Siemens unveils AI data centre blueprint with Fluence

Siemens has developed a reference architecture for NVIDIA AI data centres with Fluence and design input aligned with nVent. The design is aligned with NVIDIA DSX Vera Rubin NVL72 systems.

Aimed at hyperscalers, colocation providers and specialist cloud infrastructure operators building facilities for AI workloads, it translates NVIDIA's AI factory model into an electrical, power and controls design that can be deployed at data centre scale.

The reference design outlines a facility with total capacity of 136 MW and an IT load of 100 MW. It covers the path from the utility connection at a nominal 34.5 kV through medium-voltage distribution and modular low-voltage power blocks to the rack interface.

That focus reflects the growing pressure AI systems place on data centre infrastructure, particularly in power delivery and cooling. Operators are also facing broader questions around site selection, grid access, capital spending and how quickly new sites can begin generating revenue.

The baseline architecture targets Tier III concurrent maintainability, meaning a single component can be removed from service without disrupting IT operations. It also uses repeatable electrical building blocks aligned with NVIDIA deployment units, allowing operators to expand capacity in phases from tens of megawatts to hundreds of megawatts without redesigning the core layout.

The design includes electrical parameters aligned with nVent requirements for NVIDIA workloads and system architectures. A later supplement is expected to add more detail on thermal management.

High rack densities are emerging as one of the main constraints in AI infrastructure projects. Traditional air-cooled data centre layouts are under pressure as operators prepare for systems that require far greater power and liquid-cooling support at rack level.

"nVent has deployed more than two gigawatts of liquid cooling capacity globally," said Sara Zawoyski, President of nVent Systems Protection.

"That operational experience allows us to help partners like Siemens translate reference architectures into deployable thermal solutions that perform reliably from day one at this scale. Platforms like NVIDIA Vera Rubin NVL72 are pushing rack densities well beyond what traditional air-cooled infrastructure can support."

The design is intended to support NVIDIA DSX MaxLPS and help operators increase computing output within fixed power limits. A key goal is to shorten deployment schedules and reduce execution risk on large AI data centre projects.

Factory-built equipment is central to that approach. Siemens is using prefabricated and factory-tested medium-voltage and low-voltage skids to reduce on-site construction work and shorten commissioning.

"Siemens' deep expertise in power systems and controls engineering, modular infrastructure, protection, and industrialized delivery is really evident in this latest joint reference architecture design," said Ruth Gratzke, President of Siemens Smart Infrastructure USA.

"Our pre-engineered, prefabricated, and factory-tested medium- and low-voltage skids help minimize on-site construction complexity, shorten commissioning cycles, and improve quality, safety, and repeatability across deployments. Further, our automation and digital twin strategies deployed in this reference help ensure that facilities are brought online faster and with greater potential to produce tokens at scale."

Storage role

Fluence's contribution to the design centres on battery storage. Battery energy storage is included to give AI facilities more flexibility and resilience where grid connections are constrained or where operators need greater stability as loads change rapidly.

Fluence's Smartstack platform is intended to support functions including voltage and frequency ride-through, black start, demand response and AI load smoothing. These measures are designed to help data centres maintain operations and manage power quality while scaling up.

"Our Smartstack platform is central to this new architecture, transforming the grid into an accelerator for compute," said Jeff Monday, Chief Growth Officer at Fluence.

"By providing essential capabilities like voltage and frequency ride-through, black start, grid demand response, and AI load smoothing, we are enabling our customers to build the AI factories of the future faster and more reliably."

Single control layer

The blueprint also includes a centralised Integrated Data Centre Management Suite. The system gives operators a single view across power, cooling and compute infrastructure, reflecting the tighter coordination required as data centres move to denser AI installations.

The project highlights how suppliers to the data centre market are reshaping products around AI infrastructure rather than traditional cloud facilities. As GPU clusters become larger and more power-hungry, vendors are placing greater emphasis on prefabricated electrical systems, on-site energy storage and liquid-cooling designs that can be repeated across multiple sites.

For operators, the attraction is not only scale but also the ability to add capacity in stages while keeping the wider site architecture intact. The design supports phased expansion without fundamental redesign, with the facility model extending beyond 100 MW of IT load.

The architecture is integrated with a centralised management suite that provides what Siemens describes as a single-pane-of-glass view across power, cooling and compute infrastructure.