Liquid Cooling Gains Ground in AI Data Centers, But One Size Doesn't Fit All

Data Centers Heat Up: Why AI Chips Are Driving a Cooling Overhaul

As artificial intelligence workloads push data centers to install ever more powerful graphics processing units and central processing units, a hard thermal limit is emerging. The newest generation of chips generates significantly more heat than the hardware it replaces—heat that conventional air-based cooling cannot remove quickly enough. When fans and air handlers fall short, chip performance throttles, and the efficiency gains from upgrading evaporate.

The answer, many in the industry now agree, is liquid cooling. By circulating a coolant directly past heat sources, liquid systems can carry away thermal energy far more efficiently than air. But adoption has been anything but automatic. A mix of misunderstanding about water consumption and the complexity of retrofitting existing facilities has kept some operators on the sidelines—even as chip temperatures continue to climb.

One company trying to smooth the path is Nortek Data Center Cooling, which recently highlighted its “C-Force” team—a unit built to deliver custom and semi-custom cooling architectures rather than off-the-shelf hardware. The message: data centers running different chip configurations need different flow rates, temperatures, and footprints, making a tailored approach essential.

Beyond the Hype: Misconceptions, Hybrid Systems, and the Customization Imperative

Closed-Loop Reality: Liquid Cooling Often Uses Less Water

A persistent myth is that liquid cooling systems gulp water, undermining data center sustainability goals. In fact, the closed-loop designs now being deployed recirculate the same fluid repeatedly. Heat transfers from the chip to a coolant distribution unit, which then rejects it to a secondary loop, often also closed. This dramatically reduces net water intake compared to once-through evaporative cooling that some air-cooled setups rely on. For operators under pressure to cut water usage, liquid cooling can actually be a water-saving move—if designed correctly.

Hybrid at Heart: Why Air Systems Won't Disappear

Despite the liquid push, no one is throwing out air handlers. In a typical modern rack, liquid cooling targets the largest heat sources—the GPUs and CPUs—while fans still manage the remaining electronics. This hybrid model lets facilities balance capital cost and cooling performance. The challenge, however, is that the ratio of liquid to air cooling varies by rack and by workload, demanding tight integration from the start.

The Customisation Imperative

Because flow rates and inlet/outlet temperatures depend on the specific chips installed, a one-size-fits-all cooling module is impractical. Operators need suppliers who can engineer around the unique thermal profile of each deployment—from prototyping in a thermal lab through to commissioning on site. That level of hand-holding adds time and cost but is fast becoming a requirement for hyperscale and colocation providers who cannot afford cooling-related throttling.

Verification note: This analysis draws on information provided by Nortek DCC, a commercial player in the space, and reflects the company’s perspective. The broader trend toward liquid cooling is supported by independent industry data, but specific claims about water savings and performance are based on the source’s representations.

What This Means for Data Center Operators and Equipment Buyers

• Audit your chip roadmap. If upcoming GPU and CPU purchases will push rack density beyond 20–30 kW, start modeling liquid cooling requirements now. Involve facilities teams alongside procurement.
• Question water claims. Insist that vendors provide water-usage figures for both open- and closed-loop modes, and compare them against your current evaporative cooling consumption.
• Demand a physical prototype. Before signing a large order, arrange for a thermal lab test of the exact configuration you will run—flow rates, temperatures, and failover scenarios—to avoid performance gaps after installation.
• Plan for hybrid operation. Evaluate whether your air-handling system can be scaled back once liquid loops are installed; over-cooling is wasteful, but under-cooling risks equipment degradation.