor This. “We have introduced something called intelligent power smoothing, and we expect that to contain almost all of the fluctuations within the rack. We actually strive to continue doing that with our future generation chips as well, as we move toward an 800 Volt DC system.” The second item is how NVIDIA is thinking about the power density in the rack. “As the power density within the rack scales, we are projecting up to maybe a megawatt per rack, Somayajula revealed. “We are actively working on the 800 Volt DC architecture, engaging with the industry and then the wider community and openly sharing the reference architectures. We’re working with OEM partners (as well) to make sure we get there responsibly.”