Johannesburg, 10 Apr 2025
The complexity of the infrastructure required to support global artificial intelligence (AI) ambitions was highlighted at last week’s Vertiv AI Innovation Roadshow, in Sandton, where data centre power and cooling expert Vertiv outlined the key considerations for next-generation data centre infrastructure.
The roadshow, one of a series for CIOs and data centre managers across Africa, highlighted key emerging trends and technologies for data centres, with AI being the biggest game-changer.
Wojtek Piorko, managing director for Africa at Vertiv, said: “AI is transforming our IT, our businesses and our lives. Bloomberg predicts that generative AI alone will be a $1.3 trillion business by 2032, and over 34% of organisations globally are actively adopting AI.”
In Africa, according to the GMSA, agriculture accounts for around 49% of AI adoption, followed by the climate action (26%) and energy (24%) sectors, with most of this adoption relating to generative AI.
GPUs change data centre architecture
Jon Abbott, EMEA technologies director, colocation and hyperscale at Vertiv, outlined how GPUs, at the heart of the AI revolution, have to be clustered together.
“You're unable to distribute GPU processors – they have to sit close to each other. This means it’s a lot more power dense. This is a sea change in terms of critical infrastructure,” he said.
Abbott highlighted five imperatives to prepare critical infrastructure for AI workloads:
- Be transformative: “It needs an ecosystem to make this work. It must have a holistic design,” Abbott said.
- Be efficient: “You have to streamline processes, eliminate unnecessary costs and minimise the space required,” he advised.
- Be first: Abbott said data centres should increase their speed to market and accelerate innovation to gain a competitive advantage.
- Be confident: He said organisations should rely on their AI systems to perform as needed.
- Be future ready: “Expand your use cases and prepare for continued evolution,” he said.
“Cooling, power and deployment are the common thread running through all of these,” Abbott said.
AI demands next-generation cooling
AI is driving advances in liquid cooling technology, and bringing new complexity to managing data centres.
Jonathan Duncan, application engineering and technical solutions director – Africa at Vertiv, said: “In the AI era, air cooling won’t be enough. You have to prepare for liquid cooling.” However, he noted that data centres would likely run hybrid air and liquid cooling infrastructure for the foreseeable future.
He clarified that with advanced liquid cooling, cooling distribution units (CDUs) are the ‘beating heart’ of the cooling system, and have to be powered continuously, so systems must be in place to enable reliable power supply. Another potential challenge is that coolant flows through channels as narrow as 27 microns, with the risk of flow being fouled or obstructed.
“Fluid creates considerable complexity,” Duncan said.
He highlighted the ‘Five Fs’ that are critical for liquid cooling systems: feed-in temperatures, faults, fluids, filtration and flow rate.
AI changes the game in data centre power
Haroun De Almeida, AC power technical solutions manager at Vertiv, noted that the GPUs needed for AI compute impact the entire data centre power train. “With high density compute, the picture changes completely. You may have a requirement of close to 1MW in just one aisle. So instead of having single energy centres, we now need multiple power modules to cater for continuous cooling, with UPSes for both mechanical and IT loads." In addition, most data centres would have to cater for both traditional compute and AI compute.
On the issue of grid constraints and sustainability issues, De Almeida noted that data centres have the opportunity to adopt the ‘bring your own power’ approach, integrating renewables to supplement their power requirements.
Vertiv’s comprehensive AI-ready portfolio powers and cools even the highest-density workloads in the most challenging deployments. Vertiv 360AI solutions include a complete portfolio of power, cooling and service solutions for AI. Because it is crucial to have a robust and continuous power availability for the GPUs and CPUs that run AI compute, Vertiv recently launched Vertiv Trinergy, an uninterruptible power supply (UPS) engineered to handle the fluctuating load demands of data centres, and Vertiv PowerNexus, an integrated solution that provides space-saving close coupling of the UPS and system switchgear. Vertiv also recently announced the launch of Vertiv Liquid Cooling Services, providing the tools and expertise to enhance system availability, improve efficiency and navigate the challenges of managing advanced liquid cooling systems.
Share