AI has moved rapidly from experimentation to core infrastructure, bringing with it new demands on data centres worldwide. Since the launch of ChatGPT in November 2022, the AI conversation has accelerated dramatically – evolving from early curiosity to a strategic priority for organisations across industries.

Speaking at a recent joint event hosted by Vertiv, a global leader in critical digital infrastructure solutions, and its customer Open Access Data Centres (OADC), trend analyst and Flux Trends founder Dion Chang described the pace of change.

“In just three years, the landscape has evolved dramatically. At first, generative AI was all about writing e-mails, summarising documents and handling tedious tasks. Then last year, we started seeing agentic AI, with systems that don’t just follow instructions but begin taking stewardship over tasks. It’s exciting, but it’s also challenging to navigate,” Chang said.

Growing importance of available power for data centres

Every era has its defining structures, Chang noted: the 19th century had railway stations and the 20th had skyscrapers – the 21st, in the context of the AI explosion, may be defined by data centres.

Wojtek Piorko, managing director for Africa at Vertiv, explained why, pointing to the transformative impact that graphics processing units (GPUs) have had on data centres.

“GPUs have evolved from graphics processors into high-performance parallel engines. To give a simple analogy: think of a CPU as a single-lane road where cars move one after another, very efficiently but sequentially. A GPU, on the other hand, is like a multi-lane highway, where thousands of cars can travel in parallel at the same time. That ability to process thousands of simultaneous calculations is what makes them so critical for AI.

“The result is a step change in density. You can now consolidate what once required multiple racks into a single rack. While that improves space efficiency, it fundamentally changes the demands on power and cooling, and ultimately how data centres are designed.”

With the advent of AI, rack densities are moving from 30-40kW to potentially hundreds of kilowatts per rack in the next few years, making the supporting infrastructure – power, cooling and fibre – more critical than ever.

“AI deployments raise important questions,” Piorko said. “The first is power availability. AI data centres require hundreds of megawatts of additional capacity. Locally, we need to find where that power will come from and how it will interact with the grid.

“Across Africa, power generation has roughly doubled over the past decade, but is that enough? Probably not. We do have significant natural resources, including hydropower, but the challenge currently lies with power distribution – something many of us in South Africa know particularly well.”

Africa in focus: connectivity, edge and flexibility

Piorko added context on the African data centre landscape: “Africa currently has less than half a gigawatt of active white space capacity for a population of over a billion. To put this into perspective, Dubai alone has the same capacity as our entire continent. This shows that, while the opportunity in Africa is enormous, we need factors like stable power, regulatory alignment, connectivity and skills in place to catch up.”

Marc Matthews, engineering director and head of projects at Open Access Data Centres, said it is possible to meet these requirements in Africa, pointing to OADC’s own recent achievements on the continent.

“Our business model is slightly different, with OADC’s site in Amanzimtoti, South Africa, functioning primarily as a cable landing station. This highlights a key point: data centres are nothing without connectivity. You can have huge computing capacity, but without access to subsea cables and global networks, the data cannot reach users. That’s why our facilities across Africa are strategically located near subsea cable landing points and terrestrial fibre networks.”

One of the biggest shifts in the local industry is the move towards edge computing, Matthews said. “If someone in Johannesburg asks an AI system for restaurant recommendations, they shouldn’t have to wait minutes for a response from a data centre on another continent. AI inference needs to happen closer to the user, and this is why we’re seeing the rise of edge data centres across Africa.”

Environmental consciousness need not be just a buzzword, he added. “In Nigeria, OADC is exploring gas-based power generation as an alternative to diesel, while in South Africa we have solar panels that offset energy use for one data hall during daylight hours. Operational tweaks, like adjusting cooling setpoints slightly, can also yield significant energy savings.”

Water concerns in Africa are often misunderstood, Matthews said, since modern cooling systems operate in closed loops, meaning the same water is reused repeatedly.

Piorko agreed, referring to the recently launched Vertiv SmartRun, a converged overhead IT infrastructure system that integrates high-density power distribution, closed-loop liquid cooling, networking and containment infrastructure in an all-in-one platform. In this design, all infrastructure above the rack is pre-designed and modular.

“This allows much faster and more efficient deployment,” Piorko said. “In fact, our broader philosophy is modularity. Components are pre-designed, manufactured and tested in factories before being shipped to site. That reduces risk, speeds up deployment and minimises construction waste.”

This type of modularity has played a key role in the design of OADC’s facility in Lagos, Nigeria, which Matthews said reflects a flexible, phased approach to infrastructure deployment. “Initially planned as a 24MW facility, we opted to start with 6MW, scalable to 24MW later. This decision is based on the concept of designing facilities that can support both today’s workloads and tomorrow’s AI demands.”

Looking ahead

Together, these approaches point to a more scalable and resilient model for data centre development in Africa – one that balances current demand with the evolution AI will place on infrastructure in the years ahead.

To find out more about Vertiv’s AI offering, visit the Vertiv AI Hub.