Inside Africa’s Push To Bring The AI Era Home

A fuel truck idles next to generators, its hum rivalling air handlers inside the slab-gray walls as the midday heat envelops in waves. A row of new racks blinks to life, status LEDs cascading. Observing it are engineers in antistatic coats monitoring a dashboard. This is what Africa’s AI era looks and feels like at birth: steel, silicon, and stubborn logistics converging in a single room. The scene replicates from Midrand to Mombasa.

After a decade spent watching hyperscalers and private equity carve up the globe’s compute, African operators, telcos, and policymakers are now sprinting to build AI-grade capacity. The bet is simple. If models are the new factories, then racks of GPUs are the assembly lines—and presence on the factory floor is the ticket to the next economy.

A year ago, this all sounded aspirational. Then the announcements started, each bigger than the last. In April 2024, Microsoft and Abu Dhabi’s G42 pledged a billion-dollar digital ecosystem push in Kenya that includes AI-enabled data centre capacity. A signal that geopolitics and investment were shifting toward East Africa. In June 2025, Korea’s Naver, alongside NVIDIA and a new developer, Nexus Core Systems, unveiled a plan for a 500MW AI campus in Morocco, aiming squarely at “sovereign compute” for EMEA.

Come September 2025, Cassava Technologies said it would roll out NVIDIA-powered AI factories. The first facility is set to launch in South Africa, with plans to expand to other key locations like Egypt, Kenya, Morocco, and Nigeria within the next 12 months.

Beneath the headlines, however, sits a hard truth baseline. Africa hosts less than 1 per cent of the global data centre capacity. Any credible move on AI then means changing that denominator, and fast. That is what makes this moment different. Money, hardware, and policy are finally arriving together, even as stubborn constraints such as power, water, land, fibre, and skills threaten to pull the brakes.

Let’s Meet The Builders

An apt place to start would be Cassava Technologies, parent to Africa Data Centres. In 2025, Cassava announced a programme widely reported at around $700 million to deploy GPU-powered facilities across key markets. The first phase delivered over 3,000 NVIDIA GPUs to a site in South Africa this June, with plans to fan out over the next three to four years. The ambition: create a distributed mesh of AI-capable facilities so banks, telcos, start-ups, and governments can fine-tune and serve models locally.

In Nairobi, iXAfrica and Safaricom announced what they call Kenya’s first “AI-ready” infrastructure for enterprise innovation. Their NBO1 campus targets 22.5MW, a scale that would have sounded improbable in East Africa not long ago. It positions the site to host GPU-dense clusters alongside network and storage tiers. It’s proven to be a template. Pair a carrier with a neutral colocation operator, then layer services and ecosystems on top.

Further south, Vantage continues to bulk up Johannesburg’s role as Africa’s densest market. Its Joburg I campus is designed for 80MW IT load across three data centres, and the company is building a second site near OR Tambo to chase fresh demand. Not every MW will be AI, but AI is the gravitational pull behind new density, new power feeds, and new cooling budgets.

Nigeria is moving briskly, too. In mid-October, a spate of projects was described as a billion-dollar data centre race, powered by AI demand and a young, vibrant population that practically lives online. Lagos has the subsea cables, the banks, the media houses, and the developers. All it needs is GPU-ready halls and firm power to turn this hive of activity into low-latency inference and training at home.

Then there is Morocco, which has put down a marker with plans for a 500MW AI campus. The Naver–NVIDIA–Nexus consortium’s first 40MW phase should be up and running shortly. Geography helps. Morocco can plug renewables into a campus aimed at Europe, the Middle East, and West Africa, while building domestic capacity in language AI, public administration, and industry.

MTN Group is currently in talks with the US and European partners to build AI data centres across the continent under a venture dubbed Genova. It is a push to turn network reach into compute revenue. If the model sticks, it could redraw the roles of carriers, who have fibre, land, power interconnects, and customer access, but historically lacked the platforms.

It’s About Skills, Not Just Steel

Compute without talent is a warehouse. Here, South Africa offers a signal. Microsoft has committed to train one million South Africans in AI and cybersecurity by 2026, and followed with an additional ZAR 5.4bn plan to expand local cloud and AI infrastructure while funding 50,000 technical certifications. Training does not wire GPUs into jobs on its own, but it does move the bottleneck from ‘no talent’ to ‘place the talent.’

Across smaller markets, development finance is trying to fill gaps. In April 2025, the IFC put $100 million into Raxio Group to expand neutral colocation in countries like Ethiopia, Angola, Côte d’Ivoire, Mozambique, DRC, and Uganda. The logic is not fancy. Bring workloads closer to users, improve latency, cut costs, and keep more data under local regulatory oversight. That’s table stakes for AI in health, finance, and government.

BY THE NUMBERS

• Africa’s share of global data centre capacity: <1 per cent.
• Johannesburg I campus design: 80MW IT load.
• Nairobi NBO1 campus target: 22.5MW.
• Cassava/NVIDIA first tranche: ~3,000 GPUs delivered in South Africa.
• IFC Raxio debt funding: $100 million.

But Why Now?

Three catalysts converged. Geopolitics. Hardware. Demand.

First, geopolitics. America’s scrutiny of chip flows and cloud partnerships has collided with Gulf capital’s appetite for AI platforms. The Microsoft–G42 deal in Kenya shows how those currents can align with African priorities when structured well—skills, safety commitments, and compute on the ground rather than only in distant regions. A separate G42–EcoCloud announcement in Kenya back in March 2024 sketched a vision for up to 1GW of green-powered capacity built on geothermal, a reminder that the continent’s energy profile can be an asset if financing and interconnects line up.

Second, hardware scarcity is easing. The floodgates did not open, but the pipeline of GPUs to non-traditional markets improved in 2025. Cassava’s June delivery is a case in point, and the Morocco commitment signals that suppliers are willing to back large campuses in North Africa if power and permitting look credible.

Third, local demand got specific. Banks want real-time fraud models at lower latency. Miners want predictive maintenance and route optimisation at the pit face. Health ministries want language models aligned to local data governance. None of that runs well when the GPU is 6,000 kilometres away, and the fibre is congested.

And Now For The Constraint Stack

Power is the first and last question. South Africa’s grid has been unreliable; countries that depend on hydro face drought risk; diesel is a budget-killer; and solar alone won’t chill a 30MW hall at noon and midnight. That’s why developers are wiring deals that pair the grid with power purchase agreements. PPAs, and where possible, geothermal, wind, or gas. Morocco’s 500MW concept leans on renewables. Kenya’s geothermal gives Nairobi a story few peers can tell. But none of this is ready-made. It takes land rights, substations, and yes, a lot of patience.

Water is the next constraint. High-density AI clusters are a heat problem, which makes cooling a water problem unless operators go air-cooled or adopt heat-recovery and non-heat-transferring systems carefully. Expect more designs that push liquid cooling deeper into the rack.

Skills matter twice: to build and to use. The training commitments in South Africa offer a roadmap. So do vendor academies, university partnerships, and on-site apprenticeships. Without that, you get idle GPUs.

Finance is changing shape. Traditional colocation spreadsheets can’t keep up when one data hall’s power bill matches that of a small power company. That’s attracting infrastructure investors and new consortium partnerships. Especially when global money is paying attention. When a BlackRock-led group can stitch together mega-deals for AI-optimised platforms elsewhere as the AI Infrastructure Partnership (AIP), it forces African developers to sharpen their own funding models and partnerships.

THE PLAYERS: WHO’S WHO

• Cassava Technologies / Africa Data Centres — rolling out NVIDIA-based AI “factories” across multiple markets.
• iXAfrica + Safaricom: Nairobi AI-ready campus and services at 22.5MW design capacity.
• MTN Group (Genova): telco-led AI data centre venture in talks with U.S./EU partners.
• Naver–NVIDIA–Nexus (Morocco): 500MW AI campus targeting EMEA and sovereign compute.
• IFC → Raxio: DFI capital pushing neutral colocation across underserved markets.
• Microsoft–G42 (Kenya): skills + infrastructure package with regional implications.
• Vantage: large-scale capacity anchor in South Africa’s core market.

What About The Data Sovereignty Question?

AI revives old arguments with new stakes. Where should public data live? Who sets safety rules for models used in courts, clinics, and classrooms? Countries eager to ride the AI wave are not eager to give up agency. This is why ‘sovereign compute’, a nation’s ability to control and manage its own computing infrastructure, especially cloud, AI, and data processing within its borders and under its laws, has crept into policy speeches from Rabat to Pretoria. Local capacity does not solve trust overnight, but it moves the conversation from renting everything abroad to running critical workloads at home.

Enter the money equation. A CFO will want to know three things: how much each token costs, what faster speed is worth, and how much unused capacity might cost. Renting H100S overseas hides power and cooling costs, while owning or leasing locally makes them visible but gives control over data, speed, and system integration. For customer apps, every millisecond counts; for regional model training, transferring huge data across oceans is slow and costly. The main risk is unused hardware. Something best solved through shared platforms, strong ecosystems, and workloads that stay local.

THE TIMELINE

• May 2024: Microsoft & G42 announce a $1B digital ecosystem initiative for Kenya that includes AI-enabled capacity.
• Jan–Mar 2025: Microsoft pledges to train 1M South Africans by 2026 and commits ZAR 5.4bn to expand cloud/AI infrastructure.
• Apr 2025: IFC invests $100M in Raxio to expand data centres across six African countries.
• May 2025: iXAfrica + Safaricom announce Kenya’s first AI-ready data centre services; NBO1 targets 22.5MW.
• Jun 2025: Naver–NVIDIA–Nexus unveil a 500MW AI campus in Morocco; first 40MW phase slated for Q4 2025.
• Jun 2025: Cassava begins rollout with ~3,000 NVIDIA GPUs delivered to South Africa facility.
• Sept–Oct 2025: MTN confirms talks to build AI data centres continent-wide; Nigeria projects stack up, billed as a billion-dollar race.

What Gets To Run First?

Expect the money people (BFSI – Banking, Financial Services, and Insurance) to lead with fraud models that must learn from local patterns. Telcos will follow with network optimisation and customer care co-pilots. In mining and energy, predictive maintenance pays for itself. In health, localised language models can triage and translate, but they will need guardrails and on-prem options for sensitive data. Governments will chase service delivery gains such as permits, registries, and call centres, where latency and uptime matter.

The big unlock is African languages and dialects. Training and fine-tuning models on Swahili, Yoruba, Amharic, isiZulu, Arabic dialects, and mixed code speech requires curated data and nearby compute. That is not just cultural equity, either. It is product quality. The more an assistant or agent understands the way people actually speak, the more it adapts, and the more the economics of local compute make sense.

The Risk Of A Hollow Boom

Hype is cheap. MW is not. A dozen press releases do not make a market if the grid falters, permits drag, or capital costs spike. There is also a governance risk. If a few players control all the new capacity, the goal of wider access and inclusion disappears. The better outcome is what you can already see in places like Nairobi and Johannesburg with neutral colocation, carrier participation, open peering, strong IXPs, developer programmes, and skilling at scale.

The other risk is that the world moves on. NVIDIA’s Blackwell and its successors will change the power and cooling math again. If Africa locks into the wrong generation at the wrong density, it could spend years paying for sunk costs. Mitigation means using modular designs, liquid-ready cooling, and flexible contracts that let operators upgrade without rebuilding.

WHAT TO WATCH IN 2026

• Grid deals: Long-dated PPAs tied to renewables and flexible baseload.
• Cooling: Wider adoption of liquid to handle rack densities for next-gen GPUs.
• Talent: Proof that skilling programs flow into jobs on the floor.
• Policy: Clearer rules for data residency, AI safety, and public procurement.
• Flagships: The first named, at-scale African AI workloads running in-region for banks, ministries, and platforms.

Back in the data hall, the lights pulse steadily, and the power levels even out. A technician notes a successful test and wheels the ladder away. It doesn’t look like a revolution, but it is. These rooms are where Africa claims its future, not by renting power from others, but by creating spaces where her own ideas can run at full speed.