The digital world is often described in weightless terms. Clouds. Streams. Virtual spaces. In reality, the internet sits on concrete, copper, and cooling fans. As the race for Artificial Intelligence accelerates, Africa is ready to innovate, but the question on the table is harder: are we ready to power the machines?
In April 2026, President William Ruto admitted publicly what had been suspected inside Kenyan government circles for over a year. The flagship 1 billion US dollar data centre project backed by Microsoft and Abu Dhabi-based G42 could not proceed at full scale because the country did not have the electricity to run it. The project, first announced during Ruto’s May 2024 state visit to Washington, was meant to be powered by geothermal energy at Olkaria and to launch Microsoft’s first East African Azure cloud region by May 2026. That deadline is gone.
This is not just a delayed business deal. It is a continental warning. Africa’s AI ambitions are outpacing the grids underneath them, and unless that is fixed, the continent will host the announcements but not the infrastructure.
The Power Paradox
The challenge is not uniform across the continent. The result, however, is consistent: a bottleneck between announcement and execution. Three country profiles set out the problem clearly.
Nigeria: the capacity gap
Nigeria is a tech giant with a structural power problem. Installed generation capacity stands at 13,625 MW, yet available generation for dispatch averaged just 5,151 MW in December 2025, a Plant Availability Factor of only 38 per cent. For a data centre that requires 99.9 per cent uptime, relying on the Nigerian grid is not an option. Operators are forced to bring their own power, mainly captive gas and diesel plants, which drives up the cost of every local cloud service Nigerian businesses consume.
Uganda: the distribution gap
Uganda offers a different irony. The country produces more electricity than it currently consumes. Installed capacity climbed to 2,007 MW after the 600 MW Karuma hydropower plant came fully online in June 2024, while peak demand remains below 900 MW. On paper that is a surplus. In practice, transmission and distribution losses average over 18 per cent, and the network struggles to deliver high-voltage, stable loads to specific industrial sites. A hyperscale data centre in Entebbe or Kampala cannot run on national pride. It runs on conditioned power at the meter.
Kenya: the half-the-country warning
Kenya’s position is the bluntest of the three. Olkaria’s geothermal complex generates roughly 950 MW today against a national installed capacity of around 3,000 MW. The G42 facility was scoped to scale to 1,000 MW. Even if every available megawatt at Olkaria was diverted to one customer, households and industries already connected to that supply would be priced and physically out. A single hyperscale facility commanding a third of national capacity is no longer an infrastructure project. It is a national security decision.
Why AI Changes the Maths
Traditional data centres are energy-hungry. AI data centres are gluttons. According to the International Energy Agency and the Electric Power Research Institute, a single AI query can consume up to ten times the electricity of a standard Google search. Goldman Sachs Research projects that global data centre power demand will rise by 160 per cent by 2030.
Three pressures combine:
- Density. AI workloads run on specialised GPUs that generate immense heat. GPU-accelerated servers consume between 3,000 and 5,000 watts compared with 300 to 500 watts for traditional CPU servers.
- Cooling. In Africa’s tropical climates, cooling these machines can consume close to the same power as running them.
- Baseload. AI models that train 24/7 cannot tolerate load shedding. A ten-second flicker can erase hours of computational work and corrupt expensive training runs.
The implication for African operators is direct. The grid does not just need more megawatts. It needs cleaner megawatts, more reliable megawatts, and a transmission network that can move them precisely to where the racks are.
Case Studies: How Four African Markets Are Responding
The continent is not standing still. Four case studies show the range of approaches now in play, what is working, and what is not.
The Microsoft and G42 Olkaria project
Microsoft and G42 announced a 1 billion US dollar geothermal-powered data centre near the Olkaria fields in Naivasha in May 2024. Phase one targeted 100 MW, with a long-term build-out to 1,000 MW. The project was structured on a build-operate-transfer model, with the Kenyan government committed as anchor tenant for Azure cloud services.
By August 2025, the May 2026 launch was already lost. The National Treasury declined to clear the concept note, sovereign payment guarantees demanded by lenders were never issued, and Kenya’s electricity system could not absorb the load without disrupting households and industry. President Ruto’s office has since clarified that the project is not cancelled, but the original scale and timeline are not coming back.
Teraco and the wheeling model
South Africa is now the largest data centre market in Africa, valued at 2.16 billion US dollars in 2024 and projected to reach 3.4 billion by 2030. Teraco, owned by NYSE-listed Digital Realty, has reached 189 MW of total capacity across Johannesburg, Cape Town, and Durban. The CT2 hyperscale expansion in Cape Town was completed in November 2025 at 50 MW, using a zero-water closed-loop cooling system suited to a drought-prone city.
The substantive innovation sits behind the racks. Teraco is building a 120 MW solar PV plant in the Free State, the largest clean energy project for an African data centre, and is wheeling that power across the national grid into Ekurhuleni and Cape Town under formal agreements with the municipalities. The company has committed 3.5 billion rand to 200 MW of utility-scale solar by 2035.
GERD, Bitcoin miners, and the rural blackout problem
The Grand Ethiopian Renaissance Dam was inaugurated on 9 September 2025, doubling the country’s installed capacity to around 5,300 MW, with 92 per cent from hydropower. Faced with a transmission network that cannot yet deliver this generation to the population, Ethiopian Electric Power began selling power to Bitcoin mining operators at roughly 3.14 US cents per kilowatt-hour, well below the global average. By 2024, Ethiopia was accounting for around 2.5 per cent of global Bitcoin hashrate, drawing about 600 MW, or 11 per cent of installed capacity.
The economics looked attractive. In the 2023/24 fiscal year, EEP earned over 25 million US dollars from mining contracts, with some estimates suggesting figures rose to over 220 million in 2024. Then the cost became visible. The Ethiopian Energy Outlook 2025 reported that crypto mining was on track to consume one-third of Ethiopia’s total power output, while 15 million households remained off-grid and more than 40 per cent of the population still lacked reliable access. In August 2025, EEP announced the gradual phase-out of all crypto mining contracts.
Rack Centre, Equinix, and the bring-your-own-power model
Nigeria hosts around 20 data centres, the majority concentrated in Lagos for its proximity to submarine cable landing stations. The market is projected to grow from 288 million US dollars in 2025 to over 1 billion US dollars by 2031. New capacity coming online tells a clear story about how operators are working around the grid.
Rack Centre’s 12 MW LG2 facility, commissioned in March 2025, runs on a power mix of gas and diesel, with solar in the planning pipeline. MTN Nigeria is constructing a 240 million US dollar data centre in Lagos. Equinix announced a 22 million US dollar LG3 facility in November 2025, the first phase of a 100 million US dollar Africa expansion. None of these facilities are designed to rely on the national grid as primary power. Captive gas generation, on-site diesel, and increasingly PPAs with independent solar producers are now the default architecture.
The Path That Actually Works
If Africa is to host the next wave of AI infrastructure rather than rent it from Frankfurt, Dublin, and Northern Virginia, the answer is not to repeat Olkaria-style announcements. It is to build the energy backbone that lets the announcements survive contact with reality. Four interventions matter.
Renewable integration with transmission, not without it
Kenya’s geothermal and Ethiopia’s hydropower are world-class resources. Both remain trapped behind transmission networks that cannot move power at scale to industrial off-takers. Generation investment without parallel transmission investment is generation that bills the public and serves nobody.
Decentralised power, energy islands, and wheeling
South Africa’s model is instructive. Build data centres as energy islands with dedicated renewable assets, wheel surplus back into the grid, and structure direct municipal contracts. The state does not need to fund every megawatt. It needs to license and regulate the model that allows private capital to do it.
Treat data centres as industrial utilities, not buildings
Planning approval, zoning, water rights, and high-voltage line allocation must be handled as critical national infrastructure. Egypt and South Africa are already moving in this direction with special economic zones and dedicated licences. The countries that drag their permitting timelines will simply lose the projects to those that do not.
Anchor demand before announcing capacity
The Microsoft and G42 deal failed in part because the commercial off-take was never secured. Future African data centre announcements should be preconditioned on signed enterprise and government anchor tenancy, sovereign payment guarantees where required, and demonstrated grid headroom. No anchor demand, no announcement.
The African AI infrastructure story is not over. It is being rewritten in real time. The capital is there, the talent pipeline is there, and the geographic logic of regional cloud hubs in Lagos, Nairobi, Cape Town, and Cairo is sound. What is missing is the energy architecture beneath them. The investors who back transmission, wheeling, captive renewables, and gas-to-power infrastructure will own the rails on which the continent’s digital economy runs.
The Bottom Line
The cloud is not in the sky. It is in the ground, and it is thirsty for volts. Africa cannot leapfrog the industrial age of energy if it intends to lead the digital age of AI. Without a radical fix for distribution networks and baseload capacity, the continent’s AI revolution will remain buffered, hosted abroad, and billed back to African enterprises at a premium.