How advanced technologies are helping AI Data Centres keep their cool

The rapid advancement of artificial intelligence (AI) is revolutionising the digital landscape, bringing with it unprecedented demands on data centre infrastructure. As AI applications become more sophisticated and widespread, the computing power required to support them is skyrocketing, presenting significant challenges for data centre cooling systems.

Data centre managers are faced with the challenge of how to accommodate increasingly power-hungry and heat-generating hardware, whilst ensuring they maintain an energy efficient data centre. This is not just an incremental change. The data centre cooling market is experiencing a paradigm shift that necessitates a fundamental change in data centre cooling solutions.

From air to liquid cooling systems - the change in cooling techniques

For decades, air cooling has been the predominant method for managing heat in data centres. This approach typically involves using computer room air conditioning units to circulate cold air throughout the facility, with the excess heat in the hot air exhausted through raised floors or ceiling plenums. Free cooling is an air cooling system innovation which takes advantage of cool external temperatures in the natural environments outside of the data centre to help cool the system.

While air cooling has proved to be effective for traditional IT loads in data centres, it faces significant limitations when dealing with modern data centres with high-power AI hardware and becomes increasingly inefficient at high densities.

A hot and cold aisle based cooling systems

Hot and cold aisle based cooling systems are unable to remove excess heat at the rate that an AI data centre facility requires. Hot air and cold air aisle containment is an advanced structural layout that improves this traditional data centre cooling system. It promotes a cyclical air flow, expelling the hot air in ventilation and pumping cool air down the cold aisles. The hot aisle containment ensures the cold and hot air do not mix, reducing the excess energy previously required to maintain temperatures.

The shift to a liquid cooling system

The only practical alternative is to use liquid cooling methods for this next-generation hardware. Liquid cooling technologies are much more energy efficient, can remove the heat more quickly and support greater equipment densities that modern data centres require. 

Technologies including immersion and direct liquid cooling are competing to address this challenge. As the name suggests, immersion cooling completely covers the equipment in non-conductive cooling liquid, while direct liquid brings coolant to the specific devices that are generating the heat.

Comparing immersion and direct-to-chip cooling systems

Immersion cooling has some unique benefits, it can handle high heat loads and since the IT devices are immersed it cools all components uniformly. It can eliminate the need for any air-cooling infrastructure, which has the additional benefit of reducing the noise usually generated from air circulating fans. However, with immersion cooling, initial setup costs can be high and in the case of upgrading existing data centres, floor loading issues may prevent its use. It has the additional disadvantage of making the IT components less accessible and harder to service, as they need to be removed from the liquid.

Direct-to-chip and data centre cooling efficiency

Direct-to-chip cooling delivers heat absorbing liquid directly to the devices via cold plates. While this method is highly efficient, it does require a complex network of pipes and pumps to circulate the coolant, which increases risk of failure due to leaks.

But unlike its immersion counterpart, it does permit ready access to components for servicing and it is more easily integrated into existing data centres. Direct-to-chip cooling has the disadvantage that it must be combined with air cooling because the technology only cools the components that are major heat sources.

Which liquid technology wins the race for next generation data centre cooling remains to be seen and it is likely to be heavily influenced by the major IT manufacturers, who will dictate the cooling interfaces for their equipment.

Sustainability benefits

Liquid cooling has distinct sustainability advantages. The water used always remains in a closed circuit, so once the system is filled, it doesn’t require a constant supply. This is a far more efficient cooling system in terms of water usage.

Data centre energy efficiency

With liquid cooling less energy is required overall, which reduces the CO2 footprint of data centres. In 2023, Data Centre Dynamics cited a study in which full implementation of direct to chip liquid cooling was compared with a 100% air cooled solution. The comparison revealed an 18.1% reduction in facility power and a 10.2% reduction in total data centre power.

Upgrading existing data centres

Data centre operators are facing significant challenges when upgrading existing facilities to liquid cooling. The original designs employed for lower power densities and air-cooled data halls are not easily retrofitted to accommodate the changes needed for the new technology. In addition, the scale of AI data centres is generally much larger than their conventional IT counterparts. These two factors combined mean that it makes more sense to build new capacity to meet the growing demand, as it is more efficient and cost-effective.

There is considerable change happening in the data centre industry. The concept of AI data centres is still relatively new and it may be some time before the optimal solution for liquid cooling in these facilities is established. The infrastructure changes are challenging, but they will be essential to support the AI computing power anticipated in the future.

How CBRE can help

The spectacular growth of cloud services and the explosion in AI technology means that liquid cooling is now an essential part of data centre infrastructure. However, upgrading and building data centres to cope with this new technology will take time. CBRE have exceptional data on where AI-ready data centre capacity is available, including liquid-cooling, power availability and networking.

CBRE are experts in finding and optimising AI data centre capacity.

Get in touch today to find out how we can help you with your IT infrastructure.