The rapid increase in artificial intelligence workloads is changing data center design and construction, creating new challenges and opportunities for mechanical, electrical, and plumbing professionals. As these demands grow, experts from across the industry provide critical insight into what’s changing and how MEP professionals must adapt.
Michael Obradovich, Ecolab’s vice president of global accounts, explained that traditional air cooling systems can no longer keep up with the heat generated by high-performance AI servers. “We’re getting to the point where just moving air over the servers doesn’t provide enough heat exchange to prevent overheating,” he said. This has led to a significant shift to liquid cooling, which transfers heat nearly 24 times more effectively than air, making it essential for cooling next-generation AI hardware.
This rapid increase in heat output is intertwined with the progression of Moore’s law. As chips become denser and more powerful, new generations of computing hardware run even hotter than before. Traditional approaches of moving more air through larger ducts cannot address this growing thermal challenge.
Travis Schumacher, senior project manager at DPR Construction, said the industry is increasingly moving away from the large-scale duct work traditionally used for air cooling. “About 80% of the cooling load in new data centers is currently water-cooled, and only about 20% relies on air,” he said. This change requires more complex piping systems and increases the use of materials such as stainless steel for durability and compatibility with liquid cooling.
What HVAC contractors need to know
Schumacher emphasized that HVAC design has changed significantly with the adoption of AI workloads and liquid cooling. Traditional ducts that flood entire rooms are being replaced by more targeted airflow systems integrated into mechanical components. “We are starting to see a shift away from heavy-duty duct work,” he explained.
He also pointed out that server fans in racks are now part of the HVAC airflow system. “The fans inside are more efficient, which means fewer fans are needed.” [HVAC airflow]. Instead of us pushing out cold air, they are pulling in cold air. That was the case 15 years ago. We use these server fans as part of our HVAC engineering system and factor that into our design. ”
HVAC contractors must adapt by incorporating server fan airflow into ventilation calculations, designing complex liquid cooling piping, and preparing commissioning protocols that differ significantly from traditional systems.
What Sheet Metal Contractors Need to Know
Sheet metal fabrication is evolving in parallel with changes in HVAC. Schumacher said sheet metal teams are increasingly involved in prefabricated insulated metal panels and modular mechanical components rather than large-scale duct work. “We started leveraging their resources in different areas, whether it was insulated metal panels or actually putting the skin on top of that, creating different shafts and other things so that the building would act as a mechanical equipment itself rather than utilizing air handlers.”
Louvers and air intake and exhaust systems remain important, especially in facilities facing environmental challenges. Schumacher recalled a project in Oregon that had large louvers and dedicated passageways designed to keep sand and dust out of the air handler intake. The components required special sheet metal fabrication and installation.
Sheet metal contractors should expect to work closely with prefabricated multi-trade racks and modular assemblies where ductwork equipment may be integrated with piping, fire protection, and safety conduits.
What every MEP professional should prepare for
Prefabrication and modular construction are becoming central to data center construction strategies. “The multi-trade rack is actually one of the things I like to do the most. It creates a lot of simplicity. But one thing you definitely have to do is you have to build it into the design,” Schumacher said, highlighting the early adjustments. Prefabricated assemblies of electrical conduit, mechanical piping, fire sprinkler pipe, and ductwork are manufactured off-site and delivered ready to install, reducing on-site labor and schedule risk while also preventing trade-offs from trampling in a more controlled manufacturing environment.
Commissioning an AI data center is more complex. Schumacher explained, “Traditional commissioning methods are not effective for AI liquid cooling designs. So instead of 40 load banks on the floor, we could potentially have 200 to 400 load banks on the floor just to test different systems.” Early involvement of the commissioning agent and planning for dedicated test equipment is essential.
Change requests are increasingly driven by design revisions and supply chain constraints. Schumacher said, “Most of the change requests we see are design-driven or tenant-driven…Equipment replacements due to availability require redesign of ductwork and piping. These replacements may require rebuilding systems in the field.” Flexibility and proactive value engineering are essential to managing these challenges.
The choice between steel and concrete is increasingly influenced not only by performance but also by lead time. Prefabricated concrete can shorten schedules, but steel provides more usable floor space, so MEP teams must adapt to the chosen structural approach.
Water use remains a sensitive topic. Obradovic revealed that while many data centers use air-cooled chillers to minimize water demand, data centers that use water-cooled chillers often use alternative water sources such as recycled gray water or treated black water. He emphasized that data center water consumption accounts for a portion of global industrial water use, and judicious use of on-site water can reduce overall energy consumption.
A future-proof cooling infrastructure is essential. Obradovic recommended designing for flexibility to accommodate rapidly evolving workloads, including separate cooling loops with separate temperature setpoints for air and liquid cooling systems. Schumacher noted that modular construction allows for gradual growth, but cautioned that long-term costs and customization needs may favor a more conventional concrete structure for some customers.
Concludes insights from global technology
Chris Campbell, senior director at World Wide Technology, brought the conversation full circle by highlighting how fundamentally different the demands of AI infrastructure are compared to traditional data centers. He explained that modular data centers are in the spotlight due to the rapid evolution of computing power and cooling needs.
“The requirements for AI infrastructure are completely different,” Campbell says. “The amount of power delivered to the racks varies. Modular allows you to build all of these capabilities at a moment’s notice.”
He highlighted how modular data centers play a key role in enabling rapid deployment and meeting intense power and cooling demands that traditional facilities often cannot meet. By providing pre-engineered, integrated solutions, modular units can be installed anywhere power is available, whether in a parking lot or in a remote location, allowing organizations to scale quickly without the multi-year timelines of traditional construction. Campbell’s outlook is clear. Modular data centers are not a passing trend.
