AI Factories Go Utility-Scale as Siemens Debuts 100 MW IT Design

retained text

AI Factories Go Utility-Scale as Siemens Debuts 100 MW IT Design
New reference architecture combines battery storage, electrical systems, cooling, and Nvidia Rubin technology for 100 MW-class AI facilities.
AI infrastructure vendors are pushing beyond servers and cooling as data centers grow into utility-scale campuses. Siemens on Tuesday introduced a reference architecture developed with Nvidia, Fluence, Vertiv, and other partners that combines electrical infrastructure, battery energy storage systems (BESS), liquid cooling, and Nvidia’s upcoming Vera Rubin platform.
Siemens said the design aims to provide operators with a validated blueprint for large-scale AI campuses while reducing engineering complexity and deployment risk. The new reference architecture will support 100 MW of IT load across a 136 MW facility. Beyond compute infrastructure, the design incorporates Siemens electrical systems, Fluence’s BESS technology, and Vertiv’s liquid cooling.
AI Factories Diverge from Conventional Builds
Alex Cordovil, research director at Dell’Oro Group, said the announcement illustrates how AI infrastructure is diverging from conventional data center designs.
“This isn’t an incremental upgrade of a conventional data center – it confirms that the AI factory has become its own class of facility, engineered end to end around the compute platform rather than assembled from separately optimized power and cooling systems," he told Data Center Knowledge.
Cordovil described the development as a “bifurcation” between traditional data centers and purpose-built AI factories designed around platforms such as Nvidia’s Vera Rubin NVL72. “Tellingly, this is first a power, electrical, and controls architecture, with advanced cooling following as a supplement – and that sequence is the real signal,” he added.
“Power and grid access, not cooling, are the binding constraint on AI infrastructure today, and a battery layer built to smooth AI load and ride through grid events in power-constrained sites makes that explicit.”
Designing from the Substation Inward
Siemens design spans utility interconnection and medium-voltage distribution through facility controls, electrical protection systems, energy storage, and AI compute. As AI cluster sizes increase, operators must manage far more than servers and cooling systems. Utility service, power quality, resilience, and electrical distribution are becoming central design considerations for facilities measured in tens or hundreds of megawatts.
The architecture’s scale reflects that shift. At 100 MW of IT load and 136 MW of total capacity, the design targets deployments that increasingly resemble industrial facilities in their power requirements. According to Siemens, the architecture was developed to provide a repeatable deployment model that helps operators shorten planning cycles and reduce integration risk.
In this blueprint, battery storage occupies a larger role than traditional backup power. Fluence said its Smartstack platform is designed to support functions such as voltage and frequency ride-through, black start, demand response, and AI load smoothing, positioning storage to support grid interaction, resilience, and workload-related fluctuations rather than serving solely as backup.
The inclusion of battery energy storage alongside power distribution, controls, and cooling systems highlights how energy infrastructure is becoming a more active component of AI facility operations.
Industrializing AI Infrastructure
The move also reflects Nvidia’s effort to build an ecosystem around the “AI factory” concept. Over the past several years, the company has expanded from GPUs and servers into networking, cooling, power, and facility-level integration through partners.
“The deeper shift is industrialization,” Cordovil said. “Prefabricated, factory-tested building blocks that scale from tens to hundreds of megawatts move the competition toward time-to-revenue and execution risk rather than bespoke engineering.”
Rising power density is accelerating that shift. “We are seeing the increased power density [means] the architecture must evolve, and we are reaching a point where the new architecture is not upgradable from the previous,” said Roy Illsley, chief analyst at Omdia, in comments to Data Center Knowledge.
As operators pursue larger AI campuses, the industry is confronting a reality that underpins Siemens’ new architecture: the infrastructure supporting AI workloads is becoming as strategically important as the compute systems themselves.

document fetched 2026-06-04 09:47