The rapid rise of AI workloads is forcing data centre operators to rethink the way they design, power, and cool their facilities, industry experts told a webinar hosted by Automation News this week.
“Just a few years ago, five-megawatt data centres were the norm,” said Danel Turk, Data Centres Global Portfolio Leader at ABB addressing the free webinar on 10 February.
“Now we’re talking about 200-megawatt facilities, and discussions of gigawatt-scale deployments are happening. The challenges of scaling at that speed are enormous. You need to build faster, standardise your modules, and anticipate future requirements.”
Pradeep Shenoy, Compute Power & Controls Technologist at Texas Instruments, highlighted the strain that AI workloads place on energy infrastructure. “Historically, power supplies converted AC to DC in small stages, maybe a few kilowatts at a time,” he said. “Today, AI workloads are orders of magnitude higher. We’re moving towards 400–800 volt architectures, and any energy lost along the way is energy not used for compute. Efficiency and reliability are now critical considerations in system design.”
Cooling presents another major challenge. “Air-based cooling was sufficient when processors drew hundreds of watts,” Shenoy said. “Now, racks can draw megawatts, with multiple high-density AI processors. Liquid cooling is no longer optional; it is essential to remove heat where it is generated.” Turk added: “It’s not just the racks. UPS and rectifier systems also need liquid cooling. At these power levels, air simply won’t do. We are looking at higher temperature liquids, integrated heat reuse, and even district heating in Europe to improve efficiency.”
Modular design is helping operators deploy AI infrastructure quickly while maintaining reliability. “We’re building in three-megawatt blocks,” Turk said. “You can scale up in increments, replace components easily, and maintain uptime. Standardisation across sites allows teams to train consistently and optimise operations using AI.” Shenoy stressed the importance of secure supply chains. “Operators need assurance that components will be available at scale. At TI, we own a large portion of our manufacturing, providing reliability and flexibility for rapid deployment.”
Automation and AI are increasingly deployed to manage complex data centre operations. “The first step is high-fidelity telemetry,” Shenoy said. “You need to monitor voltages, currents, and temperatures continuously. With this data, AI can optimise energy use, predict faults, and schedule maintenance before problems arise.” Turk added: “Even if you don’t yet know how you’ll use the data, install sensors. Collecting it early allows predictive maintenance, energy optimisation, and system-wide coordination.”
Emerging technologies are further improving efficiency. “Wide bandgap semiconductors, like gallium nitride and silicon carbide, enable smaller, faster, and more efficient power converters,” Shenoy said. Turk pointed out gains in mechanical systems as well: “More efficient motors for cooling can reduce energy consumption by 30–40%, a far greater impact than changing a single material in memory chips.”
Experts also warned that regulatory, cybersecurity, and grid integration challenges must keep pace with AI demand. Turk noted that the European Union is investing in gigawatt-scale facilities and cross-border transmission, while Shenoy emphasised grid stability: “Sudden AI workload ramp-ups can create hundreds of megawatts of transient load. The grid cannot tolerate that without buffering and energy storage.”
Looking ahead, both experts stressed collaboration across the industry. “The pace of AI innovation is unprecedented,” Dirk said. “You need modular, future-proof infrastructure, and you need to work with partners who understand the full system.” Shenoy added: “AI is not just the problem—it’s also part of the solution. By embedding intelligence into power, cooling, and monitoring systems, we can make data centres more resilient, sustainable, and efficient.”
As AI workloads continue to grow, it is clear that the next generation of data centres will look very different from the legacy facilities of the past. “It’s a complex challenge,” Turk said, “but with smart design, modularity, and automation, we can scale from megawatts to gigawatts without compromising reliability or efficiency.”
To see a recording of the free webinar click on the link below.
