The same AI investment wave reshaping cloud infrastructure in Singapore, Frankfurt, and northern Virginia is reaching Africa — slowly, unevenly, and weighted toward two countries. Africa’s data centre market, valued at approximately $1.94 billion in 2025, is projected by industry analysts to reach $4.36 billion by 2031: a compounding growth rate that reflects both genuine demand expansion and the acute scarcity of the base from which that growth starts. The continent that houses 18 per cent of the world’s population and some of its fastest-growing developer communities has, until very recently, had almost no purpose-built AI compute infrastructure at all.
Two investments made in the past eighteen months represent the clearest markers of what a serious data centre bet on Africa now looks like — and what it requires to work.
MTN’s Lagos Bet: The Sifiso Dabengwa Data Centre
MTN Group’s Sifiso Dabengwa Data Centre, named after the operator’s former chief executive who died in 2020, opened its Phase 1 facility in Ikeja, Lagos in July 2025. The site reached 4.5 megawatts of commissioned IT capacity at Tier III specification — meaning it is designed for concurrent maintainability, able to sustain maintenance on any component without taking the facility offline. For a data centre in Nigeria, Tier III certification is not a formality. It is a statement about the power architecture required to meet that standard in a grid environment that cannot be relied upon as a primary source.
Phase 2, expected in the second half of 2026, will expand the facility to 9MW and introduce AI-optimised GPU infrastructure — the GPU racks, high-density cooling, and power delivery systems required to run training and inference workloads at commercial scale. The total investment across both phases is $240 million. When complete, the Sifiso Dabengwa Data Centre will be the largest carrier-neutral, Tier III data centre in Nigeria and one of the most significant on the continent.
MTN’s choice of Lagos for its flagship data centre is commercially logical: Nigeria is home to Africa’s largest startup ecosystem by deal count, its most active developer community, and the continent’s highest mobile data consumption. The demand signal is clear. The infrastructure challenge — power — is equally clear.
The Power Constraint No One Can Engineer Away
Nigeria’s national grid has a generation capacity that routinely falls short of demand, with actual power delivery averaging between 3,500 and 5,000 megawatts against an estimated demand of at least 30,000 megawatts. The gap is not bridged by renewables — it is bridged by diesel. For data centres operating in Lagos, diesel generation is not a backup option. It is the operational baseline.
The financial consequence is structural. In mature data centre markets — the United States, Germany, Singapore — energy costs account for 35 to 45 per cent of total operational expenditure. The energy mix is primarily grid power, purchased at regulated or market rates under long-term power purchase agreements. A Lagos data centre running predominantly on diesel faces energy costs of 55 to 65 per cent of total opex, depending on generator configuration, diesel price movements, and whatever grid power is intermittently available. That 15-to-20-percentage-point premium on energy costs does not disappear with capital investment. It is a structural feature of operating AI compute infrastructure in Nigeria’s current energy environment.
AI workloads make this constraint particularly acute. Large language model inference — serving a single AI query — requires roughly ten times the energy of a standard web request. Training runs are orders of magnitude more energy-intensive. A data centre designed to host AI GPU clusters requires high-density power delivery — racks drawing 30 to 50 kilowatts rather than the 5 to 10 kilowatts typical of standard enterprise IT — at a facility that must, in practice, generate most of its own power.
The economics are not fatal. They are, however, a constraint that forces Lagos-based AI compute to be priced differently from Johannesburg, Nairobi, or any grid-connected market — and that limits the degree to which Nigerian data centre capacity can compete on pure cost for price-sensitive AI workloads.
Nigeria vs. Benchmark Markets: Data Centre Power Costs
| Market | Primary Power Source | Energy as % of Opex | Grid Reliability | AI Workload Viability |
|---|---|---|---|---|
| Lagos, Nigeria | Diesel (primary) + intermittent grid | 55–65% | Low | Viable at premium; Phase 2 GPU infrastructure pending |
| Johannesburg, South Africa | Eskom grid (improving) + DG backup | 40–50% | Medium (loadshedding declining) | Strongest current ecosystem; Azure A10 available |
| Nairobi, Kenya | Geothermal + hydro grid | 35–42% | High | Improving rapidly; Microsoft East Africa Cloud Region incoming |
| Benchmark (US/EU) | Grid (renewables mix) | 35–45% | Very high | Full GPU availability; lowest effective cost |
Note: Energy percentage of opex estimates are indicative ranges based on industry data centre benchmarks and Africa-specific operational cost analyses. Nigeria figures reflect diesel-primary generation environments. Nairobi figures assume primary grid power from Kenya’s geothermal-heavy generation mix. Sources: Uptime Institute Global Data Center Survey 2025; KenGen generation data; BETAR.africa reporting.
Microsoft and G42 in Kenya: Betting on Geothermal
Kenya’s data centre infrastructure story is being written differently, and the difference is geothermal. Microsoft’s $1 billion investment in Kenya — announced in partnership with Abu Dhabi artificial intelligence company G42 — anchors what will become the company’s first East Africa Cloud Region on a power source that no other African market can match. The Olkaria geothermal fields in Kenya’s Rift Valley generate approximately 800 megawatts of installed capacity, making Kenya one of the world’s top ten geothermal producers and giving Nairobi-area data centres access to baseload renewable electricity at costs that undercut diesel-dependent markets by a material margin.
The Microsoft-G42 investment targets 100 megawatts of total data centre capacity — more than ten times the Phase 1 scale of MTN’s Lagos facility. The East Africa Cloud Region will bring hyperscaler cloud services — compute, storage, AI services — to East African customers with in-region data residency for the first time, addressing a longstanding constraint for regulated industries: financial services, healthcare, and government departments that cannot store sensitive data offshore. The facility is expected to reach commercial operation in 2026.
G42’s involvement is strategically significant beyond the capital it brings. G42 — led by Peng Xiao and backed by Abu Dhabi sovereign wealth — has positioned itself as a vehicle for deploying large-scale AI infrastructure in markets where Western hyperscalers have moved slowly. Its partnership with Microsoft in Kenya follows a broader template: G42 provides Gulf capital and regional relationships; Microsoft provides cloud technology and enterprise credibility. The combination has moved faster than either party would likely have moved alone.
Kenya’s geothermal advantage translates directly into competitive data centre positioning. A facility in Nairobi drawing primarily from the Olkaria grid operates at energy costs structurally lower than Lagos, comparable to parts of Europe, and with a renewable energy profile that increasingly matters to multinational customers with Scope 2 emissions targets. Kenya Electricity Generating Company (KenGen) — the state utility that operates the geothermal plants — has for years been the unsung foundation of what is now becoming a data centre advantage.
What This Means for Africa’s AI Stack
The MTN and Microsoft-G42 investments do not, on their own, solve Africa’s AI compute gap — the shortage of training-grade GPU infrastructure that forces African AI developers to route workloads through US and European data centres at a 2.5-to-3x cost premium (as detailed in our earlier analysis of African AI compute costs). Phase 2 of the Sifiso Dabengwa Data Centre will introduce GPU hardware to Lagos, but on what terms, at what pricing, and accessible to which customers remains to be specified. Microsoft’s East Africa Cloud Region will bring Azure services to Nairobi, but the region’s GPU offering will be constrained by the same hardware availability limits that apply to Azure’s other African regions.
What these investments do represent is something more foundational: the physical layer of African AI infrastructure is being built. A continent that, three years ago, had almost no Tier III data centre capacity outside South Africa now has a $240 million facility going live in Lagos and a $1 billion hyperscaler region under construction in Nairobi. The gap between Africa’s share of global data centre capacity — below one per cent — and its share of global digital demand is beginning to narrow.
The constraint that will determine how quickly it narrows is not capital. It is power. Lagos and Nairobi illustrate both sides of that constraint: one market building AI infrastructure despite the energy disadvantage, the other building it because of an energy advantage. The difference between those two positions — between diesel-heavy Nigeria and geothermal Kenya — will shape which African markets become viable AI infrastructure hubs over the next decade, and which remain dependent on offshore compute for the foreseeable future.
Africa’s data centre investment wave is real. The continent’s power problem is equally real. How those two realities resolve against each other is the infrastructure story of the next five years.
