Only 5% of Africa’s data centres are AI-ready: what it means for South Africa and Nigeria

“This is not a retrofit.”

Johnson Agogbua said it in March, walking visitors through the nearly finished Kasi Cloud facility in Lekki, Lagos. He meant that the building had been designed for AI from the ground up, not converted afterwards. Few data centres in Africa can say the same.

The number

On October 6, the World Bank published its Africa Economic Update, subtitled “Building AI Readiness.” Africa is home to nearly 18% of the world’s population =but holds only 0.6% of global data centre capacity. And about 5% of the continent’s data centres are ready for AI.=

Business outlet Sika Finance turned that share into a count: barely eight data centres out of roughly 160.

That count should be read with care. Depending on the source and on what qualifies as a data centre, estimates range from about 160 to more than 220 facilities. The 5% is an order of magnitude, not a census.

What makes a data centre AI-ready

“An AI data centre is fundamentally about high rack density,” Krish Ranganath, Regional Executive for West Africa at Africa Data Centres, told TechCabal in January.

A standard rack running websites, business software or databases draws around 10 to 15 kW. AI workloads run on GPUs, which need far more. An entry-level AI rack might need about 25 kW. The most demanding run at 60 to 100 kW and usually require liquid cooling. Sika Finance sets the threshold for “AI-ready” at 40 kW per rack.

So the issue isn’t how new or well connected a facility is. It’s whether its power supply, electrical distribution and cooling were designed for that density from day one. Upgrading an older site means rebuilding its power and cooling, not adding servers.

South Africa: most of the capacity, now adding GPUs

South Africa hosts the bulk of the continent’s data centre capacity, about 70% according to an analysis by TechCabal Insights. It also has a head start on AI hardware.

In November 2025, Cassava Technologies launched a GPU-as-a-service offer built on Nvidia technology, with its first hardware installed near Cape Town. Its founder, Strive Masiyiwa, said the group had acquired 12,000 Nvidia GPUs for five sites across the continent.

\=Power has also become more predictable. The 2022 Energy Action Plan and private power purchase agreements helped bring load-shedding to an end. For a technology whose racks draw power around the clock, that matters as much as the GPUs.=

The other side of this lead is concentration. When the continent’s AI-ready capacity sits mostly in one country, companies elsewhere in Africa that need it send their data there, or further away.

Nigeria: building from scratch on an unstable grid

Lagos hosts nearly three-quarters of Nigeria’s commercial data centre capacity, and a new generation of facilities is being built there with AI in mind. Kasi Cloud’s LOS1, backed by the Nigeria Sovereign Investment Authority, was officially launched in May and bills itself as West Africa’s first AI-ready data centre. Rack Centre markets its 12 MW LGS2 site as AI-ready, and Airtel Africa’s Nxtra has been building capacity dedicated to AI compute.

The stakes, as Agogbua puts it: “If we do not create our own AI factory and capacity, we will simply be buying it from others and remain consumers.” His company estimates that Nigerian businesses spend around $850 million a year on foreign cloud services.

The constraint is the grid. National electricity generation fluctuates between 3,000 and 4,000 MW. According to TechCabal, energy costs for operators have risen 64% since January, and cooling alone accounts for nearly 40% of their energy bills. Most operators run hybrid systems combining gas, solar and batteries to stay online.

The Lagos State government wants to triple the city’s data centre capacity, to more than 250 MW by 2030.

Power decides

Both countries point to the same conclusion. In July, an IMF paper estimated that AI could add about 4% to sub-Saharan Africa’s GDP over a decade, but only with better electricity, connectivity and skills. Without reliable power, the gains in its models are negligible. Around 600 million people in the region still have no access to electricity.

What it means for a business

Not every company needs a 100 kW rack.

Training a large model requires concentrated GPU capacity, and very few African companies will ever do it. Most needs are about using AI: running a model on internal data, searching documents, detecting fraud, automating customer service. Many of these run on much lighter infrastructure, especially with smaller models designed to be frugal, like those built by Lelapa AI.

Three questions are worth asking before signing with any provider:

1. Do I need to train models, or to use them on my data?
2. Where will those models run, and where will my data go?
3. What does “AI-ready” mean in the offer: what power per rack, what cooling, in which country?

These questions apply to every cloud provider operating on the continent, Hodi included.

As with Tier III certification, the label matters less than what’s verifiably behind it.

The gap to watch

For now, the 5% is concentrated in a few markets, mainly South Africa, Kenya and Nigeria, where the World Bank also sees most of the continent’s AI activity. The next count will show whether the new capacity being built in Lagos and Johannesburg starts serving the rest of the continent, or stays where it is.

FAQ

What makes a data centre AI-ready?
Its ability to power and cool high-density racks of GPUs, typically from about 25 kW per rack for entry-level AI workloads to 60–100 kW for the most demanding ones, often with liquid cooling. Standard racks draw 10–15 kW.

How many data centres does Africa have?
Estimates range from about 160 to more than 220, depending on the source and on what is counted. Together they represent about 0.6% of global capacity, according to the World Bank.

Where are Africa’s AI-ready data centres?

Mainly in South Africa, which holds most of the continent’s capacity, and increasingly in Lagos, Nigeria, where several facilities designed for AI workloads have opened or are under construction.

Sources

• World Bank, Africa Economic Update “Building AI Readiness,” October 6, 2026 (via The Rio Times)
• Sika Finance, “Seuls 5% des datacenters africains sont actuellement prêts pour l’IA”
• TechCabal, “Will Nigeria get its first AI data centre in 2026?”, January 8, 2026
• TechCabal, “Lagos wants to triple its data centre capacity by 2030”, May 20, 2026
• IT Edge News, Kasi Cloud facility tour, March 2026
• Nairametrics, Kasi Cloud LOS1 launch, May 21, 2026
• Connecting Africa, Cassava GPU-as-a-service, November 12, 2025
• TechCabal Insights, African data centre capacity
• Semafor, IMF paper on AI and power, July 22, 2026