Inside the Siemens-NVIDIA-Fluence 136 MW AI Data Center Blueprint for Vera Rubin NVL72

When NVIDIA talks about "AI factories," it envisions data centers that churn out tokens the way traditional factories produce physical goods. The practical challenge is that extreme-density AI clusters do not slot neatly into conventional data center power designs. On June 1, 2026, Siemens, NVIDIA, and Fluence—with nVent-aligned design parameters—published a single reference architecture that aims to close that gap, delivering a 136 MW, Tier III electrical and controls blueprint purpose-built for the NVIDIA DSX Vera Rubin NVL72 platform .

The significance is not just another high-wattage spec sheet. This blueprint represents a deliberate shift from one-off, stick-built data center power infrastructure toward a pre-engineered, modular product that hyperscalers, colocation providers, and specialized cloud operators can deploy repeatedly across global sites .

Key specifications at a glance

The architecture defines a clear electrical and operational envelope:

• Total facility capacity: 136 MW
• IT load: 100 MW
• Target compute platform: NVIDIA DSX Vera Rubin NVL72, with explicit support for DSX MaxLPS
• Reliability tier: Tier III concurrent maintainability—any single component can be taken offline for service without disrupting IT operations
• Utility connection voltage: Nominal 34.5 kV

Electrical architecture: from the utility yard to the rack

The design covers the entire electrical path. It starts at the 34.5 kV utility interface, moves through medium-voltage distribution and modular low-voltage power blocks, and ends at the rack-level power connection . The architecture is not just a collection of one-line diagrams; it integrates a centralized Integrated Data Center Management Suite that provides a single-pane-of-glass view across power, cooling, and compute domains .

This convergence of IT and operational technology (OT) is central to the design philosophy. The management suite and embedded automation systems allow operators to monitor and control the entire chain as one system rather than stitching together separate building management and DCIM tools after construction.

Modular, prefabricated building blocks

Instead of requiring each site to undergo custom switchgear assembly and field wiring, Siemens designed the architecture around pre-engineered, prefabricated, and factory-tested medium- and low-voltage skids . These skids arrive on site as complete, tested units, which reduces on-site labor, shortens commissioning timelines, and improves repeatable quality and safety across deployments.

Capacity is also designed to be modular. The architecture uses repeatable electrical building blocks sized to NVIDIA DSX Vera Rubin deployment units. An operator can start with tens of megawatts and phase in additional blocks to reach hundreds of megawatts or more, all without a fundamental redesign of the core electrical topology .

How Fluence battery storage changes the equation

Fluence contributes grid-scale battery energy storage built on its Smartstack platform, integrated directly into the facility power architecture . The storage provides capabilities beyond simple backup:

• Voltage and frequency ride-through
• Black start capability
• Demand response participation
• AI load smoothing

The commercial logic is equally important. Large AI data centers often face multi-year utility interconnection delays because their load profiles look unpredictable—and therefore risky—to grid operators. Fluence batteries shape the load and coordinate ramp rates, making the demand profile more predictable and easier for utilities to approve. In some cases, this can convert a power-constrained location into a viable data center site by deploying storage in months rather than waiting for years of grid infrastructure upgrades .

Partner roles across the ecosystem

The blueprint draws on a deliberate division of expertise:

Partner	Primary contribution
Siemens	Lead architect—industrial-grade power distribution, automation, digital twin modeling, prefabricated MV/LV skids, and the Integrated Data Center Management Suite
NVIDIA	Defines the DSX Vera Rubin NVL72 and DSX MaxLPS compute platforms and the AI factory concept that the electrical design must serve
Fluence	Grid-scale Smartstack battery storage for resilience, grid interconnection acceleration, and load shaping
nVent	Provides aligned electrical design parameters for NVIDIA workload compatibility; a forthcoming supplement will add advanced thermal management, building on nVent's over 2 GW of deployed liquid cooling capacity globally
Emerald AI	AI-based workload orchestration that shifts compute tasks in time and location to align with real-time grid conditions; Emerald AI is a strategic Siemens investment
PhysicsX	Physics-informed AI modeling for power distribution design and predictive thermal management inside the data center

Emerald AI and PhysicsX signal that Siemens is not just delivering hardware but embedding software that actively coordinates compute workloads with power availability—a deeper IT/OT convergence than typical data center reference designs.

How the blueprint builds on the December 2025 Siemens-nVent design

In December 2025, Siemens and nVent published a joint reference architecture targeting NVIDIA GB200 NVL72 systems at a 100 MW scale, pairing Siemens electrical and automation systems with nVent liquid cooling technology . The June 2026 blueprint is not a replacement but a purposeful expansion in three dimensions:

1. Scale: Total facility capacity grows from 100 MW to 136 MW .
2. Storage integration: Fluence battery storage was absent from the original Siemens-nVent design and is now a core architectural component .
3. Target platform evolution: The design shifts from GB200 NVL72 to the next-generation DSX Vera Rubin NVL72, with explicit support for DSX MaxLPS configurations .

The nVent electrical design parameters carry forward, preserving compatibility, and a forthcoming supplement promises to extend the collaboration into advanced thermal management .

What the shift toward pre-engineered power infrastructure signals

Data center power infrastructure has long been a custom-engineered discipline: every project starts with a site survey, a unique one-line diagram, and weeks of field assembly. This blueprint breaks that pattern in several ways:

• Factory-built skids replace field construction. Prefabricated electrical skids reduce execution risk, compress schedules, and make the design repeatable from one region to another .
• Phased scalability removes the redesign penalty. Operators can commit to an initial deployment knowing that future expansion uses the same electrical building blocks, which also makes the design more bankable .
• Digital twin integration shortens commissioning. Siemens automation and simulation tools allow operators to validate system behavior before physical commissioning .
• IT/OT convergence treats the data center as one system. Emerald AI coordinates compute with grid conditions while Siemens controls manage power distribution, turning the facility into an integrated “AI factory” platform .

Ruth Gratzke, President of Siemens Smart Infrastructure USA, summarized the operational rationale: “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” . The ultimate commercial goal is straightforward: operators of Vera Rubin NVL72 clusters need to maximize computing output and token production within fixed power envelopes while compressing the time to first revenue .

This blueprint does not eliminate custom data center engineering—geology, local grid conditions, and regional permitting still demand site-specific work—but it does shrink the bespoke portion of the build to those elements that are genuinely site-unique. Everything else can arrive on a truck, pre-tested and ready to energize.