• On 16 March 2026, Schneider Electric and NVIDIA expanded an existing collaboration with three disclosed workstreams: a validated power-and-cooling reference design for NVIDIA Vera Rubin NVL72, an AVEVA lifecycle digital-twin architecture within NVIDIA Omniverse DSX Blueprint, and a Schneider Electric test of NVIDIA Nemotron for alarm management with human experts.
  • The disclosures concern designs, models and a vendor-reported test. They do not identify a customer, order, contract value or operating site, and they do not establish commercial availability, autonomous operation, delivered gigawatt capacity, revenue, measured energy savings or return on investment.

The announcement extends an existing collaboration

Schneider Electric framed the 16 March release as an expansion of its work with NVIDIA, not as the start of a wholly new partnership. The material contribution is a more specific interface between data-centre electrical and cooling design, NVIDIA compute architecture, lifecycle simulation and an experimental operations use case.

NVIDIA separately presented its NVIDIA Vera Rubin DSX AI Factory Reference Design and NVIDIA Omniverse DSX Blueprint as an ecosystem architecture. Both vendors use the term “AI factory” for infrastructure designed to produce and operate AI workloads; the term does not mean that the announcement documents a factory already built for a named customer.

The NVL72 reference design joins compute racks to facility systems

The Schneider Electric reference design covers NVIDIA Vera Rubin NVL72 compute racks and the shared networking, storage, CPU and support racks around them. Its disclosed parameters include 480 VAC distribution and a 45°C TCS loop. It also models MaxP and MaxQ operating modes, making the power-and-thermal assumptions visible to designers before site-specific engineering begins.

The validation named in the release is model-based: ETAP was used for the electrical model and ITD CFD for computational-fluid-dynamics analysis. That is useful evidence that the reference configuration was checked against stated assumptions. It is not field validation of a constructed facility, customer acceptance testing, certification of every site condition or proof that the design has reached commercial deployment.

AVEVA and Nemotron address simulation and operations

Schneider Electric-owned AVEVA contributed a lifecycle digital-twin architecture to NVIDIA Omniverse DSX Blueprint. The stated purpose is to connect design, simulation, construction and operations in a shared virtual model. The releases do not disclose a named operator using the architecture in production, the data integrations required at a real site, or measured schedule, energy or maintenance gains.

Schneider Electric also said it had successfully tested NVIDIA Nemotron capabilities for alarm management in collaboration with human experts. The careful reading is a vendor-reported test of an assisted workflow: the announcement does not describe autonomous control, operation without people, a production rollout or quantified improvements in alarm accuracy, response time or staffing.

A validated blueprint is not a commissioned AI factory

“Gigawatt-scale” describes the intended design scope, not capacity that Schneider Electric or NVIDIA says it has delivered. Neither release names a customer, project location, order, contract value, commissioning date or capacity brought online. No revenue attribution or commercial-availability date is supplied for the combined design, AVEVA architecture or Nemotron use case.

The vendors likewise publish no measured reduction in power use, water use, construction time, operating cost or carbon emissions, and no return-on-investment calculation. NVIDIA’s forward-looking statement warns that expected benefits, availability, specifications and partner arrangements can change. Those variables must remain prospective until a project and operating evidence are disclosed.

Why the design boundary matters

The collaboration is strategically relevant because high-density AI systems force facility and computing decisions into the same design loop. Exposing rack mix, electrical topology, cooling temperatures and operating modes in a common reference architecture can help engineers test constraints earlier and compare alternatives on consistent assumptions.

Its current value is therefore standardisation and coordination evidence, not proof of market adoption or economic performance. The next evidentiary step is a named implementation with site-specific design changes, independent or customer acceptance, commissioning data and operating measurements that can be compared with the blueprint’s assumptions.

What to watch

  • Publication of the full design assumptions, version changes and site-specific limits for 480 VAC, the 45°C TCS loop, MaxP and MaxQ.
  • A named customer, location, order, contract value, build schedule or commercial-availability date tied to the joint architecture.
  • Field or customer acceptance results that distinguish ETAP and ITD CFD model validation from performance in a commissioned facility.
  • Evidence that AVEVA data can remain synchronized with NVIDIA Omniverse DSX Blueprint across design changes and live operations.
  • Defined human-approval, failure, security and audit controls for any NVIDIA Nemotron alarm-management workflow.
  • Measured power, cooling, reliability, schedule and cost outcomes, including the denominator and baseline behind any future efficiency or ROI claim.

Sources

  • Schneider Electric, 16 March 2026: scope of the expanded collaboration, NVIDIA Vera Rubin NVL72 power-and-cooling design, AVEVA architecture, NVIDIA Nemotron test and stated evidence boundaries
  • Schneider Electric, French release: official French version used to check terminology and the same design, digital-twin and alarm-management claims
  • Schneider Electric Brazil, 27 March 2026: official Portuguese version used to cross-check the collaboration scope, technical parameters and test boundaries
  • NVIDIA, 16 March 2026: NVIDIA Vera Rubin DSX and NVIDIA Omniverse DSX Blueprint scope, ecosystem context and forward-looking caveats on benefits, availability and specifications
  • Schneider Electric Reference Design 113: technical design scope and assumptions for electrical distribution, liquid cooling, ancillary racks, MaxP and MaxQ, ETAP and ITD CFD modelling