AI in Africa

This essay was first published at HAKI.

Conventional narratives position Africa as perpetually behind, always scrambling to catch up while real progress happens elsewhere. This framing is wrong. Africa’s position outside conventional currents of success need not always be a disadvantage. Sometimes, it is a strategic asset to be deployed.

Africa is not and should not be racing to implement policies designed in Boston or Beijing. The Swahili people say “Kutangulia sio kufika“: departing first is not arriving. Sometimes lateness allows you to choose differently, to choose better.

We have done this before. In mobile technology, Africa leapfrogged landline infrastructure entirely and built mobile-first systems that now serve as models globally. When regions lack entrenched legacy systems, they can adopt newer technologies directly without the costly transitions that incumbents face. Africa’s position amplifies this advantage, allowing individual countries to experiment without great cost.

The same logic applies in AI, only more so. As we debate graphs that draw attention to the absence of massive data centres across the African continent, we should ask the right questions. Not whether Africa will build some number of AI data centres, but what kind of AI Africa needs to build and deploy.

The answer depends on recognising what Africa has that other regions do not. The dominant AI paradigm assumes value comes from scale: training larger models on larger datasets scraped from the homogeneous digital commons of the global internet. This produces models fluent in English and Mandarin but useless in Kikuyu or Sheng. Models that know everything about American suburbs and nothing about Rongai.

Africa possesses diversity at a scale and depth vanishing everywhere else. Much of the continent has not been homogenised by the hegemonic forces that flattened regional differences in North America and Europe. Extraordinary variation still persists: over two thousand spoken languages, radically different agricultural practices adapted to microclimates, traditional medical knowledge that varies between ethnic groups, construction techniques responsive to local materials, social structures that resist western categorisation, and much more.

Africa’s polycentric structure has preserved this diversity. Where centralised states impose uniformity through standardised education, industrial policy, and cultural homogenisation, Africa’s feeble and fragmented governance has allowed rural knowledge systems to persist. The absence of strong central states that can impose standards across every inch of their territoty has inadvertently protected precisely the diversity that makes Africa valuable for AI development. A Senegalese farming community and a Tanzanian one facing drought may have developed entirely different drought-resistant crop varieties, neither appearing in global agricultural databases dominated by American industrial monoculture. A traditional healer in Malawi may possess knowledge about medicinal plants that a healer in Ghana does not, and both may know things completely absent from The Lancet.

Africa’s diversity, which has not yet been uploaded to the global internet, represents knowledge that could train fundamentally different AI systems. Systems that handle deep heterogeneity. Systems that represent local, contextual, embodied knowledge. Multimodal systems that learn from small, disparate and diverse datasets rather than massive, uniform ones.

The assumption that Africa must build AI the way Silicon Valley or Shenzhen did collapses under scrutiny. The race to train ever-larger foundation models is expensive, energy-intensive, environmentally destructive, and dependent on hardware supply chains that Africa does not control. Most high-impact problems Africa needs to solve do not require frontier models with billions of parameters. Optimising smallholder farming logistics does not need a GPT-6. Diagnosing common diseases from symptoms reported in local languages does not need massive compute infrastructure.

What these problems need is intelligent system design: the capacity to formulate problems clearly, to identify which data matters, to build small specialised models that solve specific problems efficiently, and to deploy them where needed, often in contexts with limited internet connectivity and computational resources. The bottleneck for Africa does not show up on fancy graphs because ideas, not infrastructure, constrain us.

Ideas do not require data centres. But they do require electricity to run them, connectivity to deploy them, and reliable compute to execute them. The challenge is building distributed, resilient infrastructure alongside smart design, not choosing between them. Ideas require people who understand both the technical possibilities of AI and the practical realities of African contexts. People who can translate between the abstract language of machine learning and the concrete needs of farmers, health workers, urban planners, and small business owners.

AI has also become more than a tool that large institutions deploy, it is now a capability that individuals access directly, sometimes circumventing institutional gatekeepers entirely. A Kenyan university student can use DeepSeek to learn advanced mathematics, debug code, and explore research questions without waiting for professors to answer emails or universities to update curricula. A Ghanaian entrepreneur can use locally hosted tools to analyse market data, generate business plans, build front-end systems, and even prototype products without hiring expensive services.

This is happening right now. What once required proximity to elite institutions is increasingly available to anyone with knowledge of how to use digital tools effectively. This involves knowing which tools exist, understanding what they can do, learning how to use them effectively, and developing feel for when to trust their outputs and when to doubt them.

The amplification of individual cognitive capacity depends on having these skills to use AI tools that are now becoming free and ubiquitous. When millions of students in Uganda can learn as effectively as students in the UK and prototype ideas as quickly as their peers, the constraints on African development shift from access and capability to execution.

The asymmetry between those who know how to leverage AI and those who do not already exceeds many traditional educational and economic divides of yore. And as the technology advances, this gap will widen. However, these are learnable skills. They spread horizontally, person to person, and community to community, far faster than institutional change happens, and without the involvement of any government. This means no one can gate-keep it, and innovation can emerge simultaneously from Lagos, Kigali, Nairobi, Accra, Johannesburg without requiring permission from any central authority.

Yet individual capability means little if the regulatory environment makes it impossible to start a business, if infrastructure makes it impossible to scale, if capital markets make it impossible to fund innovation, if political systems make it impossible to challenge incumbent interests. AI does not solve these problems, at least not yet.

Let’s consider how AI policy actually forms. The standard model imagines hierarchy: global bodies set the principles, national governments translate them into regulations, corporations implement them, individuals comply. That is a fiction. AI policy emerges simultaneously across overlapping spheres that neither align nor defer to each other’s authority. The EU passes the AI Act while Silicon Valley ships products that ignore it. Israel uses AI systems to conduct its campaigns against Palestine while the UN debates human rights frameworks. Google and OpenAI set their de facto standards through platforms, while individuals craft their own practical policies through daily choices about which models to use, which ones to boycott, and which companies to distrust.

This is polycentrism in practise, and Africa mirrors this structure internally. Competing regional blocs, radically different governance systems, divergent economic models. Kenya’s regulatory approach differs from Nigeria’s which differs from South Africa’s, just as the EU’s differs from China’s and differs from California’s. The East African Community coordinates on some standards while COMESA pursues others. Individual African nations cut bilateral deals with foreign powers even as the African Union declares continental principles. It all looks like dysfunction, yet it is creating multiplied sites of experimentation and is being driven by strategic choice.

When policy emerges from multiple centres, late entry means observing which approaches fail before committing. You learn from the EU’s regulatory overreach that stifles innovation and from California’s regulatory absence that enables exploitation. You study China’s surveillance infrastructure and reject it. You watch which corporate strategies succeed and which ones don’t. You take notes and move with information, failing only where you must, where failing is worth it.

Regional bodies like the East African Community have a role that national governments cannot play alone. AI does not respect borders. Data flows across jurisdictions. Talent moves in search of opportunity. Markets are regional. A Kenyan startup building legal AI tools in Swahili needs users in Tanzania to achieve scale. A Nigerian company developing health diagnostics needs regulatory approval across West Africa to make the investment worthwhile. Fragmented national approaches create friction that benefits no one except foreign entities that can afford to navigate fifty-four different regulatory regimes.

Regional coordination need not mean uniformity. Different countries will and should make different choices about data governance, about which AI applications to encourage or restrict, about how to balance innovation and risk. Polycentrism is a feature, not a bug. But coordination means recognising that some standards need to align: interoperability protocols, data sharing frameworks, mutual recognition of certifications, pooling resources for AI research, coordinating training programmes so talent can move freely within the region, harmonised approaches to cross-border data flows, and even building shared infrastructure.

This is necessity, not idealism. No African country has the market size, technical capacity, or financial resources to compete with AI ecosystems in the United States, China, or the European Union. But collectively, Africa does. Will we coordinate effectively enough to leverage our collective capacity or will we remain fragmented, each country cutting its own deals with foreign powers, then getting screwed in perpetuity? The answer lies in understanding that coordination within a polycentre does not require destroying the polycentre. It requires strategic alignment on specific issues while preserving autonomy on others.

The corporate layer of AI policy complicates this. Google, Microsoft, OpenAI, and their competitors are not passive recipients of government regulation. They shape the infrastructure on which most African AI development currently depends. They control the APIs, the platforms, the foundational models, the cloud computing resources. They set the terms of service, the pricing structures, the data policies. They decide which features to build and which markets to serve. Their decisions matter more for practical AI development in Africa than most national and regional AI strategies. But Africa’s polycentric structure creates leverage. When corporations negotiate with multiple governments, they face different abilities to impose terms than when negotiating with a single centralised authority. This can be a disadvantage or an advantage depending on whether African countries can coordinate strategically.

Africans must also think about which dependencies to accept and which to resist. Our institutions and developers will use cloud services, will fine-tune existing models, will build on platforms created elsewhere. The question is whether we do this as strategic users who are also building our own capabilities, or as permanent dependents with no alternatives.

Open-source AI development matters here. When foundational models, training frameworks, and deployment tools are open and freely available, the dependencies shift. You can build on Meta’s Llama models without being locked into Meta’s ecosystem. You can use open-source frameworks and deploy them on your own infrastructure or on competitive cloud providers. You can fork, modify, customise technologies without asking permission. Compute still costs money, hardware still comes from specific supply chains, but the nature of dependency changes from lock-in to choice.

Africa should invest heavily in open-source AI development, both as contributors to global projects and as developers of open-source tools themselves. Every African developer who contributes to open-source projects builds capability that cannot be taken away by corporate policy changes. Every open-source model trained on African data and optimised for African use cases is infrastructure that persists regardless of what happens to any individual company. Every open standard that African institutions help define is a constraint on corporate power and an enabler of African autonomy. It all creates a distributed ecosystem more resilient than any centralised alternative.

The global and multilateral layer of AI policy is where rhetorical battles happen, where broad principles get declared, where moral posturing is most common and practical impact is often minimal. Various international conferences produce frameworks and guidelines, but these tend to align with what powerful actors already intended to do. Africa should participate in these forums, but with no illusions that global consensus will solve African problems.

The value lies in the coalitions these forums enable rather than in the consensus documents they produce. When African countries coordinate with each other and with other regions facing similar challenges, such as Latin America and Southeast Asia, they can shift debates that would otherwise be dominated entirely by the US-China-Russia axis. They can resist frameworks that treat AI governance as a political problem involving choices about political powers. They can insist that AI ethics means something different in contexts of deep inequality and colonial legacy than it does in wealthy democracies with strong institutions.

The real work still happens elsewhere. It happens in the choices African governments make about data infrastructure and digital public goods. It happens in the decisions African universities make about curriculum and research priorities. It happens in the bets African entrepreneurs make about which problems to solve and which markets to serve. It happens in the daily practices of African individuals learning to use AI tools effectively and teaching others to do the same. It happens sometimes in competition, sometimes in coordination, and sometimes in indifference to what others are doing.

When policy emerges from multiple centres, you do not need to win at every level to make progress. You can build capability at an individual level when national policy is incoherent. You can coordinate regionally when global forums are deadlocked. You can choose different dependencies for different purposes, playing corporations against each other rather than being played by them.

Africa’s strategic position in AI is therefore one of optionality rather than of strength or weakness in conventional terms. We can learn from others’ mistakes without repeating them. We can empower individuals rather than waiting for institutions to catch up. The polycentre provides options.

But none of this is guaranteed. Whether Africa converts this position into actual advantage depends on the choices we make in the next few years, while AI infrastructure and norms are still being built rather than already locked in. This opportunity is real, but it is not permanent.

The work of HAKI begins here: building the tools, training the people, demonstrating the possibilities, proving that African AI developed by Africans for African contexts can be technically excellent and economically viable. Not as a political statement but as a practical necessity. Not in opposition to global AI development but as a distinctive contribution to it. Not waiting for permission, not waiting for consensus, but building the future we need, right now. Because in a polycentre, any node can catalyse change across the network.