Hans van Linschoten: Small, Smaller, Smallest

Boston Consulting Group published an article this month, authored by Hamid Maher, Patrick Dupoux and colleagues out of its Casablanca and Paris offices, setting out three priorities for Africa’s digital and AI economy: build the infrastructure and data foundations through public-private partnership, mutualise investment to reach viable scale, and pursue open architectures to grow local ecosystems. It is the clearest articulation I have read of a problem we have been building against for several years.

Their diagnosis is hard to argue with. Africa holds roughly eighteen percent of the world’s population and under one percent of its data centre capacity. Fifty-four national markets, none individually large enough in GDP terms to carry the fixed cost of modern digital infrastructure. A developer base growing faster than anywhere on earth, and being recruited away faster still. And a persistent premium – BCG puts it as high as thirty-five percent – on the same software and hardware that firms elsewhere buy cheaper.

Rather than restate the argument, it is more useful to say where we sit against it: where we are executing the same thesis, where we think the framework stops one layer too high, where it under-weights two of the five actors who actually build ecosystems, and where it makes uncomfortable reading for us.

The leapfrog has always been a question of unit size

Every genuine African leapfrog has one thing in common, and it is not the technology. Mobile overtook copper because a handset and a prepaid scratch card could be bought one at a time. M-Pesa overtook branch banking because an agent needed a float and a phone, not a vault. Solar home systems overtook grid extension because a household could buy fifty watts this month and another fifty next year. In each case the unit of deployment was small enough that no one had to approve the whole system before anyone could start.

Digital infrastructure has so far been imported at the wrong unit size. The reference design is the hyperscale region: tens of megawatts, hundreds of millions of dollars, a multi-year procurement, a sovereign counterparty and a demand forecast that has to be right at the outset. That unit works in markets that already have the demand. In markets that are still creating it, it is the reason projects stall between announcement and commissioning.

So the strategic question is not how to afford the big thing. It is how small the unit can be made before the economics break.
Small is the national node – sovereign capacity inside one jurisdiction, sized to that market rather than to a regional average.

Smaller is what we build against: a minimum viable node, roughly $339,000 of capital expenditure, deployable in a single market and expandable in place. Smallest is edge capacity attached to infrastructure that already exists – a telco exchange, an ISP facility, a university campus – where the site, power and connectivity are somebody else’s sunk cost.

The aggregate matters more than any single unit. Our modelling for a twenty-node rollout across ten markets returns an internal rate of return of approximately forty-one percent over five years. What makes that figure interesting is not its size but how it is assembled: by addition, one node at a time, each one proving demand before the next is financed. There is no single approval gate, no sovereign guarantee, and no scenario in which one bad market strands the whole programme. Failure is contained to a node. Success compounds across the federation.

This is where we would gently invert BCG’s second priority. Mutualisation, as they frame it, is about pooling resources in order to afford something large. The alternative is to make the thing small enough that pooling is not a precondition – and then federate the units, so the platform economics arrive anyway. Federation is mutualisation achieved commercially rather than diplomatically. It does not require fifty-four governments to agree on anything first.

The framework starts one layer too high

BCG’s first priority takes identity as the base layer, with payments and data exchange built on top. Aadhaar and Rwanda’s IremboGov are the reference cases, and they are good ones.

But identity is not the base layer. It runs on something. If a national identity register, a health record system or a tax platform is sovereign in law and hosted on infrastructure owned, operated and ultimately controlled outside the continent, the sovereignty is contractual rather than structural – and contracts are revised, jurisdictions reach across borders, and pricing is set elsewhere. The compute, storage and control plane beneath digital public infrastructure is where dependency is actually created or avoided.

This is the same substitution BCG warns about in its own framing: instead of minerals leaving the continent as raw material, it is data. The distinction we would add is that the substitution is settled at the infrastructure layer, before a single dataset is collected.

Five stakeholders, and the two Africa keeps under-funding

MIT’s stakeholder framework for building innovation ecosystems, developed by Phil Budden and Fiona Murray, is useful here precisely because it goes beyond the familiar triple helix. It identifies five actors: government, large corporates, universities, entrepreneurs, and risk capital. The evidence across ecosystems is that all five have to be present, and that the last two are the ones most often assumed rather than built.

Applied to sovereign digital infrastructure in Africa, the roles are reasonably clear:

• Government. Sets data classification and residency rules, mandates open interfaces in procurement, and acts as anchor tenant. BCG covers this ground thoroughly and correctly.
• Corporates. Telcos, banks and utilities are the demand aggregators, and in the smallest-unit model they are also the hosts – the exchange, the meet-me room, the existing power envelope.
• Universities. The talent pipeline, but also a deployment site. Campus edge capacity gives graduates something to operate before they are recruited abroad; it converts the brain-drain problem into an infrastructure decision.
• Entrepreneurs. In our model this is the virtual cloud operator layer – locally owned companies taking platform capability to their own market, carrying commercial risk and keeping the margin. It is where AfriQloud actually lives, and it is the stakeholder most often designed out of continental infrastructure plans.
• Risk capital. The weakest link. Development finance is indispensable, but concessional debt and blended structures are not risk capital; they price differently, move slower, and rarely fund the first node. Genuine equity risk appetite is what small unit size unlocks – a $339,000 node is a decision a regional investor can actually take.

The candid observation is that BCG’s framework leans heavily on government, corporates and development finance. Entrepreneurs and risk capital appear mostly as beneficiaries of an environment created by others. On MIT’s evidence, those are the two that accelerate ecosystems rather than merely enable them – and they are the two that a small unit of deployment brings into play.

Open code does not run itself – and this is where we sit uneasily

The most important sentence in BCG’s third priority is also its most easily overlooked: an open system still needs an operator, and that operator may be commercial, provided the protocol layer stays open and portable. OpenMRS and MOSIP are genuine public goods; neither of them runs a national cloud.

I should be straightforward about where this leaves us. AfriQloud runs on a licensed platform stack, whitesky.cloud, not on open-source code. Against BCG’s third priority as literally written, that is a mismatch, and it deserves an honest answer rather than a redefinition.

Our answer is that the harms BCG attributes to proprietary systems are the ones worth engineering against, and they are separable from the licence model. Vendor lock-in is a function of portability, not of source availability: open interfaces, standard formats, no egress penalty, no data held hostage. Value extraction is a function of ownership, not of code: the operating companies are locally held and locally profitable. Talent drain is a function of what local engineers are permitted to build on: white-label capability, a certification path, and technical depth resident inside African firms rather than inside a foreign vendor’s support queue.

Where the mismatch remains real is at the protocol layer. If the interfaces are not genuinely portable, the model reproduces the dependency it claims to solve, with a friendlier accent. That is a legitimate test, and one we should be held to publicly rather than asked to be trusted on.

Ownership rules cut both ways

BCG notes the proposed continental framework requiring at least fifty-one percent African ownership for preferential market access. We designed for that outcome before it was drafted, because it is the correct outcome: local majority ownership at operating-company level, with governance protections negotiated at shareholder level, and AfriQloud Uganda as the first working instance. But the rule does not resolve the tension it creates. Capital-intensive infrastructure needs investors, and investors price control. Local majority ownership with real governance is workable. Local majority ownership as a compliance wrapper, with the economics routed elsewhere, is the failure mode – and it will be common.

What would make me wrong

Three things, honestly stated.

• Demand aggregation may not materialise. The federation case assumes enterprise and public-sector workloads move to regional platforms at a pace the market has not yet demonstrated. If they do not, small units simply strand capital in smaller pieces.
• Power and connectivity remain the binding constraint. No control plane compensates for an unreliable grid or expensive transit. In several markets the infrastructure question sits upstream of the digital one.
• Hyperscaler regional expansion may outrun the argument. If in-country regions with credible local-control guarantees arrive at scale and at price, the structural case narrows to jurisdiction and procurement policy alone.

The point of agreement that matters most

BCG closes by observing that digital transformation almost never fails for technical reasons – it fails on alignment, incentives and accountability. That is the sentence I would put in front of every ministry, operator and investor working on this.

Africa’s digital future will not be decided by how much technology the continent adopts. It will be decided by who owns the layer everything else runs on, who is accountable when it breaks, and where the margin lands. Start small enough that those questions can be answered one market at a time, and the continental answer assembles itself.

Hans van Linschoten is the CEO of whitesky.cloud, a European sovereign cloud platform, and Co-founder of AfriQloud, a pan-African sovereign cloud and edge AI infrastructure initiative.

Follow AfriQloud on LinkedIn

Follow whitesky.cloud on LinkedIn