www.ptreview.co.uk
01
'26
Written on Modified on
Scalable Electrical and Power Architecture for Artificial Intelligence Infrastructure
Siemens has developed a standardized electrical and control reference design to streamline the deployment of high-density hardware clusters for hyperscale and cloud infrastructure providers.
www.siemens.com
Siemens, in collaboration with Nvidia, Fluence, and nVent, has engineered an electrical, power, and controls reference architecture aligned with the Nvidia DSX Vera Rubin platform. This integrated system is designed specifically for hyperscale data centers, colocation facilities, and specialized cloud infrastructure providers executing high-density artificial intelligence workloads.
Power Distribution Infrastructure and Capacity Scaling
The reference design establishes a complete power delivery pathway engineered for a total facility capacity of 136 MW, supporting a dedicated IT load of 100 MW. The architecture controls power distribution starting from a nominal 34.5 kV utility connection, regulating voltage down through medium-voltage distribution systems into modular low-voltage power blocks that connect directly to the server rack interface.
Constructed to meet Tier III concurrent maintainability standards as defined by the Uptime Institute, the system guarantees that any single component can be isolated and removed from service for maintenance or repair without disrupting IT operations.
[Utility Connection: 34.5 kV]
│
▼
[Medium-Voltage Distribution]
│
▼
[Modular Low-Voltage Power Blocks]
│
▼
[Server Rack Interface (Vera Rubin NVL72)]
The infrastructure utilizes repeatable, prefabricated electrical building blocks matched to specific deployment units. This structural modularity allows operators to scale capacity in phases, accommodating initial installations of tens of megawatts and expanding beyond hundreds of megawatts without requiring structural redesigns of the primary electrical topography.
Thermal Management and Pre-Engineered Deployment
To manage the thermal loads generated by intense computing hardware, the blueprint integrates structural electrical design parameters optimized for liquid cooling architectures, with formal thermal management specifications scheduled for subsequent release. Traditional air-cooled systems lack the thermal dissipation efficiency required for these high-density setups, which often exceed 100 kW per rack. Factory-assembled and pre-tested medium- and low-voltage skids are utilized to reduce structural layout errors, minimize on-site construction timelines, and shorten system commissioning cycles.
The reference design supports the DSX MaxLPS configuration, optimizing computational output and data token production within fixed, predetermined power allocations. Additionally, the infrastructure incorporates automated controls and digital twin emulation software to mirror physical assets, allowing operators to simulate power dynamics and accelerate the deployment pipeline.
Grid Integration, Storage, and Infrastructure Software
Battery energy storage integration provides operational resiliency in regions with power-constrained electrical grids. Utilizing the Fluence Smartstack platform, the energy storage system delivers automated voltage and frequency ride-through capabilities, ensuring continuous operation during transient grid anomalies. The storage architecture also supports black-start protocols to recover systems independently after a failure, manages grid demand response participation, and applies load smoothing algorithms to mitigate severe power spikes caused by fluctuating computing demands.
Centralized operation is maintained via an Integrated Data Center Management Suite. This software layer consolidates telemetry from power distribution hardware, cooling systems, and active computing nodes into a unified management interface, enabling real-time monitoring and resource balancing across the facility ecosystem.
Additional Context: Technical Specifications and Competitive Benchmarking
High-density computing deployments necessitate distinct modifications to traditional data center power distribution strategies. Traditional enterprise data center architectures typically support rack densities between 7 kW and 15 kW. In contrast, platforms such as the Nvidia Vera Rubin NVL72 require liquid cooling frameworks to sustain power densities that can scale from 100 kW to over 130 kW per rack.
To maintain efficiency at this density, the reference architecture minimizes transmission losses by implementing medium-voltage step-down transformations closer to the row level, rather than relying on centralized low-voltage distribution.
While standard Tier III systems rely on conventional uninterruptible power supply (UPS) systems purely for short-term energy bridges during utility outages, the inclusion of dedicated battery energy storage systems (BESS) allows for active load smoothing. This capability addresses the rapid transient power swings characteristic of large language model training phases, preventing voltage sags on the host utility grid.
Furthermore, traditional enterprise facilities operate with smaller, highly variable total capacities, typically ranging from 10 MW to 40 MW, and distribute power via centralized 480V or 400V setups. The Siemens reference design addresses these limitations by establishing a predefined 136 MW total facility architecture built specifically around native liquid cooling interfaces and immediate sub-distribution blocks.
Edited by Evgeny Churilov, Induportals Media - Adapted by AI.
www.siemens.com
Power Distribution Infrastructure and Capacity Scaling
The reference design establishes a complete power delivery pathway engineered for a total facility capacity of 136 MW, supporting a dedicated IT load of 100 MW. The architecture controls power distribution starting from a nominal 34.5 kV utility connection, regulating voltage down through medium-voltage distribution systems into modular low-voltage power blocks that connect directly to the server rack interface.
Constructed to meet Tier III concurrent maintainability standards as defined by the Uptime Institute, the system guarantees that any single component can be isolated and removed from service for maintenance or repair without disrupting IT operations.
[Utility Connection: 34.5 kV]
│
▼
[Medium-Voltage Distribution]
│
▼
[Modular Low-Voltage Power Blocks]
│
▼
[Server Rack Interface (Vera Rubin NVL72)]
The infrastructure utilizes repeatable, prefabricated electrical building blocks matched to specific deployment units. This structural modularity allows operators to scale capacity in phases, accommodating initial installations of tens of megawatts and expanding beyond hundreds of megawatts without requiring structural redesigns of the primary electrical topography.
Thermal Management and Pre-Engineered Deployment
To manage the thermal loads generated by intense computing hardware, the blueprint integrates structural electrical design parameters optimized for liquid cooling architectures, with formal thermal management specifications scheduled for subsequent release. Traditional air-cooled systems lack the thermal dissipation efficiency required for these high-density setups, which often exceed 100 kW per rack. Factory-assembled and pre-tested medium- and low-voltage skids are utilized to reduce structural layout errors, minimize on-site construction timelines, and shorten system commissioning cycles.
The reference design supports the DSX MaxLPS configuration, optimizing computational output and data token production within fixed, predetermined power allocations. Additionally, the infrastructure incorporates automated controls and digital twin emulation software to mirror physical assets, allowing operators to simulate power dynamics and accelerate the deployment pipeline.
Grid Integration, Storage, and Infrastructure Software
Battery energy storage integration provides operational resiliency in regions with power-constrained electrical grids. Utilizing the Fluence Smartstack platform, the energy storage system delivers automated voltage and frequency ride-through capabilities, ensuring continuous operation during transient grid anomalies. The storage architecture also supports black-start protocols to recover systems independently after a failure, manages grid demand response participation, and applies load smoothing algorithms to mitigate severe power spikes caused by fluctuating computing demands.
Centralized operation is maintained via an Integrated Data Center Management Suite. This software layer consolidates telemetry from power distribution hardware, cooling systems, and active computing nodes into a unified management interface, enabling real-time monitoring and resource balancing across the facility ecosystem.
Additional Context: Technical Specifications and Competitive Benchmarking
High-density computing deployments necessitate distinct modifications to traditional data center power distribution strategies. Traditional enterprise data center architectures typically support rack densities between 7 kW and 15 kW. In contrast, platforms such as the Nvidia Vera Rubin NVL72 require liquid cooling frameworks to sustain power densities that can scale from 100 kW to over 130 kW per rack.
To maintain efficiency at this density, the reference architecture minimizes transmission losses by implementing medium-voltage step-down transformations closer to the row level, rather than relying on centralized low-voltage distribution.
While standard Tier III systems rely on conventional uninterruptible power supply (UPS) systems purely for short-term energy bridges during utility outages, the inclusion of dedicated battery energy storage systems (BESS) allows for active load smoothing. This capability addresses the rapid transient power swings characteristic of large language model training phases, preventing voltage sags on the host utility grid.
Furthermore, traditional enterprise facilities operate with smaller, highly variable total capacities, typically ranging from 10 MW to 40 MW, and distribute power via centralized 480V or 400V setups. The Siemens reference design addresses these limitations by establishing a predefined 136 MW total facility architecture built specifically around native liquid cooling interfaces and immediate sub-distribution blocks.
Edited by Evgeny Churilov, Induportals Media - Adapted by AI.
www.siemens.com
