Liquid Cooling Won’t Kill Air Cooling: Stulz on the Hybrid Reality of AI Heat
28 Oct 2026 · Stories from Data Centre World Asia STULZ’s Mark Langford on cooling that moved from the room to the chip, why the real job is integrating air and liquid, and why a legacy hall isn’t...
28 Oct 2026 · Stories from Data Centre World Asia
Quick Skim
STULZ’s Mark Langford on cooling that moved from the room to the chip, why the real job is integrating air and liquid, and why a legacy hall isn’t obsolete.
Not long ago, cooling a data centre meant cooling a room. Then it meant cooling the rack. Now, increasingly, it means cooling components inside the server itself.
That migration, room to rack to chip, is how Mark Langford, Regional Technical Director for Asia at STULZ, sums up what AI and high-performance computing have done to his field. “Speed and scale,” is his two-word answer for what changed. The heat to be managed has jumped, and the pace of change around it has jumped with it.
From the Room to the Chip
The kit that follows the heat has changed with it. Liquid cooling, coolant distribution units, secondary fluid circuits and hybrid air-and-liquid systems are now, in Mark’s words, becoming part of mainstream data-centre design for AI and HPC applications, not the exotic edge case they were a couple of years ago.
What has changed alongside the hardware is what customers actually want to buy. The market has moved away from equipment selection, Mark says, towards “fully integrated infrastructure solutions, flexible and adaptable for the next generation of compute without redesigning their entire facility.” That shift, he notes, creates new opportunities and new headaches for vendors in equal measure.
The Bottleneck is the Pace, not a Part
Asked what has the longest lead time, Mark doesn’t name a component. “I wouldn’t say there is any one thing,” he says, the constraint is the speed and scale themselves. Coordinating and aligning resources for delivery, installation and commissioning against ever more aggressive completion dates is where the real localised crunches appear. On top of that, the technology cycle is moving fast enough that designers and constructors are struggling to keep pace with the market. The choke point, in other words, is less a transformer in a queue than a schedule that assumes everything lands at once.
Air and Liquid, not Air or Liquid
Here is the misconception Mark is keenest to correct: that liquid cooling means air cooling disappears. It doesn’t. “HPC servers still have a residual air-cooling requirement,” he says, “so for the foreseeable future the real engineering challenge is integrating air and liquid into one resilient cooling architecture.” The reframe he offers is sharper still: “We’re not moving from air cooling to liquid cooling. We’re moving from cooling products to prefabricated and integrated thermal architectures”, what STULZ calls cooling from “rack to roof.”
That same logic answers the retrofit question. A hall that wasn’t built for AI isn’t obsolete. Plenty of workloads don’t need high-performance GPU or CPU, Mark points out, and where HPC is coming, the move can be managed as a phased hybrid architecture rather than a rebuild. “Hybridisation isn’t about keeping an old data centre alive,” he says. “It’s about making existing infrastructure relevant to the next generation of compute.”
Not Just a Cooling Company
The other thing people get wrong, Mark says, is treating STULZ as simply a cooling company. The role has shifted from equipment vendor to technology partner, and early engagement is now a necessity rather than a nicety: thermal choices increasingly shape the performance, efficiency and flexibility of the whole facility, so they can’t be made in isolation at the end of a design.
Nor is precision cooling a data-centre-only business. STULZ has spent more than 75 years on mission-critical temperature and humidity control across telecommunications, laboratories, healthcare, education, defence and industrial sites, anywhere comfort cooling isn’t good enough and reliability is the point. It operates in more than 140 countries, and from an APAC vantage Mark sees the application shifting fastest from conventional air-cooled rooms to hybrid and liquid-cooled architectures for AI and HPC. On waste, he frames efficiency less as a feature than an obligation: modern infrastructure paired with pervasive, proactive monitoring lets capacity track demand continuously, so a data centre consumes only what it needs, and, in his words, acts as a “good corporate citizen.”
The Gemba Takeaway
Bring the thermal partner into the room early, not at the end. Cooling choices now set the performance and flexibility of the whole facility. Plan for air and liquid together rather than betting on one, because HPC still needs both. And before writing off a legacy hall, scope a phased hybrid path: making existing infrastructure relevant is usually faster and cheaper than starting over.
This interview is part of Industrial Gemba’s coverage from Data Centre World Asia. Responses have been edited for length and clarity. Watch the full interview on YouTube.

