Accelerating the demand for higher rack power densities, AI is forcing data centers to address their mounting thermal challenges. The rapid pace of industry innovation, driven by continual GPU advancements, is pushing power and thermal requirements to unprecedented levels, requiring a fundamental shift in cooling strategies.

To meet these demands, liquid cooling is increasingly being adopted, either incrementally or through hybrid air-liquid systems, enabling data centers to support high-density workloads more efficiently. However, while liquid cooling offers significant benefits, it also introduces new complexities and challenges that must be carefully navigated.

At Vertiv’s recent Cooling Innovation Day, industry experts gathered to discuss the latest advancements in direct-to-chip and immersion cooling technologies, weigh financial trade-offs, and explore how to best prepare for the future with an optimal balance of air and liquid cooling solutions.

History repeating itself

Today, GPUs have become indispensable for accelerating AI development. While early servers relied on traditional air cooling, increasing power and performance requirements have rendered air cooling insufficient. As a result, liquid cooling has emerged as a necessary evolution and is poised to become standard in future IT equipment due to its ability to manage high heat loads.

Over the past few years – and accelerating in the last two – liquid cooling solutions have shifted from general facility-level cooling to targeted approaches that deliver liquid directly to high-heat components via cold plates and immersion systems. Among these, direct-to-chip cooling has become the dominant architecture.

Eric Williams, SVP of solution engineering at Compass Datacenters, who brings 25 years of industry experience, reflects on this trend:

“It feels like history is repeating itself – liquid cooling from the old mainframe days is making a comeback, coming full circle.”

Historically, the industry has struggled with limited collaboration between chip manufacturers, hardware vendors, and data center operators, resulting in technology mismatches and infrastructure readiness gaps. But as AI data centers grow increasingly complex, thoughtful design becomes critical to optimizing infrastructure and operational workflows.

Key considerations focus on timing and integration methods, monitoring and serviceability, and striking the right balance between performance and cost – especially as liquid cooling technologies and CDU selection evolve rapidly in response to these new challenges.

Best practices for implementing CDUs

Over the course of the Cooling Innovation Day, it became clear that successful CDU implementation hinges on strong collaboration, transparency, and the use of modular, standardized building blocks that enable scalability and adaptability.

Adaptability, redundancy, and customization remain essential to meeting varied customer needs, with transparency becoming a more prominent industry focus. Steve Madara, VP of Vertiv’s thermal data center department, draws on decades of HVAC expertise to guide global clients toward optimal thermal solutions. He notes:

“Companies like Nvidia are helping by sharing long-term roadmaps, allowing infrastructure providers to prepare for the future. While those plans aren't always fixed, they give needed direction.”

George Hannah, director of global core thermal management, focuses on compact chilled water systems engineered to support rising liquid cooling needs. As Hannah emphasizes:

“We often talk about features and efficiency, but in data centers, reliability and uptime are king – and it's the control systems that make that possible by orchestrating everything, even through failure scenarios.”

Vertiv offers a comprehensive suite of cooling technologies, including direct-to-chip liquid cooling, hybrid solutions, and advanced chillers. What sets Vertiv apart is its sophisticated control systems that seamlessly coordinate cooling across diverse technologies – chillers, DX systems, and liquid cooling – enabling precise management of temperature, flow, and pressure. This tight integration with facility infrastructure maximizes efficiency while maintaining the reliability and uptime critical to demanding data center operations.

Flexible infrastructure that evolves with you

Flexibility in data center infrastructure is fundamentally about adaptability – being prepared for changing, often unpredictable future requirements. According to Williams, many customers struggle to precisely define flexibility, but as he puts it:

“One customer put it simply: ‘We don’t know what the change is, we just know change is coming – so give us options.’”

Liquid cooling’s rapid adoption has challenged traditional data center designs, and with workloads and technology roadmaps evolving rapidly, infrastructure must enable seamless swapping and upgrading of components with minimal disruption.

Vertiv addresses this by managing targeted temperature differentials, allowing efficient handling of high return temperatures – up to 60°C (140°F) – while balancing heat exchange sizing, power consumption, and overall system efficiency. Nigel Gore, senior director of liquid cooling at Vertiv, oversees systems that capture and transfer heat directly from racks and chips, explaining:

“Managing a 4°C (39.2°F) approach temperature difference while handling chip return temps up to 60°C is key to running liquid cooling efficiently – and that’s where our integrated technology orchestration really delivers.”

Madara highlights the complexity of the transition phase, where hybrid environments blending air- and liquid-cooled equipment require careful management to maintain flexibility throughout refresh cycles:

“Currently, pods might be around 70 percent liquid-cooled,” he notes, “but future compute racks may reach 95 percent liquid cooling, meaning significantly greater cooling liquid volume and flow in the coming years.”

Despite uncertainties about exact future demands, the priority is building adaptable, modular infrastructure that can be easily customized, upsized, or reconfigured as needs evolve.

Tailored cooling solutions

One of the key lessons from the evolving data center landscape is that uncertainty about future needs has often led to oversized infrastructure. Vertiv takes a different approach, focusing on helping customers right-size their cooling systems with adaptable, scalable solutions that balance performance and cost effectively. As Hannah notes:

“If it’s not in our portfolio, we’ll design it – working hand-in-hand with customers from concept to deployment to deliver exactly what they need.”

Hybrid cooling solutions, which combine liquid and air cooling, are essential today since no data center is fully liquid-cooled. These hybrid approaches optimize liquid cooling for high-heat components like chips, while using air cooling to support other equipment. Designed with adaptability in mind, they accommodate shifting cooling demands over time.

Ben Smith, VP of DX thermal systems, leads refrigerant-based and hybrid cooling technologies such as Vertiv CoolPhase Flex. Highlighting this flexibility, Smith says:

“Today, customers might deploy 20 percent liquid cooling and 80 percent air cooling, but with our modular solutions like Vertiv CoolPhase Flex and Vertiv CoolLoop Trim Cooler, they can easily scale to reverse that ratio within a few years and avoid costly overhauls.”

Madara explains the careful planning involved:

“We’re working closely with customers to model and envision future data centers, exploring various cooling methods to find the best fit. Our goal is to help customers transition thoughtfully from air to liquid cooling while avoiding costly wholesale replacements.”

Scalable, proven cooling solutions for high-density demand

At the forefront of liquid cooling innovation, Vertiv’s experts are shaping efficient, scalable solutions to meet the soaring thermal demands of modern data centers. This evolution is driven by silicon vendors demanding more efficient cooling options, notably through direct-to-chip single-phase cooling and exploring two-phase and immersion methods.

The rapid growth of liquid cooling has accelerated collaboration across silicon and IT vendors, bridging industry knowledge gaps in coolant chemistry, system cleanliness, and material compatibility.

Srdjan Mutabdzija, Vertiv’s head of product management for liquid cooling, spearheads efforts to align existing and emerging technologies with these goals. Mutabdzija highlights the deepening partnerships:

“What excites me most is the daily collaboration with GPU and IP vendors – they know that deploying high-performance chips depends on the right infrastructure. That partnership is the best part of my job.”

As GPUs and CPUs drive new cooling requirements, data center design is shifting from an outside-in approach to focusing first on server-level needs. Mutabdzija emphasizes:

“Think about what you need at the server level first, then design outward so that your solution works today and scales as rack densities double.”

To succeed, industry-wide standardization, scalability, and close alignment with silicon roadmaps are critical, alongside careful attention to environmental regulations and innovative approaches to future thermal challenges. Gore adds:

“Success in liquid cooling hinges on deep technical expertise, rigorous quality control, and seamless facility integration – areas where Vertiv leads.”

Looking ahead

Future trends in data center cooling increasingly focus on heat recovery and reuse, signaling a shift toward fully integrated thermal ecosystems. With rising water temperatures and a growing emphasis on on-site power generation, data centers now have greater opportunities to capture and repurpose waste heat. As Hannah notes:

“Heat recovery and reuse is literally heating up as a topic. Advancements in chip design and rising water temperatures have brought this to the forefront.”

Over the past six months, momentum has accelerated alongside the global push for localized power solutions to ease grid demand, making the pairing of liquid cooling with heat-rich power sources a promising avenue for responsible operations.

Vertiv’s portfolio already includes water-cooled chillers and CDUs that enable heat recovery, and ongoing development aims to maximize the potential for heat reuse. Hannah reflects:

“Some deployments already feature a full suite of thermal solutions integrated under one control system. While heat recovery is still niche today, it’s poised to play a much larger role moving forward. We’re delivering a comprehensive thermal chain, supported by a strong power ecosystem.”

Together, these trends point toward a more environmentally responsible and adaptable future for data center thermal management – one where heat is not wasted but becomes a valuable resource within an integrated energy ecosystem.

See more from Vertiv’s Cooling Innovation Day here.