Why cooling is now a strategic design function for AI data centers

As AI infrastructure continues to scale, data center operators face a clear challenge: delivering more compute while working within increasingly constrained power, cooling and sustainability requirements.

During a recent Data Center Dynamics webinar, Davin Sandhu – Global Portfolio Director of Data Center Solutions at Johnson Controls – discussed how thermal management is evolving to meet the needs of AI Factories and why data centers are adopting integrated thermal management strategies.

The wide-ranging conversation touched on several key developments that are redefining data center thermal management, including:

• The growing need to view power and cooling as a single system
• The role of absorption chillers and the environmental, societal and operational benefits of recovering waste heat
• The impact that increasingly frequent GPU cycles and ever-evolving liquid cooling technology is having on data center design
• How leaning on reference design guides allows data center operators to consider the design opportunities and capabilities available to them
• How sustainability metrics are evolving beyond PUE to also include water usage and carbon usage effectiveness

Cooling is no longer just about removing heat

Traditional data center cooling strategies were primarily designed to remove heat from the data hall. Higher density AI workloads have fundamentally changed that approach. Today's operators face growing pressure to maximize the amount of available energy directed toward compute. As AI rack densities continue to increase, cooling decisions increasingly influence the overall productivity of a facility.

As Sandhu explained, "The name of the game is to maximize the amount of power that is either coming from the grid or being produced on site and trying to drive as much of it towards AI compute and the AI Factories. And the best way to do that is with a very efficient AI Factory design."

Integrated thermal management is becoming essential

One of the central themes of the webinar was the industry's transition to an integrated thermal management system.

Rather than optimizing cooling equipment individually, operators are increasingly evaluating the entire thermal chain – from heat capture at the chip level through heat transfer, heat rejection and potential waste heat reuse opportunities.

According to Sandhu, taking a holistic view enables operators to improve overall efficiency, maintain environmental conditions for high-performance compute and better prepare for future infrastructure requirements.

Why power and cooling must be designed together

Power constraints have emerged as one of the defining challenges facing AI infrastructure development. As a result, more operators are evaluating on-site power generation as part of their long-term strategy.

The webinar explored how this has created new opportunities for operators to optimize their thermal management system, with a great example being the approach to waste heat.

Rather than treating waste heat from on-site power generation as a byproduct, operators can potentially capture and reuse it through technologies such as absorption cooling. This integrated approach can help reduce cooling-related electrical demand while maximizing the energy available for AI workloads.

As Sandhu noted, "You're actually being a lot more efficient when you're considering power and cooling in the same integrated approach.”

Designing for future AI workloads

Another major topic was the pace of change occurring within AI hardware itself. Today's high-density AI environments are pushing cooling infrastructure into whole new territories. Future generations of compute are expected to demand even greater power densities and the pace of change will continue to accelerate.

Building flexibility into thermal infrastructure today can help organizations adapt to future requirements without major redesigns.

The webinar underscored a clear message: cooling is no longer a standalone infrastructure consideration. By taking an integrated approach to thermal management, operators can create more efficient, resilient and sustainable data centers capable of supporting the next generation of AI innovation.

FAQs

What is an integrated thermal management strategy for data centers?

An integrated thermal management strategy for data centers is a holistic, end-to-end approach that connects cooling infrastructure and power generation in one unified system. Harnessing waste heat is a core component of an integrated thermal management strategy.

What is waste heat recovery for data centers?

Waste heat recovery is an engineering approach where data centers capture excess heat from on-site power generation and re-use it as an asset. When natural gas is used to generate electricity, only 35-50% of the input fuel becomes usable power. The rest is lost as waste heat. Rather than rejecting this thermal energy into the atmosphere, operators can potentially utilize waste heat on secondary resources or even supply it to local communities.

What are absorption chillers and how do they work?

Absorption chillers are cooling systems that use waste heat. They can recover the waste heat from on-site generation and use it as their primary energy source instead of grid electricity.