Data center cooling ain’t what it used to be. Not only is liquid cooling finding its way into more and more facilities, but the fundamentals of airflow management have never been more important.

With so much money being spent on GPUs, high-density servers, liquid cooling systems, and supporting infrastructure, even small amounts of airflow inefficiency can:

1. Add millions annually in energy costs
2. Become a major contributor to expensive data center outages

The AFCOM State of the Data Center Report 2026 notes that cooling has become one of the most immediate constraints in data center operations. AI and high-performance workloads are placing mounting pressure on existing cooling architectures. While 39% of survey respondents said current cooling solutions do not meet all their operational requirements, 36% have deployed liquid cooling and another 28% plan to adopt it within the next 12 to 24 months. Some believe that rack densities exceeding 20 kW signal the point where liquid-based thermal management should be considered. Others believe air-based cooling can scale as high as 80 kW. Time will tell who is right.

But the shift toward liquid cooling now appears inevitable. According to the report, more than one in four racks use liquid cooling. Additionally, 37% said they had adopted or were considering rear door heat exchangers (RDHX). For now, RDHX appears to be the dominant form of liquid cooling because it requires less infrastructural upheaval. Cold plates for specific servers are next in adoption.

“Cooling is becoming a primary design driver that will define how the next generation of data centers are built, scaled, and operated,” said Kleyman.

Liquid Baby Steps

Massive hyperscale data centers, large colocation providers with deep pockets, and government laboratories are expected to drive advanced liquid cooling adoption and innovation over the next year or two.

But most data centers are taking a more cautious approach. Many are testing the waters with RDHX and limited deployments of cold plates. A few are moving ahead with end-to-end data center reconfigurations involving major plumbing and electrical changes. However, don’t expect that to become the norm anytime soon. Economic realities still matter.

It costs millions to install GPU-dense racks. Even deploying a few rows is beyond the means of many data centers. Anyone submitting a PO for investments of that size had better be ready to demonstrate strong gains in data center efficiency, PUE, and airflow management. It would be extremely wasteful to install millions of dollars of state-of-the-art computing infrastructure in an inefficient data center environment.

Modularization of Cooling and Power

A major takeaway from the AFCOM report is that facilities are becoming significantly larger, construction volume is accelerating alongside demand, and hybrid and prefabricated models are becoming core strategies. This means:

• Hyperscalers will continue building massive AI factories supported by extensive liquid cooling infrastructure
• Prefabricated power and cooling modules for both small and large data centers are gaining momentum

Some vendors are introducing large data center “chunks” to simplify the deployment of power and cooling infrastructure in increasingly large facilities. Initial modules are aimed at larger operators, but the broader trend is moving away from small, server- and rack-based cooling equipment toward plug-and-play units that integrate plumbing, electrical, cooling, and other supporting systems.

“While traditional construction remains common, hybrid approaches that combine conventional builds with prefabricated components have emerged as the most widely planned strategy,” said Kleyman. “Prefabrication efforts are focused on power and cooling modules, reflecting where standardization and off-site assembly provide the greatest time and risk reduction.”

According to the AFCOM survey, data center managers are planning to deploy power modules (77%), followed by cooling modules (69%). However, IT, physical security, and auxiliary modules are also in demand.

All of these trends are expected to accelerate as data centers continue scaling upward. The more GPUs packed into facilities, the more liquid cooling will be required. Each year, liquid cooling adoption continues to rise in lockstep with increasing rack density. To support higher densities, organizations are adding more liquid cooling while also finding ways to improve airflow within existing facilities.