Dive Brief:

• Schneider Electric has co-developed two data center reference designs with chipmaker NVIDIA that will “significantly accelerate time to deployment and aid operators as they adopt AI-ready infrastructure solutions,” the French building infrastructure firm said Thursday.
• One design delivers what Schneider calls “first and only critical framework for integrated power management and liquid cooling control systems,” and operates hand in glove with NVIDIA Mission Control, the chipmaker’s orchestration platform for high-powered “AI factories.”
• Schneider said the second provides a framework for data halls using NVIDIA’s next-generation Blackwell Ultra architecture across four technical areas: facility power, facility cooling, IT space and lifecycle software. It can support rack power densities up to 142 kilowatts, exceeding most data centers’ power densities.

Dive Insight:

The new releases will streamline the design, deployment and operation of advanced AI infrastructure, Joe Simonelli, Schneider’s senior vice president and chief technology officer, said in a statement. 

“Our latest reference designs, featuring integrated power management and liquid cooling controls, are future-ready [and] scalable … enabling data center operators to keep pace with surging demand for AI,” he said.

The first reference design focuses on power and cooling control systems in AI data centers, or “AI factories” as NVIDIA calls them. Schneider said data center customers can use the design alongside Schneider’s existing data center reference design for facilities using Grace Blackwell chips, the first generation of NVIDIA’s Blackwell line.

The controls design can integrate a range of power management and liquid cooling solutions, including Schneider’s Motivair liquid cooling technologies. Schneider took a 75% stake in Motivair, a formerly independent thermal management solutions company, earlier this year. It says it plans to acquire the remaining 25% in 2028.

The power management and liquid cooling control system framework also builds on Schneider’s work on digital twins that replicate AI factory architecture and simulate operations in real time, highlighting opportunities to improve efficiency and reliability. In March, Schneider and power systems firm ETAP rolled out digital twin capabilities that leverage NVIDIA’s Omniverse Blueprint, enabling power load modeling at the chip level — what the companies call a significant improvement over the previous rack-level standard.

The result of these integrations, Schneider says, is an end-to-end controls system that provides a standardized format interface supplying data for use by “upstream” tools such as AI infrastructure management software, digital twins and other enterprise systems while mitigating redundancy across data centers’ cooling and power distribution infrastructure.

The second data center reference design details deployment of NVIDIA GB300 NVL72-based clusters with maximum rack densities of 142kW, Schneider says. It envisions a data hall that is purpose-built to support NVIDIA technology and can host “up to 1,152 GPUs using liquid-to-liquid [coolant distribution units] and high-temperature chillers,” Schneider says.

With the GB300 reference design, Schneider says users can leverage digital twins to optimize application design for specific power and cooling scenarios.

Schneider is one of scores of companies developing reference designs and other data center optimization initiatives with NVIDIA, the chipmaker says. 

At a Silicon Valley infrastructure summit earlier this month, the executive leading NVIDIA’s “accelerated computing” initiatives said the designs it’s developing with infrastructure firms like Schneider, Siemens and Vertiv would bring fully-integrated AI factories closer to fruition.

“At the heart of this vision lies a fundamental challenge: how to optimize every watt of energy that enters the facility so that it contributes directly to intelligence generation,” Madison Huang, NVIDIA’s senior director of product and technical marketing for physical AI platforms, wrote in a Sept. 9 blog post.