The data centre industry faces a transformation driven by artificial intelligence workloads that demand fundamentally different approaches to power and cooling infrastructure. 

Richard Whitmore, President and CEO of Motivair by Schneider Electric, oversees the delivery of liquid cooling technologies designed for AI, high-performance computing (HPC) and next-generation data centre environments. 

Motivair develops cooling solutions focused on scalability, efficiency and sustainability for digital infrastructure operating under increasing thermal loads. With Schneider Electric, the organisation is positioned to help global customers design and operate resilient systems that balance performance, reliability and environmental responsibility.

A new era of AI demand

Richard explains that data centre operators working with existing brownfield sites or developing new greenfield facilities must recognise the intensity of power and cooling requirements that differ from previous generations of infrastructure. 

"That shift is being driven by AI, which has been coined the fourth industrial revolution touching every aspect of our lives,” says Richard. “It's advancing rapidly, and newer GPUs leading to rack densification are fundamentally changing the way data centres and AI factories are designed, now and in the future.”

The market trajectory shows AI platforms pulling more than 20 times the power of traditional cloud servers, with roadmaps pointing towards rack densities approaching one megawatt. 

This shift elevates infrastructure from a technical consideration to a business decision. The industry previously built data centres to accommodate around 80% of customers at modest density levels, then created bespoke solutions for the 20% requiring higher density or high-performance computing. That model no longer applies.

"Today, data centres need to be designed with the initial intent of supporting large-scale, very dense, high-performance computing systems from day one. They also need to be able to scale at the speed hyperscalers and AI customers are demanding," Richard says.

Cooling as foundational infrastructure

The evolution of AI infrastructure places cooling at the nexus of data centre design rather than as a secondary consideration. 

Richard draws a direct line between cooling capability and operational viability. "You can have the most advanced computing systems ever developed, but if you can't power and cool them reliably and effectively, they're basically rendered useless," he says.

Silicon manufacturers including NVIDIA and AMD now pay close attention to power and cooling infrastructure as AI platforms drive rack densities higher and create dynamic power profiles. According to Richard, this attention reflects the reality that cooling must function correctly from initial deployment.

Air-cooled data centres offered flexibility, allowing operators to move cold air around, mix different compute platforms and work within wide margins of error. Liquid-cooled infrastructure eliminates those margins, he adds.

"With how intense these heat loads are, if the cooling isn't right, the data centre simply won't work,” explains Richard. “You turn everything on, and if the power and cooling aren't there, the servers won't operate. It becomes very black and white – basically it's either right or it's wrong.”

Cooling decisions now dictate deployment options, scaling capacity and whether facilities can support AI workloads. The design stage becomes the point where cooling considerations take precedence, shaping subsequent infrastructure decisions.

Liquid cooling tech: analysing the risks

Liquid cooling technology itself has existed for decades. IBM implemented liquid-cooled systems in the 1980s, and Motivair by Schneider Electric has been cooling three of the most advanced liquid-cooled platforms for more than a decade. 

Richard identifies the real risk not as the technology itself, but as underestimating its complexity. 

"Liquid cooling is a highly scientific, deeply engineered discipline and it's not something you can adopt without experience,” he says. “These systems take years of development and close collaboration with silicon manufacturers and server OEMs. If that work hasn't been done properly, risks begin to surface.”

Those risks include fluid contamination, performance degradation and poorly designed cooling architectures. Any reduction in cooling loop performance directly impacts server performance. The Coolant Distribution Unit serves as the heart of the system, as its absence means no cooling for servers. The Technology Cooling System loop, which comprises the network of piping and connections between the CDU and the racks, governs how thermal energy moves in and out of the compute environment.

"If these elements aren't designed, manufactured and deployed correctly you introduce risk into an environment where there is zero room for failure," Richard says.

But what are the rewards?

Properly implemented liquid cooling delivers efficiency advantages based on the physics of heat transfer, Richard notes. 

Liquid removes heat thousands of times more efficiently than air, expanding performance and efficiency possibilities. 

The technology enables higher rack densities while improving data centre operations by removing heat directly at chip level, allowing cooling systems from rack to plant to run more efficiently and predictably.

System architecture presents another opportunity. The CDU combined with a well-designed TCS loop provides a structured and scalable method for moving thermal energy between the compute environment and facility infrastructure. 

When designed correctly, operators can deploy capacity and scale as AI densities rise without reworking core infrastructure.

"There is also longer-term potential around reducing or even eliminating chillers, although that remains a utopian ambition rather than an immediate reality. For now liquid cooling is the only way to efficiently, reliably, and sustainably run the chip and server technologies required for AI," Richard says.

He identifies the elimination of risk at scale through proven, repeatable liquid cooling solutions aligned with AI infrastructure development as the primary opportunity.

AI server requirements mandate cooling technology adoption

Data centre operators no longer face a choice about liquid cooling adoption for AI deployments. "If you want to deploy advanced AI systems and leverage the power of AI in your business, liquid cooling is the only way those servers are being delivered," Richard says.

This reality represents a tipping point for the industry. Current and future silicon roadmaps design chips with liquid cooling as the primary thermal solution. Direct-to-chip liquid cooling will remain the dominant method at least through 2030, Richard insists, requiring organisations competing in AI to align infrastructure strategies accordingly.

The necessity of adopting liquid cooling for AI brings efficiency benefits. Liquid cooling enables higher efficiency outside the data centre through higher water temperatures, reduced chiller reliance or more efficient heat rejection. 

While AI requirements drive the adoption of liquid cooling, efficiency, sustainability and performance benefits accompany the transition.

No pressure: navigating scale and execution for trillion-dollar AI investments

Looking at the next 12 to 18 months, Richard points to the scale, speed and capital intensity of infrastructure development. With over a trillion dollars of investment flowing into AI infrastructure, he argues that there is no room for error.

“Power and cooling decisions made today will determine whether these projects succeed or stall,” says Richard.

Operators must consider proven, end-to-end solutions delivered at global scale rather than evaluating individual technologies. Partner selection becomes equally important for consistent execution across geographies as timelines compress and expectations rise.

"Operators need to get it right and think beyond individual technologies to proven, end-to-end solutions delivered at global scale. Just as important is choosing partners who can execute consistently, across geographies, as timelines compress and expectations continue to rise," Richard says.