For years, liquid cooling has dominated the spotlight as the industry’s innovative answer to rising AI-driven heat loads. But what was once viewed as an emerging technology has quickly become essential infrastructure inside modern high-density data centers.. As rack densities rapidly surpass 100kW – with some deployments pushing far beyond this threshold – the cooling conversation has shifted once again..
Today, the challenge is no longer simply how to deploy liquid cooling on day one – it’s designing systems that can scale efficiently for tomorrow, adapt to changing compute demands, and remain reliable over the long term.. As a result, reference architectures and modularization are emerging as essential tools for long-term success.
Fluid quality, filtration, and material compatibility are also climbing the priority list as operators look to protect the longevity of compute infrastructure.. In a recent DCD>Broadcast episode, experts from nVent explore why liquid cooling is now unavoidable, and what it takes to build scalable cooling infrastructure capable of supporting the next generation of AI deployments..
More than an upgrade. Legacy approaches are out the window. The increases in power aren’t incremental – they’re seismic Matthew Archibald, nVent.
“This isn’t just a paradigm shift – it’s brand new,” says Matthew Archibald, director of technical architecture at nVent. “We’re now dealing with hundreds of kilowatts per rack, and that’s only going to rise.”. The shift in data center cooling isn’t about simply replacing air with liquid.
It’s altering how facilities are designed, how systems are serviced, and the ways in which infrastructure is scaled.. “We’re deploying these systems at the largest compute scale the industry has ever seen,” continues Archibald. “Legacy approaches are out the window.
The increases in power aren’t incremental – they’re seismic.”. Each increase in rack density introduces a new layer of complexity into the infrastructure mix. More CDUs, more pumps, more plumbing infrastructure, and more demanding service procedures all become part of the equation..
And as a result, operators must carefully consider flow rates, pressure differentials, pipe sizing, and long-term scalability from the earliest planning stages.. Crucially, liquid cooling is no longer an enhancement layered onto existing infrastructure – it’s now foundational to system functionality.
Compute simply cannot run unless the liquid is flowing.. Jason Matteson, director of product management at nVent, argues that this interconnectedness fundamentally changes how facilities must be planned:. “The liquid cooling and the CPU or GPU inside the rack have to connect all the way out through the plumbing to a CDU and ultimately to the facility water.
The level of pre-planning this requires is now far more critical.”. Deployment challenges now begin long before equipment reaches the data hall floor. Decisions around infrastructure, fluid selection, material compatibility, and scalability must often be made months – or even years – in advance to support not only smooth installation, but long-term flexibility..
Managing the scale. Despite the scale of change, the industry has little time to adjust gradually. AI demand continues to accelerate, making scalable and reliable liquid cooling infrastructure an immediate requirement.
Achieving this at scale requires standardization.. “We need to build this technology at scale, and we need repeatable manufacturing to do that,” says Patrick McCarthy, staff engineer at nVent. “Using a reference architecture enables us to build the same thing over and over again and take advantage of economies of scale.”.
Reference architectures are becoming increasingly important as critical roadmaps for scalable deployment. They allow operators and manufacturers to align around proven designs, standardized components, and predictable manufacturing processes.. The scalability challenge is compounded by the speed at which rack power continues to rise.
As densities increase, so do cooling flow requirements, placing additional pressure on infrastructure design.. “As rack power goes up, flow requirements increase dramatically. You need visibility into what the ultimate power threshold will be so you can properly size the infrastructure,” says Matteson..
Here, reference architectures become especially valuable. They help operators balance today’s deployment requirements with future anticipated growth.. “What we’re putting in today needs to work for tomorrow,” says Archibald.
“The new reference architectures are designed to strike that balance between current and future deployments.”. Without this foresight, operators risk either underbuilding infrastructure that quickly becomes obsolete, or overprovisioning systems in ways that are unnecessarily expensive and inefficient..
Reference architectures also help stabilize supply chains during a period of rapid industry transition. Matteson explains:. “We’ve gone from an air-cooled world to a liquid-cooled world almost overnight.
Everybody is trying to figure out who to engage with, how to design their cooling loops, and how to deploy these systems.”. At the same time, operators must prioritize fluid compatibility – ensuring liquid flow won’t negatively impact cold plates, piping, or the silicon itself over time – as well as managing contamination risk..
A cold plate affected by biological growth or particulate contamination can trigger major operational disruption and potentially lead to full-loop replacement.. By aligning around common architectures and standards, the data center ecosystem can focus resources more efficiently, improve manufacturing consistency, and reduce deployment uncertainty as a whole..
The right building blocks. For many, modularity is emerging as an effective strategy for balancing scalability, ongoing maintenance, and speed of deployment.. “One of the best things you can do is choose something modular,” says McCarthy.
“You want flexibility so you can make changes quickly without tearing out and replacing infrastructure.”. Telemetry and intelligent monitoring are becoming increasingly important within these modular architectures. Flow, pressure, and temperature monitoring provide critical operational visibility, helping operators manage increasingly complex cooling environments..
“In the rush to deploy quickly, we still need to ensure systems remain protected and clean,” says Archibald. “That’s where factory-built modularity becomes extremely valuable.”. Prefabrication in controlled manufacturing environments allows components to be cleaned, tested, and validated before arriving onsite.
This minimizes exposure to construction debris, welding residue, and particulate contamination that can compromise cooling loops.. “When fabrication happens entirely onsite, debris from cutting, welding, and general construction can eventually enter the system,” explains McCarthy.
“Doing that work in a controlled factory environment dramatically lowers risk.”. Matteson emphasizes just how critical cleanliness has become: “The technology cooling system loop has to be sanitary grade. It has to be clean enough that you’d be willing to drink from it.”.
Factory-prefabricated modular systems also help accelerate deployment timelines by reducing flushing requirements and simplifying installation. They arrive ready to connect, enabling operators to scale deployments in phases while minimizing downtime and disruption.. “This is where AI factory reference architectures become so valuable.
You can deploy in chunks as you grow,” says Archibald.. Disrupting the market. Liquid cooling is no longer confined to hyperscale environments.
AI is reshaping cooling requirements across enterprise, financial services, and colocation providers alike.. The transition is especially disruptive for enterprise operators accustomed to traditional air-cooled environments. Moving from 10kW air-cooled racks to 250kW liquid-cooled racks introduces entirely new operational challenges, particularly for facilities never designed to handle these thermal loads..
“It’s different strokes for different folks,” says Archibald. “Hyperscalers are building large-scale architectures that work economically for them, but enterprise deployments will often look very different.”. Many enterprise operators are likely to begin with smaller AI footprints and modular cooling deployments before investing in larger liquid cooling infrastructure..
Simultaneously, operators still need the same reliability, scalability, and operational assurance demanded at hyperscale. Even smaller deployments can benefit from standardized modular solutions and established reference architectures.. “You may only be buying a handful of units,” adds McCarthy.
“But if they’re built on the same modular platform being produced at scale, you still benefit from those economies of scale.”. – Getty Images. Prepared for what’s next. As the industry continues to navigate this complex transition, preparation and planning are becoming increasingly important..
Reference architectures help reduce uncertainty, but equally critical is working with experienced partners that understand the complexity of modern liquid cooling deployments.. “This is core infrastructure,” says Matteson. “Once water is running through that loop, replacing it becomes the worst nightmare if something was designed incorrectly.”.
This makes expertise, standards awareness, and long-term planning essential from the earliest stages of deployment.. Rather than acting solely as equipment providers, companies such as nVent are increasingly positioning themselves as long-term solution partners – helping customers navigate decisions spanning fluid selection, infrastructure design, modular deployment strategies, and future scalability..
As AI and the infrastructure required to support it continue to evolve with full force, liquid cooling will be increasingly central to how interconnected systems are designed, deployed, and operated across the data center.. This means that building with flexibility, scalability, and standardization at the core – via strong reference architectures, modular systems, and trusted partners – will become as fundamental to operational strategy as liquid cooling itself..
To hear more about liquid cooling readiness for the AI era, watch the full DCD>Broadcast episode with nVent’s experts, here.. More from nVent
