The AI Leapfrog Is a Myth

Why Africa Cannot Skip the Infrastructure That Makes Artificial Intelligence Work

The leapfrog narrative is one of the most seductive stories in African technology. It goes like this: Africa skipped landlines and went straight to mobile phones. Africa skipped bank branches and went straight to mobile money. Therefore, Africa will skip the intermediary stages of technological development and leap directly to the frontier. The narrative has been applied to clean energy, to financial services, to healthcare. And now it is being applied to artificial intelligence.

The AI leapfrog thesis argues that Africa can bypass the decades of institutional and infrastructure development that underpins AI capability in advanced economies and move directly to deploying artificial intelligence at scale. The thesis is appealing. It is also, in its current form, dangerously wrong.

What the Mobile Money Leapfrog Actually Was

To understand why the AI leapfrog is a myth, it is necessary to understand what the mobile money leapfrog actually was — and was not. M-Pesa did not emerge from a vacuum. It emerged from a specific set of conditions: near-universal mobile phone penetration, a regulatory environment that permitted non-bank financial services, a telecommunications company with the distribution infrastructure to support agent networks, and a population with an acute unmet need for basic financial transactions.

Critically, the mobile money leapfrog did not skip infrastructure. It substituted one form of infrastructure (bank branches, physical payment networks) for another (mobile networks, agent distribution systems). The infrastructure was different, but it was infrastructure nonetheless. M-Pesa required Safaricom's network coverage, tens of thousands of physical agents, regulatory permissions that took years to negotiate, and technology systems that processed transactions reliably at scale.

The leapfrog narrative mischaracterises what happened. It was not a skip. It was a substitution — replacing expensive, legacy infrastructure with cheaper, more appropriate infrastructure that could be deployed at scale given the continent's specific conditions. This distinction matters enormously for AI, because the infrastructure that AI requires cannot be easily substituted.

The Infrastructure AI Actually Requires

Artificial intelligence is not a product. It is a capability that sits atop multiple layers of infrastructure, each of which must function for the capability to be useful. These layers include compute infrastructure (data centres with GPU capacity), data infrastructure (clean, labelled, representative datasets), connectivity infrastructure (reliable high-bandwidth networks), energy infrastructure (consistent power supply), and human infrastructure (engineers, data scientists, domain experts who can build and deploy AI systems).

Africa's position across each of these layers is stark. The continent hosts less than one percent of the world's data centres. Only about three percent of global AI talent is based in Africa. One in two Africans does not have reliable electricity — the most basic prerequisite for any computing infrastructure. International bandwidth costs remain multiples of what is available in developed markets. And the datasets that exist are overwhelmingly generated by and for populations in North America, Europe, and East Asia, meaning that AI models trained on global data perform poorly on African contexts.

These are not gaps that can be leapfrogged. They are prerequisites that must be built. You cannot run GPU clusters without reliable electricity. You cannot train locally relevant AI models without locally representative data. You cannot deploy AI systems without engineers who understand both the technology and the context in which it will operate. You cannot serve AI applications without connectivity that can deliver model outputs to end users.

The Hyperscale Trap

The infrastructure gap has attracted significant attention from global technology companies and development institutions. The response has been to propose hyperscale solutions: massive data centres, continental AI strategies, billion-dollar infrastructure investments. The African Union has declared AI a strategic priority. Hyperscale cloud providers are exploring African data centre locations. Development finance institutions are mobilising capital for AI infrastructure.

But the hyperscale approach carries its own risks. Hyperscale data centres designed for North American or European contexts require hundreds of megawatts of continuous power — far exceeding available grid capacity in most African countries. The result is costly infrastructure with low utilisation and limited spillover into local innovation ecosystems. Many existing data centre operators already report underutilised capacity, with the constraint shifting from physical capacity to uncertain workload demand and the high cost of GPU access.

This is the hyperscale trap: building infrastructure designed for the scale and conditions of other markets, then discovering that it does not match the scale and conditions of the market in which it is deployed. A 100-megawatt data centre in Lagos is not a smaller version of a 100-megawatt data centre in Virginia. It is a fundamentally different proposition, facing different power constraints, different demand profiles, different cost structures, and different use cases.

The alternative — modular, distributed infrastructure matched to actual demand and available power — is less glamorous but more appropriate. Containerised GPU clusters that can be deployed incrementally as demand develops. Edge computing infrastructure that pushes AI inference closer to users rather than concentrating it in distant data centres. Smaller-scale training facilities optimised for the specific models and datasets that African use cases require, rather than general-purpose compute designed for frontier model training that will never happen on the continent.

The Data Problem

Even if compute infrastructure were available at scale, Africa faces a data problem that cannot be solved by hardware. AI systems are only as useful as the data on which they are trained and the data on which they operate. The global AI ecosystem is built on datasets that overwhelmingly represent the developed world — English-language text, European and North American imagery, financial data from mature markets, healthcare data from well-resourced institutions.

These datasets produce models that perform poorly in African contexts. Natural language processing models that work well in English struggle with the over 2,000 languages spoken across the continent. Computer vision models trained on North American or European data misidentify African environments and populations. Financial models calibrated to developed-market conditions produce irrelevant predictions for economies structured around informality and mobile money.

Building Africa-relevant AI requires Africa-generated data — in African languages, reflecting African economic structures, representing African populations and environments. This data does not exist at the scale required for modern AI systems. Creating it requires investment not in GPU clusters but in the unglamorous work of data collection, labelling, cleaning, and governance across dozens of countries, hundreds of languages, and thousands of distinct economic and social contexts.

This is not a problem that leapfrogging solves. It is a problem that requires sustained, patient investment in data infrastructure — the kind of investment that does not attract venture capital attention because it does not produce the exponential growth curves that the industry rewards.

The Talent Bottleneck

The scarcest resource in Africa's AI development is not compute or capital. It is talent. With approximately three percent of global AI talent based on the continent, the human infrastructure required to build, deploy, and maintain AI systems is profoundly thin.

This is not merely a quantity problem. It is a depth problem. Building AI applications requires not just data scientists who can train models but domain experts who understand the contexts in which models will be deployed, engineers who can build production systems around model outputs, and product managers who can translate between technical capability and user need. This multi-disciplinary talent stack does not exist at scale in any African market.

Training this talent takes time — years, not months. And it requires institutional foundations that are themselves underdeveloped: universities with AI research programmes, companies with AI engineering teams that can mentor junior talent, and an ecosystem of practice that allows knowledge to compound across organisations rather than remaining locked in individual companies.

The leapfrog narrative implicitly assumes that talent can be acquired or imported at the speed required. It cannot. Talent development is inherently a compounding process that cannot be accelerated beyond certain limits. Africa can and should invest aggressively in AI education and training. But expecting this investment to produce the talent density required for transformative AI deployment within a few years is not ambition. It is delusion.

What Can Actually Be Done

Rejecting the leapfrog myth does not mean rejecting AI in Africa. It means rejecting the fantasy that AI can be deployed in Africa the same way it is deployed in markets with fundamentally different infrastructure endowments. The opportunity is real. The path to capturing it is different from what the leapfrog narrative suggests.

First, investment must flow to the boring infrastructure that makes AI possible — not GPU clusters for frontier model training, but reliable electricity, affordable connectivity, and data centre capacity matched to actual demand. Without this foundation, AI deployment is not just inefficient. It is impossible.

Second, data infrastructure must be treated as a strategic priority. This means investing in the collection, labelling, and governance of Africa-relevant datasets across languages, sectors, and geographies. It means building the institutional frameworks — data trusts, governance standards, privacy regulations — that allow data to be shared and used responsibly. And it means recognising that data infrastructure is not a one-time investment but an ongoing process that requires sustained funding and institutional commitment.

Third, AI deployment in Africa should prioritise applications that work within existing infrastructure constraints rather than assuming those constraints will be solved. Edge AI, small language models, offline-capable inference, and applications that deliver value at low bandwidth and intermittent connectivity — these are not compromises. They are the appropriate technological choices for the environment.

Fourth, talent development must be treated as the binding constraint it is, with investment timelines measured in decades rather than quarters. This means funding AI research at African universities, supporting the development of AI engineering teams within African companies, and creating the institutional conditions for a self-sustaining AI talent ecosystem to emerge.

The Honest Assessment

The AI leapfrog narrative serves a purpose. It attracts attention, mobilises capital, and creates urgency around an important technological transition. But it also creates expectations that cannot be met, strategies that cannot be executed, and disappointment that undermines confidence in the genuine opportunity that AI represents for African markets.

The honest assessment is this: Africa will not leapfrog into AI leadership. It will build its way there — slowly, incrementally, through the unglamorous work of constructing the infrastructure, generating the data, and developing the talent that AI requires. This path is less exciting than the leapfrog narrative promises. It is also the only path that leads anywhere real.

The mobile phone revolution that the leapfrog narrative celebrates did not happen overnight. It took two decades of investment in telecommunications infrastructure, regulatory reform, and ecosystem development. The AI opportunity in Africa will follow a similar trajectory — real, significant, and transformative, but measured in decades rather than years, and built on infrastructure rather than aspirations.

Those who understand this will build for the long term and capture the genuine opportunity. Those who believe the leapfrog myth will invest in castles built on sand and wonder why the foundation keeps shifting.