What is liquid cooling, and why does it matter for AI data centres?

As the data centre industry continues to undergo rapid digital change, liquid cooling is no longer an alternative technology but is now marking a fundamental shift in planning and operation.

Liquid cooling has become more essential for data centres as companies invest more in AI. The hardware for these workloads (ie, GPUs) generates significantly more heat, which means that traditional air-cooling systems are unable to cool the technology fast enough.

With liquid cooling, data centre operators can deliver the greater performance and density that modern AI infrastructure requires.

What is liquid cooling?

Liquid cooling is a heat management technology that removes heat from computer components by circulating a specialised coolant through or around the hardware that generates heat.

The coolant, which is normally water or a water-based solution, absorbs the heat and carries it away to a heat exchanger or cooling tower and then dissipates it. This leaves the hardware cooler than if it were relying on air cooling alone.

There are several key types of liquid cooling:

Direct liquid cooling (DLC) – cold plates that sit on individual chips, which has quickly become the most common industry approach
Immersion cooling – servers are submerged in tanks of non-conductive dielectric fluid
Rear-door heat exchangers – a liquid-cooled door on the back of a server rack that captures heat as air exits

In comparison to air cooling, liquid cooling has been touted as being more effective at removing heat from components than air cooling. For one thing, it doesn’t rely on air flow, so it’s not affected by dust or any other obstructions.

Likewise, it’s less likely to cause noise pollution, as the fans are typically quieter than those used for air cooling.

Why has it become so popular in the data centre industry?

AI workloads are the key driver pushing liquid cooling further into the mainstream, particularly as modern AI accelerators like Nvidia and Google chips have significantly higher power density. Chips are able to run cooler and more consistently with liquid cooling, which extends their lifespan.

In simple terms, AI chips produce more heat than air can carry away. A single rack of AI servers can now surpass 100 kW, meaning that air cooling is unable to remove heat fast enough. Liquid cooling can help data centres run more efficiently and improve their Power Usage Effectiveness (PUE), cutting cooling-related energy significantly.

Another aspect of liquid cooling that the data centre industry struggles to publicise is, although an oxymoron, it helps data centres to reduce water consumption. It’s no secret that AI is continually thirsty and increases energy and carbon emissions for data centres. However, while public perception is that these power-hungry facilities are guzzling gallons of water, specialised closed-loop cooling systems can help replace evaporative cooling towers.

For hyperscalers that have strong net zero targets, liquid cooling is perhaps one of the most impactful solutions for cutting the carbon footprint of AI infrastructure.

Who are the industry leaders?

In response to AI and heat-related challenges across the data centre industry, cooling companies like Vertiv, Schneider Electric, Johnson Controls, Ecolab and LiquidStack/Trane Technologies have emerged with significant solutions to cool facilities down.

Johnson Controls in particular has recently tapped more into the cooling space, finding new ways to support its customers across the data centre industry. It did this through its acquisition of Alloy Enterprises, a thermal, mechanical and materials sciences technology innovator.

Speaking with Capacity recently, Mihir Nandkeolyar, director business development global data centre solutions at Johnson Controls, shared how cooling technologies will usher in the next era of data centre growth.

“We incorporated the latest refrigerants to achieve higher temperatures and decarbonise more buildings,” Nandkeolyar explained. “Pairing that technology with data centres to support community decarbonisation falls perfectly within both our strategy and our capabilities.”

Schneider Electric also demonstrated how organisations can deploy AI-ready infrastructure responsibly with liquid cooling at Datacloud Global Congress. The newc comes after Motivair by Schneider Electric announced two new CDUs at the end of last year to deliver scalability and efficiency for high-performance computing (HPC), AI Factory and data centre environments.

Are there any risks?

Although a popular solution, liquid cooling inevitably has some risks associated with it. These are primarily around maintenance, as custom liquid cooling systems require regular upkeep to avoid points of failure. They are also associated with higher costs.

However, liquid cooling failures and leaks are not very common, and coolers are designed specifically to be durable and are tested for leaks.

What’s the long-term reward?

Major data centres are now designing new facilities specifically with liquid cooling in mind, or retrofitting older ones to suit the technology. According to Global Market Insights, the liquid cooling market for data centres is expected to expand from US$6 billion in 2026 to $27.1 billion by 2035.

For those eager to engage with the technology, the promise is greater efficiency, faster innovation and more net zero-positive outcomes.

“While there is no one-size-fits-all solution, liquid cooling, particularly two-phase systems, is emerging as a promising option for high-density AI workloads in certain scenarios, depending on system requirements,” said Fabian Ipfelkofer, EMEA business development manager for data centre applications at global specialist materials firm Solstice Advanced Materials, earlier this year.

“Cooling approaches are set to play an increasingly strategic role, evolving to support greater efficiency and more sustainable power use.”

How Veolia plans to power sustainable AI data centres

Beyond FLAP-D: Why Europe’s next wave markets are moving centre stage