AI and compute economics 2026-04-26 9 minute read

The African data center buildout 2026: Lagos, Nairobi, Cape Town, and Cairo

Africa hosts roughly 1.5 percent of global colocation capacity but is on track to triple installed megawatts by 2028, with Lagos, Nairobi, Cape Town, and Cairo absorbing the bulk of new builds and forcing fresh thinking on power, regulation, and inference latency.

African data center capacity is finally inflecting in 2026. Installed colocation power across the continent now sits near 700 megawatts, still a small share of the global 55 gigawatt market, but contracted pipelines suggest a path to roughly 1.8 gigawatts by 2028. South Africa, Nigeria, Kenya, Egypt, and Morocco anchor that growth. The story is no longer just connectivity; it is power, water, regulation, and the new question of where AI inference physically lives. This brief maps the operators, the grid constraints, the subsea cable build, and the localization rules shaping siting decisions, then offers three trajectories for 2026 to 2028 and how Athena and Promethean help clients act on them.

The continental picture in 2026 #

Africa entered 2026 with roughly 700 megawatts of operational colocation capacity, a figure that has roughly doubled since 2022 but still represents only about 1.5 percent of global installed power. The concentration is striking. South Africa alone accounts for around 60 percent of that total, with Johannesburg and Cape Town hosting most of the hyperscale-grade availability zones. Nigeria, Kenya, Egypt, and Morocco follow, and together those five markets capture more than 90 percent of contracted megawatts in the active pipeline.

What changed in 2024 and 2025 was the entry of generative AI demand on top of a previously enterprise-driven cloud migration story. Hyperscalers, sovereign agencies, and a new wave of African AI startups have all started signing multi-megawatt offtake contracts. Where a 5 megawatt deal felt large in 2022, several 2025 announcements ranged from 20 to 60 megawatts at a single campus. Lagos has emerged as the surprise demand center, driven by fintech, telecoms, and the absence of nearby alternatives for West African workloads.

The 2026 to 2028 outlook depends on three variables: how quickly utilities can deliver new connections, whether subsea capacity matches inland buildout, and whether regulators converge on workable cross-border data flows. None of these is guaranteed, which is why the dispersion across plausible scenarios is wider than for any other regional compute market we track.

Who is building, and where the hyperscalers have planted flags #

The operator landscape is consolidating around a handful of pan-African platforms backed by infrastructure capital, plus the local arms of global colocation names. Africa Data Centres, owned by Cassava Technologies and backed by US development finance, remains the largest pure-play, with anchor sites in Johannesburg, Cape Town, Lagos, Nairobi, Accra, and Kigali. Equinix entered through its 2022 acquisition of MainOne and now operates connected campuses in Lagos, Accra, and Abidjan. Open Access Data Centres, part of WIOCC, has scaled aggressively in Lagos and is expanding into Dakar and Mombasa. Raxio focuses on emerging markets including Uganda, Ethiopia, Mozambique, and Cote d'Ivoire. Liquid Intelligent Technologies operates a Pan-African fiber and edge footprint that complements its Africa Data Centres affiliate.

Hyperscaler availability zones remain scarce, which is the central asymmetry shaping how AI workloads are placed. Microsoft Azure operates a full region in Johannesburg with three availability zones, AWS launched its Cape Town region in 2020 and continues to expand it, and Google Cloud Platform opened its Johannesburg region in 2024. North Africa is served indirectly through Madrid, Milan, and Dubai, with Egyptian and Moroccan workloads typically routing to Europe. There is no operational hyperscaler region in West Africa yet, though both AWS and Azure have publicly indicated study work, and Lagos is the most likely candidate for a 2027 to 2028 announcement.

The table below summarizes the four flagship markets as of early 2026.

MarketLive capacity (MW)2026 to 2028 pipeline (MW)Hyperscaler regionLead operators
Johannesburg and Cape Townapprox. 420approx. 550Azure JNB, AWS CPT, GCP JNBAfrica Data Centres, Teraco (Digital Realty), NTT, Vantage
Lagosapprox. 110approx. 380None (study)Open Access, Equinix MainOne, Africa Data Centres, Rack Centre
Nairobiapprox. 55approx. 180None (edge zones)Africa Data Centres, IXAfrica (Digital Realty), iColo (Digital Realty)
Cairoapprox. 45approx. 220None (Madrid routed)Telecom Egypt, Benya, EDCH
Table 1. Operational capacity and 2026 to 2028 pipeline for the four flagship African markets. Pipeline figures combine announced campuses, signed offtake, and platform expansions; not all will deliver on schedule.

Power, water, and the grid constraint #

Power is the binding constraint everywhere, but the constraint takes a different shape in each market. South Africa carries the legacy of Eskom load shedding, which through 2023 routinely hit stage 6 and forced operators to run on diesel for hundreds of hours a year. The grid stabilized through 2024 and 2025 as private renewable PPAs scaled and Eskom's energy availability factor improved, but operators still design every Johannesburg and Cape Town campus with full N+1 generation, on-site fuel storage, and increasingly with solar plus battery hybrid systems sized for at least four hours of cover. Wheeling reforms allow contracted renewable power to move across the grid, and most new builds in Cape Town and the Western Cape now anchor a wind or solar PPA at signing.

Kenya enjoys the most attractive structural picture. Roughly 90 percent of grid generation comes from geothermal, hydro, and wind, giving Nairobi operators a credible green narrative without renewable energy certificate gymnastics. The drawback is connection lead times, where Kenya Power can take 18 to 30 months for a new 20 megawatt feed in Athi River or Konza. Nigeria sits at the opposite extreme. Grid reliability in Lagos is too poor for hyperscale workloads, so every operator runs primary captive generation on gas, with grid power treated as a backup or secondary source. This inverts the usual cost stack: fuel and maintenance dominate operating expense, and access to reliable gas pipelines becomes a siting criterion. Egypt offers heavily subsidized industrial electricity tariffs and a stable grid, but the subsidy regime is under IMF-driven review, and operators are pricing in tariff normalization risk over the contract life. Morocco, less discussed, combines renewable abundance with proximity to European markets and is winning a quiet share of overflow demand.

Water is the next constraint. Cape Town's 2018 Day Zero memory still shapes cooling design, and most new South African builds specify air-cooled or closed-loop systems even at the cost of higher PUE. Cairo and Lagos face less acute water stress today but will need to plan for it as liquid-cooled AI training clusters appear in the 2027 pipeline.

Subsea cables and the inference latency map #

The compute story would not be possible without the recent subsea cable boom. 2Africa, the Meta-led system, completed its main ring through 2024 and 2025 and now lands at more than 30 sites encircling the continent, adding tens of terabits per second of design capacity. Equiano, Google's West Africa system, lit traffic from Lisbon to Lagos and Cape Town and has driven measurable price drops in Nigerian wholesale bandwidth. Africa-1, led by a consortium including e&, lands along the East Africa, Red Sea, and Mediterranean corridor and is targeted for full service in 2026. EllaLink, while primarily a Brazil to Europe system, anchors transatlantic latency that matters for African operators routing to Sao Paulo or Fortaleza.

For AI workloads the relevant question is no longer raw bandwidth but inference latency to the user. A user in Lagos hitting a model hosted in Frankfurt sees roughly 90 to 110 milliseconds of round-trip latency on a good day, which is fine for chat but degrades agentic tool-use and voice. Hosting that same model in Lagos collapses latency to under 15 milliseconds. This is the economic case for in-country inference even when training continues to happen in North America or Europe. We expect the 2026 to 2028 buildout to be dominated by inference-first capacity rather than training campuses, which require power densities and water access that only South Africa and selected Egyptian sites can credibly support at scale.

Regulation, localization, and the forex problem #

Data localization rules are tightening across the continent and increasingly drive siting. Kenya's Data Protection Act, in force since 2019 and actively enforced by the Office of the Data Protection Commissioner from 2022 onward, requires that processing of sensitive personal data and certain regulated categories occur on servers physically located in Kenya, with cross-border transfer subject to adequacy or contractual safeguards. Nigeria's NDPR, supplemented by the 2023 Nigeria Data Protection Act, takes a similar posture and has been used to require local hosting for fintech and health data. South Africa's POPIA, fully effective since 2021, allows cross-border transfer under prescribed conditions but is increasingly read by sectoral regulators, particularly in financial services, as favoring in-country processing.

Egypt's 2020 Personal Data Protection Law adds a licensing requirement for data controllers and processors, which has slowed some cross-border deals but also created a clearer playing field for operators willing to register. Morocco's CNDP regime is the most aligned with European GDPR norms, which is part of why Casablanca and Rabat are emerging as a hedge for European-facing African workloads.

Foreign exchange repatriation is the under-discussed risk. Nigeria's chronic dollar shortages through 2023 and 2024 stranded operator earnings and forced several global vendors to write down receivables. The 2024 to 2025 naira reforms improved access but did not eliminate the friction. Egypt has run a parallel pattern, with the March 2024 devaluation finally clearing a multi-year backlog. Operators now structure contracts with hard currency clauses where possible, price in a forex haircut, and increasingly route receivables through offshore special purpose vehicles. Any compute economics model for Africa that ignores the forex line item will overstate returns by 200 to 400 basis points.

Three trajectories for 2026 to 2028 #

The range of plausible outcomes is unusually wide, so we frame three named trajectories rather than a point forecast.

Trajectory2028 installed capacityDefining assumptionWinnersStress points
Lagos Leapfrogapprox. 2.0 GWAWS or Azure announces a Lagos region by 2027 and Nigerian gas supply holds; localization rules push regional fintech and AI workloads onshoreWest African operators, gas IPPs, Equiano-connected carriersForex repatriation, Lagos land cost, talent supply
Southern Anchorapprox. 1.7 GWEskom stability holds, wheeling scales, and Johannesburg captures most AI training demand while Lagos and Nairobi absorb inferenceTeraco, Africa Data Centres, renewable PPA developersWater constraints in Cape Town, slower West African growth
Fragmented Driftapprox. 1.3 GWHyperscalers delay new African regions; localization rules diverge; forex stress returns in Nigeria and EgyptEdge operators, sovereign clouds, European hyperscaler regions serving Africa remotelyStranded pipeline, write-downs, talent flight
Table 2. Three trajectories for African data center capacity through 2028. The base case sits between Southern Anchor and Lagos Leapfrog.

How we help: Athena, Promethean, and the next twelve months #

Two of our offerings map directly to the decisions clients face right now. Athena, our compute economics and AI infrastructure practice, builds bottom-up unit economics for African data center investments, including power stack modeling, forex sensitivity, and inference latency analysis tied to specific user geographies. We have run Athena engagements for hyperscaler siting committees, sovereign infrastructure funds, and African operators raising growth capital. Promethean, our energy and grid advisory, models the power side: PPA structuring, captive generation economics, wheeling design in South Africa, and gas supply diligence in Nigeria. The two practices increasingly run in tandem because the compute and power decisions can no longer be sequenced.

Over the next twelve months we expect three discrete decision moments to drive client demand: hyperscaler region announcements that will reset the West African competitive map, Egyptian tariff reform that will reprice a generation of Cairo investments, and the first wave of liquid-cooled AI training campus proposals in South Africa. Clients positioning for any of these should have their compute, power, and regulatory diligence stitched together now, not after the announcement.

If you are evaluating an African data center investment, financing a power stack, or stress-testing a localization strategy, /engage with our Athena and Promethean teams and we will scope a focused diligence sprint.

Sources #

Cite this brief

@misc{hossen2026africandatacenterboom2026,
  author = {Hossen, Md Deluair},
  title  = {The African data center buildout 2026: Lagos, Nairobi, Cape Town, and Cairo},
  year   = {2026},
  url    = {https://deluair.com/consultancy/insights/african-data-center-boom-2026},
  note   = {Deluair Consultancy briefs}
}