A blueprint that removes friction from deployment
As artificial intelligence reshapes the demands placed on physical infrastructure, Schneider Electric and NVIDIA have joined forces to offer data centers something rarely available in moments of rapid technological change: a proven path forward. Their jointly developed reference designs address the compounding challenges of power density, thermal management, and system integration that accompany the deployment of next-generation AI hardware. In offering standardized blueprints capable of supporting up to 142 kilowatts per rack, the two companies are not merely solving an engineering problem — they are attempting to set the tempo for how the world builds the physical foundations of artificial intelligence.
- Data centers face mounting pressure as AI workloads outpace the infrastructure designed to support them, creating costly delays and compatibility failures between power, cooling, and computing systems.
- The divide between operational technology and information technology — long treated as separate domains — has become a critical bottleneck, slowing deployment and eroding efficiency in facilities trying to scale AI.
- Schneider Electric and NVIDIA are responding with validated reference designs that unify these systems under standardized controls, removing the need for operators to engineer solutions from scratch.
- Supporting up to 142 kW per rack, the designs are built specifically to accommodate NVIDIA's Grace Blackwell processors, reflecting the raw electrical and thermal demands of modern AI training and inference.
- Beyond speed, the collaboration carries a sustainability dimension — by optimizing power and liquid cooling from the ground up, the designs aim to reduce the environmental footprint of infrastructure that is rapidly becoming one of the world's largest consumers of electricity.
Schneider Electric and NVIDIA have unveiled a set of reference designs aimed at accelerating how data centers deploy artificial intelligence systems. The collaboration addresses a persistent structural problem: as AI workloads grow more demanding, the power systems, cooling apparatus, and control mechanisms required to support them must evolve together — yet these components have historically operated as separate, poorly integrated worlds.
At the core of the partnership is the challenge of bridging operational technology and information technology. Data centers have long managed these domains independently, each governed by its own protocols and management systems. When forced to coexist without proper integration, the result is inefficiency, compatibility failures, and slower deployment timelines. The new reference designs resolve this by enabling seamless interoperability between the two, managed through unified interfaces rather than a patchwork of vendor systems.
The designs support up to 142 kilowatts per rack — a capacity calibrated to the real demands of NVIDIA's Grace Blackwell processors and the liquid cooling systems required to manage the heat they generate. Rather than asking each operator to design bespoke infrastructure, Schneider Electric and NVIDIA are offering pre-validated blueprints: proven configurations that reduce friction and compress the time between planning and deployment.
The announcement arrives as data centers worldwide race to build or upgrade facilities capable of supporting large language models and generative AI applications. Every delay carries competitive cost. By standardizing the path to high-density AI infrastructure, the two companies are also making an implicit argument about sustainability — that optimizing power management and cooling from the outset, rather than retrofitting existing systems, can meaningfully reduce both operating costs and the environmental footprint of an industry under growing scrutiny.
Schneider Electric and NVIDIA have jointly unveiled a set of reference designs intended to accelerate how data centers deploy artificial intelligence systems. The partnership addresses a practical problem: as AI workloads grow more demanding, the infrastructure required to run them—the power systems, cooling apparatus, and control mechanisms—must evolve in tandem, and often these components don't communicate well with one another.
The designs center on a specific technical achievement: they enable operational technology (the machinery and sensors that run a data center) and information technology (the computing systems themselves) to work together seamlessly. This interoperability matters because data centers have historically treated these two domains as separate worlds, each with its own management systems and protocols. When they're forced to coexist without proper integration, efficiency suffers and deployment takes longer.
At the heart of the collaboration is a focus on power and cooling—the two constraints that limit how densely you can pack AI hardware into a single rack. The reference designs accommodate up to 142 kilowatts per rack, a capacity that allows data centers to integrate NVIDIA's Grace Blackwell systems, which represent the company's latest high-performance AI processors. This power density is not arbitrary; it reflects the reality that modern AI training and inference require enormous amounts of electricity, and that electricity generates proportional amounts of heat that must be managed through liquid cooling rather than traditional air systems.
What Schneider Electric and NVIDIA are offering, in essence, is a blueprint. Rather than each data center operator designing their own infrastructure from scratch—a process that invites inefficiency, compatibility problems, and delays—they can adopt these validated designs and move faster toward deployment. The designs come with standardized controls, meaning operators can manage power distribution, cooling loops, and monitoring through unified interfaces rather than juggling multiple vendor systems.
The partnership reflects a broader shift in how infrastructure companies approach AI. The technology is no longer a future concern; it is now the primary driver of data center investment and design. Companies building or expanding facilities need to know, with confidence, that their power systems can handle the load, their cooling can dissipate the heat, and their management systems can orchestrate it all without constant manual intervention. Schneider Electric, which has long specialized in energy management and automation, brings decades of expertise in exactly these domains. NVIDIA brings the credibility of knowing precisely what its own hardware demands.
The timing of this announcement reflects market pressure. Data centers worldwide are racing to build or upgrade infrastructure capable of supporting large language models, generative AI applications, and other compute-intensive workloads. Every month of delay in deployment translates to lost revenue or competitive disadvantage. By offering pre-validated designs, Schneider Electric and NVIDIA are removing friction from that process. A data center operator can now point to these designs and say to their engineering team: this is proven, this works, build it.
The designs also carry an implicit promise about sustainability. High-density computing environments are notorious for their power consumption and heat generation. By optimizing power management and cooling efficiency from the outset—rather than bolting these systems onto existing infrastructure—the designs aim to reduce waste. This matters not only for operating costs but for the environmental footprint of AI infrastructure, which has become a topic of increasing scrutiny as the technology scales.
La Conversación del Hearth Otra perspectiva de la historia
Why does it matter that OT and IT systems work together? Aren't they already in the same building?
They are, but they've historically been managed separately—different teams, different software, different assumptions about how things should work. When you're trying to push 142 kilowatts through a single rack and cool it with liquid, you need real-time coordination between the power systems and the computing systems. Without that, you get bottlenecks and inefficiency.
So this is really about speed—getting AI systems deployed faster?
Speed, yes, but also confidence. A data center operator can now point to a design that NVIDIA has validated with its own hardware and say: we know this works. That removes a huge source of uncertainty and risk.
What's the actual constraint here? Is it power, cooling, or something else?
Both, and they're linked. You can't just add more power without adding more cooling. The designs optimize both simultaneously, which is harder than it sounds. Liquid cooling is more efficient than air, but it requires different infrastructure and different thinking about how to manage heat.
Who benefits most from this—the data center operators or the chip makers?
Both, but in different ways. NVIDIA benefits because faster deployment means more of their chips get sold sooner. Data center operators benefit because they can build faster and with less risk. The real winner is probably the customer waiting for AI services to become available.
Is this a one-time thing or the start of a pattern?
It's the start of a pattern. As AI infrastructure becomes more specialized and demanding, you'll see more of these kinds of partnerships—companies that understand hardware working with companies that understand infrastructure to create validated blueprints.