Africa’s artificial-intelligence plans will ultimately depend on a physical system of electricity, cooling, connectivity and resilient data centres. Without that foundation, announcements of new models and computing capacity will remain difficult to translate into working services.
Artificial intelligence may appear to exist in software, but every model depends on physical infrastructure.
Applications run on processors housed in data centres. Those processors require continuous electricity, advanced cooling, resilient connectivity and skilled teams capable of maintaining increasingly complex equipment.
For African countries seeking to develop local artificial-intelligence capabilities, the central question is therefore not only how many graphics processing units they can acquire. It is whether they can power, cool, connect and protect that equipment at a cost businesses can afford.
“No power and no cooling means no revenue,” Gary Chomse, Sales Director, Vertiv, told the “From AI Consumer to AI Producer: Building Africa’s Intelligence Infrastructure” panel at Hyperscalers Convergence Africa 2026. Moderated by Prof. Nkem Ihenachor, Professor of Strategic Management, Lagos Business School, the panel also featured Wole Abu, Managing Director, West Africa, Equinix; Kanwulia Okafor, Director, Industry Services, GSMA; Lars Johannisson, Chief Executive Officer, Rack Centre; and Guy Zibi, Chief Executive Officer, Xalam Analytics.
The discussion demonstrated why Africa’s AI strategy must also be an energy and infrastructure strategy.
AI is an energy-intensive industry
Training and running artificial-intelligence models requires large amounts of computing power. The electricity demand comes not only from processors, but also from storage, networking equipment, cooling, security and the supporting systems needed to keep a facility operating.
The International Energy Agency estimates that data centres consumed approximately 415 terawatt-hours of electricity in 2024, or about 1.5 per cent of global electricity consumption. It projects that this could rise to approximately 945 terawatt-hours by 2030, with artificial intelligence among the main drivers of that growth.
A typical AI-focused data centre can consume as much electricity as 100,000 households, while some of the largest facilities under development could use substantially more.
These figures are global, but they have particular significance for Africa. In many markets, electricity supply remains unreliable, grid capacity is constrained and businesses already depend on private generation.
Conventional data-centre operators have learned to combine grid connections with gas generation, diesel backup, battery systems and, in some cases, renewable energy. These arrangements can provide the reliability customers require, but they add to construction and operating costs.
AI infrastructure increases that pressure because it concentrates unusually high electricity demand within a relatively small location.
Reliability is as important as availability
A country may possess sufficient generation capacity in theory while still struggling to deliver reliable electricity to a specific data-centre site.
AI operators require power that is continuous, stable and capable of supporting sudden changes in computing demand. Interruptions can affect services, damage equipment and reduce the utilisation of expensive infrastructure.
“The AI engine is an extremely expensive infrastructure asset, so that equipment must be protected at all costs,” Chomse said.
Resilience therefore requires more than connecting a facility to the grid. Operators may need multiple electricity feeds, backup generators, uninterruptible power systems, batteries, fuel-storage arrangements and detailed recovery procedures.
Every additional layer increases costs. These costs are eventually reflected in the price of cloud and computing services, making local capacity less affordable for startups, research institutions and smaller enterprises.
This is why the electricity discussion cannot be separated from the commercial case for African AI. If local computing is considerably more expensive than capacity in foreign cloud regions, businesses may continue hosting workloads abroad even when domestic facilities are available.
Location will follow power
AI infrastructure may change how data-centre locations are selected.
Traditional facilities are often situated close to customers, submarine-cable landing points, fibre networks and major commercial centres. Lagos, Johannesburg, Nairobi and Cairo have attracted infrastructure partly because they concentrate enterprise demand and international connectivity.
Large computing facilities may place greater emphasis on proximity to reliable and affordable electricity.
“Large-scale AI infrastructure needs a great deal of energy in one place,” Chomse said. “That energy must either be brought to the facility or the facility must be located close to the energy source.”
This creates opportunities for African countries and regions with renewable-energy potential, natural gas, hydropower or improving electricity markets. A facility does not necessarily need to sit in the centre of a capital city if it has sufficient power and fibre connectivity to serve customers elsewhere.
However, locating computing infrastructure close to generation will require coordination among data-centre developers, electricity companies, transmission operators, telecommunications providers, regulators and local authorities.
Land, power and connectivity must be planned together. Developing any one of them in isolation may produce infrastructure that cannot operate competitively.
Cooling is becoming more complex
Electricity is only part of the challenge. Almost all the power consumed by processors eventually becomes heat, which must be removed to keep equipment within safe operating temperatures.
Traditional data centres rely heavily on air cooling. The higher density of AI equipment is increasing the need for liquid-cooling systems that move heat away from processors more effectively.
These systems require different facility designs, technical skills, maintenance procedures and, in some cases, water-management strategies.
“An AI facility is more complex to manage than the IT facilities that came before it,” Chomse said. “You have to consider liquid cooling, rapidly fluctuating power demand and the clean environment in which the equipment must operate.”
African facilities must adapt cooling systems to local temperatures, humidity, water availability and environmental risks. A design imported from a cooler region may not perform efficiently in Lagos, Accra or Nairobi.
Sustainable cooling should therefore be considered at the design stage. The World Bank’s guidance on green data centres identifies climate resilience, sustainable energy, cooling, water efficiency and electronic-waste management as important dimensions of data-centre development. World Bank
Connectivity completes the infrastructure chain
Computing capacity has limited value if users cannot reach it reliably.
Data centres need multiple high-capacity fibre routes connecting them to businesses, telecommunications networks, internet exchanges, submarine cables and other facilities. A single cable cut should not isolate a critical computing platform.
Within countries, inadequate metropolitan and long-distance fibre can prevent businesses outside the largest cities from accessing computing capacity, even when a facility has been built.
Kanwulia Okafor described connectivity as the distribution layer of Africa’s AI economy. She identified affordable, high-quality broadband extending beyond major commercial centres as the single intervention most likely to widen access.
The World Bank similarly notes that fixed fibre provides the backhaul connecting users to internet exchanges, data centres and submarine cables. Weak fibre penetration and high connectivity costs limit the types of cloud and digital services that facilities can provide.
Governments can reduce deployment costs by including fibre ducts in new roads and requiring major buildings to provide appropriate pathways for digital connectivity. This would prevent operators from repeatedly excavating completed roads or retrofitting buildings that were designed without digital access.
Operators have built around unreliable grids
Africa’s power challenges have not prevented the development of data centres. Operators have innovated around them.
Facilities use combinations of grid electricity, gas generation, diesel backup, batteries and renewable energy to achieve the reliability required by banks, telecommunications companies, cloud providers and other customers.
But building around the grid is not the same as solving the underlying power problem.
Self-generation requires additional capital, maintenance and fuel. It can also make smaller facilities less competitive because operators must reproduce much of the power infrastructure normally supplied by a reliable electricity system.
AI increases the scale of that burden. A conventional enterprise facility may be able to support its requirements through private generation. A much larger AI campus could require energy on the scale of an industrial plant.
African countries seeking major computing investment must therefore move beyond asking data-centre operators to solve power individually. The energy and digital-infrastructure sectors need coordinated investment plans.
Dedicated and regional power arrangements
Several models could support the next phase of development.
Data-centre operators could enter long-term power-purchase agreements with independent energy producers. Governments could create clear rules for dedicated generation and direct procurement while ensuring that large facilities contribute fairly to shared infrastructure.
Facilities could combine renewable generation with gas, storage and grid supply to balance reliability, cost and sustainability. Industrial parks could also be planned around shared power and connectivity infrastructure.
Regional electricity markets offer another opportunity. The West African Power Pool has connected the grids of 15 countries through more than 4,000 kilometres of high-voltage transmission lines. Approximately eight per cent of regional electricity is now traded across borders, improving access to lower-cost generation and strengthening reliability.
As regional power trading develops, digital infrastructure could be situated where power is most abundant while serving customers across several connected markets.
This approach would require resilient cross-border fibre and compatible data-governance rules, but it could allow Africa to combine regional energy resources with regional digital demand.
Power investment requires bankable demand
Energy providers face the same problem as data-centre investors: they need credible demand before committing capital.
A proposed AI facility can serve as a large, predictable customer for a power project. But the computing facility itself needs customers prepared to use its capacity.
The two investment cases are therefore linked.
A credible data-centre project could underpin a new electricity development. A reliable power agreement could, in turn, make the computing facility more attractive to customers and financiers.
Governments can help coordinate this process by aggregating public-sector cloud and computing requirements, supporting clear procurement arrangements and creating consistent rules for electricity, land, environmental approval and fibre deployment.
The World Bank argues that AI readiness depends on the coordinated development of connectivity and power, computing infrastructure, quality data and skills. It also stresses that supply, demand and regulation must move together if promising AI pilots are to scale.
Africa needs an infrastructure plan, not only an AI strategy
The African Union’s Continental Artificial Intelligence Strategy recognises reliable electricity, broadband, data centres, cloud services and computing platforms as necessary foundations for AI development. It also notes that frequent outages have forced many African operators to depend on generators and uninterruptible power systems.
Countries now need to translate that recognition into coordinated national and regional plans.
Those plans should identify:
- Where reliable electricity can support computing clusters
- How additional generation and transmission will be financed
- Which fibre routes will provide redundancy
- How facilities will manage cooling and water
- What environmental and efficiency standards will apply
- Which customers will provide anchor demand
- How smaller businesses and researchers will access the infrastructure
- How regional power and digital markets can support scale
The objective should not be to build the largest possible facility. It should be to build infrastructure that can operate reliably, attract customers and provide computing services at a competitive price.
Africa’s AI ambitions will be tested not by the number of processors announced, but by whether those processors can be kept running and put to productive use. That begins with reliable power, effective cooling and resilient connectivity – the physical foundations on which every AI service depends.
