Data centres are the foundation of our digital world, powering everything from cloud computing to AI. But this immense capability comes at a cost – substantial energy consumption, much of it tied to cooling systems.
As demand for computing grows, so too does the urgency to find more efficient, sustainable solutions, writes Tony Bartlett, director of data centre compute at Dell Technologies South Africa.
AI is shifting the energy narrative, driving smarter, leaner consumption
Data centres account for about 1% of the world’s total energy consumption, according to the International Energy Agency. With the rapid growth of AI, edge computing, and 5G networks, this figure is expected to rise. Cooling systems alone can consume 30% – 40% of a data centre’s energy, making them an essential area for improvement.
AI is changing this narrative by introducing intelligent, adaptable, and automated solutions that optimise conditions while minimising energy use. One major application is in energy efficiency optimisation. As AI workloads increase, data centres face rising energy demands and associated emissions.
Data centre power consumption in South Africa is experiencing significant growth, driven by increasing demand for cloud services, data consumption, and digital infrastructure expansion.
AI-driven systems also respond to external environmental conditions in real time. They adjust cooling strategies based on current weather, humidity, and temperature, maintaining efficiency while adapting to changing surroundings. This flexibility helps data centres meet sustainability targets without compromising computing performance.
Cooling solutions: From air to liquid and beyond
To meet evolving demands, data centres are embracing a mix of cooling solutions:
- Air cooling: Air cooling, a traditional method, uses fans, heat sinks, and air circulation to dissipate heat from crucial server components like processors and memory. Data centres have traditionally used air flow to cool servers; approximately 99% of data centres still use this approach in one form or another. Advanced techniques like hot/cold aisle configurations and intelligent fan zoning help optimise airflow and reduce energy use. It is cost effective, easy to maintain, and widely compatible with most standard servers and racks, making it a convenient option. However, it is less energy-efficient, struggles to scale with high-density AI workloads, and heavily relies on HVAC systems, which can increase electricity consumption and carbon emissions. While suitable for many setups, air cooling may fall short as computational demands rise, often prompting a shift to more advanced solutions.
- Liquid cooling: Direct liquid cooling (DLC) uses small heat exchangers known as cold plates to deliver liquid directly to heat-generating components, leveraging liquid’s superior thermal conductivity to efficiently dissipate heat. It supports higher-density server configurations, reduces noise by minimising fan use, and handles increased thermal loads with lower energy spikes. The South African data centre cooling market is estimated at $60-million in 2025, and is expected to reach $120-million by 2031, growing by a CAGR of 13,43% between 2025 and 2031. Large commercial data centres in South Africa typically consume 12 – 20MW of power. Just like their international counterparts, local facilities are sophisticated in their approach to PUE and using the optimal mix of energy sources.
- Hybrid cooling solutions with integrated rack scalable systems: Hybrid cooling combines elements of both air and liquid cooling systems. It involves direct-to-chip liquid cooling for high-priority components (like CPUs or GPUs) along with air cooling to manage ambient temperature and cool other components like memory. Even DLC-enabled racks of dense compute can generate intense air heat loads up to 80 kW per rack. Traditional air-cooling approaches limit rack density and create space and power challenges, even with direct liquid cooling on CPUs and GPUs. Standard industry rear door heat exchangers rely on cold water from a power-hungry refrigerated chiller system to extract heat from the air before blowing chilled air into data centres. Placing cold plates on additional server components can be done in a 100% DLC system, but that increases costs, dramatically reduces serviceability and configurability, and means that all the cooling system components must be replaced with each generation of new hardware.
AI and cooling: A smarter approach to energy efficiency
The demands on data centres are only growing, and so is the need for innovative cooling solutions.
By adopting AI-powered cooling strategies, South African businesses can reduce energy consumption, cut operational costs, and meet their sustainability goals without compromising performance.
Now is the time to embrace smarter, AI-enabled technologies that ensure data centres can power our digital world responsibly.