The Stack Sovereignty Test: Why Compute Is the Infrastructure Africa Cannot Afford to Outsource
Your data is being processed right now. Just not in Africa.
By Oluwaseyi Ayodeji | June 9, 2026
Series note:This is the third article in the Sovereign Stack series, which introduces the Stack Sovereignty Test - a five-pillar framework for assessing whether a nation truly controls its AI future. Last week, we explored Talent - and why Africa's most urgent AI problem is the one that walks out of the airport. This week, we turn to Compute: the physical layer on which everything else depends.
Every time a Nigerian bank runs a fraud detection model, every time a Kenyan startup queries a large language model, every time a South African government agency searches its own citizen database - there is a very good chance that transaction is being processed in a data center that is not on African soil.
That is not a metaphor. It is an engineering fact.
And it is the second most important question the Stack Sovereignty Test asks: Does your nation own the machines that think for you?
The Numbers You Need to Know
Let me give you a sense of the scale of the problem in plain terms.
Africa currently accounts for just 0.6% of global data center capacity. Active capacity across the entire continent stands at approximately 360 megawatts, with another 238 megawatts under construction and 656 megawatts in the planning stage. That last number is promising - until you consider the continent's actual need.
The Africa Data Centres Association estimates that the continent requires at least 1,000 megawatts of new capacity across 700 facilities just to meet current demand. We are not even halfway there. And demand is not standing still.
McKinsey projects that data center capacity in Africa's five largest markets will need to grow from roughly 400 megawatts today to between 1.5 and 2.2 gigawatts by 2030. That is a fivefold increase in under six years, in a continent that has historically struggled to build and sustain basic power infrastructure.
Here is what those numbers mean in practice: the overwhelming majority of AI workloads generated on the African continent - customer service bots, credit scoring engines, medical diagnosis tools, language translation models - are being computed offshore. The electricity that powers those calculations is paid for by African businesses. The data that trains those models originates from African users. But the economic value of that compute - the jobs, the tax revenue, the engineering capability - flows to data centers in Amsterdam, Virginia, and Singapore.
Nigerian enterprises alone spend an estimated $850 million annually on foreign cloud infrastructure - capital that exits the economy and sits under foreign legal jurisdiction.
That number is for one country. Multiply it across 54 nations and the figure becomes a structural drain that no trade policy or currency regime can fully compensate for.
What a Data Center Actually Is - And Why Tiers Matter
Before we go further, I want to take a moment for the people reading this who are new to the concept. This is not a conversation for engineers only. It is a conversation for anyone who wants to understand where AI actually lives.
A data center is, at its simplest, a building full of computers. But not the kind of computers you have at home. These are racks of servers - machines that run continuously, process millions of calculations per second, and generate enormous amounts of heat that must be cooled around the clock. They need uninterrupted power, high-speed internet connections, physical security, and trained staff to keep them running.
The industry classifies data centers in four tiers:
Tier I and II facilities are basic. They have some redundancy - meaning if one component fails, another can take over - but they are designed for simple workloads. A small business server room might qualify. These are not suitable for running AI workloads at scale.
Tier III facilities are where the real infrastructure begins. They are concurrently maintainable, meaning engineers can service systems without shutting them down. Most enterprise-grade and government data centers in Africa aspire to Tier III certification. Kasi Cloud's recently commissioned LOS1 facility in Lagos is built to Uptime Institute Tier III standards, with a hybrid power system combining gas, solar, and battery storage.
Tier IV is the gold standard: fully fault-tolerant, with 99.995% uptime. These are the facilities that train the world's largest AI models. There are none in sub-Saharan Africa.
The gap between Tier III and Tier IV is not just technical - it is economic and strategic. AI model training at scale requires Tier IV environments. Which means Africa, even as it builds out Tier III capacity, remains dependent on foreign facilities for its most advanced AI development needs. We can run inference. We cannot yet train.
The KASI Moment: A Milestone That Reveals the Gap
In May 2026, something significant happened in Lagos.
Kasi Cloud Datacenters commissioned its Lekki campus - West Africa's first hyperscale-ready, AI-capable data center - at a ceremony attended by Lagos State Governor Babajide Sanwo-Olu, Nigeria's Minister of Finance Taiwo Oyedele, and NSIA Managing Director Aminu Umar-Sadiq.
Located on four hectares in the Maiyegun area of Lekki, adjacent to six subsea cable landing stations including the Equiano and 2Africa systems, the campus is designed to scale to approximately 100 megawatts of critical IT capacity. The first building, LOS1, is purpose-built for high-density AI and accelerated computing workloads. It delivers sub-50 millisecond latency for in-country data processing - meaning Nigerian data, processed in Nigeria, arriving at Nigerian users faster than anything routed through London or Frankfurt.
The CEO of Kasi Cloud, Johnson Agogbua, put it plainly: "For too long, Africa's data has powered someone else's economy. Today, that changes."
I want to sit with that statement for a moment, because it captures something important. Agogbua is not just describing a business. He is describing a sovereignty problem. When your data is processed abroad, it is not just your bandwidth bills that suffer. It is your legal standing. The laws governing what can be done with that data are not your laws. The courts that would adjudicate a dispute are not your courts. The regulators who set the rules are not accountable to your government.
The Kasi LOS1 commissioning is genuinely significant. Nigeria's data center market was valued at $278 million in 2024 and is projected to reach $671 million by 2030, growing at over 15% annually. NSIA - the Nigerian Sovereign Investment Authority - is a foundational investor, a signal that the federal government understands this is not just commercial infrastructure. It is national infrastructure.
And yet I will say something that the celebration does not quite capture: Kasi LOS1 is a beginning, not an arrival. Africa currently hosts approximately 230 data centers across the entire continent. South Africa leads with 56 facilities, followed by Kenya with 20 and Nigeria with 19. The United States alone has tens of thousands. A single hyperscale facility in Northern Virginia can carry more compute than everything currently operational on the African continent combined.
The milestone reveals the milestone's own limitation. We should celebrate every KASI - and we should be clear-eyed about how far the road still runs.
The Kenya Warning: When Ambition Meets Reality
While Lagos was celebrating, Nairobi was grappling with a very different story.
In May 2024, Kenya signed what was presented as a landmark deal: a $1 billion investment by Microsoft and its partner G42 to build an AI data center campus in Kenya. The announcement generated exactly the kind of headlines that governments dream about - global technology partnership, cloud services for East Africa, AI infrastructure arriving on the continent.
What followed was a masterclass in the gap between digital ambition and physical reality.
At full scale, the Microsoft-G42 facility was expected to require approximately 1 gigawatt of electricity. That figure represents nearly one-third of Kenya's current total installed power capacity, which stands between 3,000 and 3,200 megawatts.
President Ruto, to his credit, was blunt about it. His words: "To switch on that one data centre, we would need to shut off power for half the country."
The project was originally slated to be operational by May 2026. By late 2025, progress had already slowed significantly, stalled by energy constraints and administrative and financial hurdles. As of this writing, Kenyan officials describe the talks as still active - but what is under discussion is a scaled-down first phase starting at 100 megawatts rather than 1 gigawatt, with full buildout contingent on grid expansion.
There is a deeper issue buried inside this story that does not get discussed enough.
Microsoft and G42 sought sovereign guarantees that power would be available at a predictable tariff over the life of the facility. At the heart of the delay is a relatively standard but politically sensitive ask: the companies wanted the Kenyan government to underwrite a baseline level of cloud consumption over several years - pledging to purchase a set amount of cloud capacity annually, regardless of whether government agencies or local businesses actually consumed the full allocation.
Read that again slowly.
A foreign technology company building a data center in Kenya, on Kenyan soil, was asking the Kenyan government to guarantee them a revenue floor - regardless of actual usage. That is not a partnership. That is a concession agreement dressed in partnership language.
The question this raises is not whether Kenya should or should not have signed. The question is structural: what does it mean to have a data center "in Africa" when the ownership, the contractual terms, the legal jurisdiction of the data, and the guaranteed revenue stream all flow outward? Is geographic presence the same as sovereignty?
This is where the Stack Sovereignty Test asks its hardest question. A data center on African soil run under foreign terms, processing African data under foreign law, with African governments bearing the financial risk - that is infrastructure colonialism with a fiber optic cable.
True compute sovereignty requires not just the building, but the governance.
The Wall No One Talks About: Why Data Centers Are Hard to Build in Africa
So why don't African governments and entrepreneurs just build more? The demand is real. The need is urgent. The investment case, on paper, is compelling.
The honest answer is that building data centers in Africa is harder than almost anywhere else on earth - and the reasons are worth understanding.
Power is the fundamental constraint. A Tier III data center does not run on a municipal grid that experiences daily interruptions. It requires a dedicated, high-voltage power supply, backup generation, and consistent voltage regulation. Investment up-cycles are accelerating because grid constraints now favor operators who can bundle renewable power and compliance expertise - meaning the cost of energy self-sufficiency is now baked into the business model rather than left to the national grid. That adds capital cost before a single server is racked.
Financing is the second wall. Building a Tier III facility of meaningful scale costs between $50 million and $300 million. Commercial banks in most African markets do not have the appetite or the instruments for this kind of long-tenor, capital-intensive project. The blended finance toolkit - where development finance institutions like the African Development Bank, the IFC, or the DFC take a first-loss position to de-risk private capital - has worked in telecom and energy. The financing framework is familiar: blended finance structures combine development finance institutions with private capital, allowing public risk capital to absorb early-stage uncertainty while enabling commercial lenders to fund assets at appropriate tenors and pricing. What is missing is the systematic application of that framework to compute infrastructure specifically.
The precedent exists. The U.S. Development Finance Corporation committed up to $300 million to Africa Data Centres, with an initial disbursement of $83 million to expand ICT infrastructure in South Africa, Kenya, and other countries. That model - sovereign development finance unlocking private build - is replicable. It needs to be replicated faster and at greater scale.
The local talent gap for data center operations is the third wall that the first two walls produce. Building a Tier III facility requires specialized engineering: power systems, cooling design, fiber interconnection, network operations. These are not skills that can be hired off a general engineering degree. Africa's universities are producing computer scientists. The continent needs data center engineers, and right now there are too few of them - a direct consequence of the fact that there were too few data centers to train against.
This is where last week's Talent article and this week's Compute article are actually the same article told from different directions. You cannot build the talent without the infrastructure to train against. You cannot justify the infrastructure without the talent to run it. This is not a chicken-and-egg problem to be solved philosophically. It is an infrastructure investment problem that requires sequencing: build the facilities, co-develop the curriculum with operators, run apprenticeship pipelines from commissioning day one.
The Investment Case: Why the Numbers Are Better Than They Appear
If you are reading this as an investor - and I am speaking to you directly now - let me make the case for why compute infrastructure in Africa is among the most structurally sound infrastructure investments available on the continent.
The demand curve is locked in. Africa's digital economy is not speculative. The African data center market was valued at $1.94 billion in 2025 and is projected to grow to $4.36 billion by 2031 at a CAGR of 14.46%. In terms of IT load capacity, the market is expected to grow from 1.17 gigawatts in 2025 to 3.46 gigawatts by 2030 - a compound annual growth rate of over 24%. Fintech alone is producing millions of daily transactions that require AI-powered fraud detection. Healthcare, agriculture, logistics - every sector that is digitizing creates compute demand. And Africa is digitizing fast.
The competitive moat is geographic. Latency is physics. A model query answered from Lagos in 8 milliseconds cannot be replicated from Frankfurt in 50 milliseconds. As data sovereignty laws tighten across the continent - and they are tightening; Nigeria's National Cloud Policy 2025 mandates in-country hosting for sensitive government and financial data - the demand for local compute becomes regulatory rather than just commercial. Regulation is a moat.
The exit environment is improving. The global hyperscalers - AWS, Google Cloud, Microsoft Azure - are not going to build their own Tier IV facilities on the African continent in the near term. They are going to lease capacity from local operators who have already done the hard work of navigating local power, local permitting, and local community relations. Being the operator that a hyperscaler partners with is a known and highly valuable exit in this industry. 52 upcoming data centers in Africa are expected to add over 1,900 megawatts by 2030, with Nigeria, Kenya, Egypt, and South Africa contributing as the top markets. The infrastructure is being built. The question is who owns it.
The risk premium is being repriced. For years, the perceived political and operational risk of African infrastructure investment drove up cost of capital to levels that made deals unviable. That perception is shifting - because the actual track record of data center operators like Teraco, Africa Data Centres, and now Kasi is producing real returns. Risk perception lags reality. That lag is where value lives for investors who do their homework.
What Sovereignty Actually Requires
Let me bring this back to the Stack Sovereignty Test.
The Compute pillar does not just ask: do you have data centers? It asks:
Who owns them? A foreign-owned data center in Lagos is better than no data center in Lagos - but it is not sovereign infrastructure.
Under whose law does the data operate? Geographic presence without legal jurisdiction is a building without a flag.
Who has the skills to run and iterate them? Infrastructure that can only be maintained by imported expertise is infrastructure that can be switched off by another country's export controls.
Who financed them - and on whose terms? Sovereign guarantee agreements that lock African governments into purchasing minimums for foreign-operated facilities are not a solution. They are the next version of the problem.
With over 40 African countries now enacting data protection laws, data centers are increasingly positioned as critical national infrastructure. The legal framework is catching up. The physical infrastructure is being built. The financing models exist.
What is missing is the coordinated, continental-level urgency that this moment demands. The talent leaving Africa's universities and the compute sitting in European data centers are two expressions of the same underlying failure: a continent that keeps producing the inputs for someone else's economic engine.
Kasi LOS1 in Lagos is proof that it can be different. The Microsoft-G42 stall in Kenya is proof of what happens when the physical prerequisites are not in place.
The Stack Sovereignty Test does not score effort. It scores outcomes. On Compute, Africa is not failing - but it is not yet passing either.
The machines that Africa needs are not unaffordable. They are not unfinanceable. They are not un-buildable.
They are simply not yet built.
Next week, we turn to the third pillar of the Stack Sovereignty Test: Chips. If Compute asks who owns the machines, Chips asks who controls the supply chains that make the machines possible - and why Africa's answer to that question may determine the continent's position in the AI era for the next fifty years
Oluwaseyi Ayodeji is an AI infrastructure strategist with 17+ years across AI hardware supply chains, semiconductor manufacturing operations, and workforce development. He writes the Sovereign Stack newsletter at oluwaseyiayodeji.com.
