Siemens, with Nvidia, Fluence, and nVent, released a UL-aligned reference architecture for the Nvidia DSX Vera Rubin NVL72 platform on June 1. Siemens' release describes an end-to-end electrical, power, and controls solution sized at 136 MW of total facility capacity supporting 100 MW of IT load, with a 34.5 kV utility connection, Tier III concurrent maintainability, and the whole thing prefabricated in factory-tested skids that minimize on-site construction.
The point of a reference architecture is to give every operator and EPC firm a known-good way to deploy at scale without having to design the integration from scratch. Siemens is providing the medium-voltage distribution and the modular low-voltage power blocks that hand off to the rack. Fluence is providing the battery energy storage that handles grid resilience and load smoothing. nVent is supplying liquid cooling tied into the same electrical stack, leveraging the over 2 GW of liquid cooling the company says it has already deployed globally. Nvidia is providing the platform spec the whole thing is engineered against. The output is a Vera Rubin reference plant that an EPC can quote against the next day.
Two technical features stand out. First, the architecture is scalable from tens to hundreds of megawatts in phases, which means the same design grows from a 50 MW first build to a campus without re-engineering. Second, the DSX MaxLPS support is explicitly framed around maximizing computing output within a fixed power envelope, which is the operator's actual constraint in 2026 and 2027. The grid will not give you more power. The reference design assumes that and optimizes for output per watt.
Fluence sitting in the reference design is a quiet but meaningful shift. Until recently, battery storage at the substation was an add-on the operator negotiated with the utility. Putting it inside the reference architecture, on the same controls plane as the cooling and the medium-voltage distribution, makes BESS standard equipment. It is how you smooth the load profile so the utility interconnection clears, and it is how you ride through transients without dropping the AI workload. The SoftBank battery move into AI power and the Fluence integration here point at the same conclusion. BESS is now part of the cooling and power stack, not a separate project.
nVent's named role in the reference architecture is the clearest commercial signal in the release. It tells the Vera Rubin buyer who the default liquid cooling vendor is, and it tells competing vendors that they need to be qualified into similar reference packages or be designed in by exception. That is the same dynamic that played out with cold plate vendors on Blackwell. The Vera Rubin production lock and the supply chain race are now joined by a published reference design that picks preferred suppliers. Vendors not in the package will spend the next twelve months trying to get in.