I grew up in Nigeria watching a country consistently arrive at the technological frontier after everyone else had already set the terms. We got mobile phones after the infrastructure assumptions were locked in. We got the internet after the platform economics were decided. We got social media after the attention economy's playbook was written — by others, for others.

Each time, the lag wasn't purely about access. It was about who held the conceptual framework when the technology was still plastic. The people in the room during the early, ambiguous, definitional phase of a technology are the ones who shape it. Everyone who arrives later inherits their assumptions.

AI is in that phase right now. And the window is closing.


What "Being Late" Actually Means

Late doesn't mean not having AI. Africa will have AI. The models will arrive — they already are. What late means is building on top of a foundation that was designed without you in mind, optimizing for problems you didn't have, trained on data that underrepresents your context, and governed by frameworks that treat your interests as an afterthought.

That's a different kind of disadvantage than not having the technology at all. It's subtler, harder to name, and more durable.

Consider language models. The dominant ones are trained overwhelmingly on English text, with token-level representations that handle African languages poorly. Yoruba, Igbo, Hausa, Amharic, Swahili — these are not edge cases. They are the native languages of hundreds of millions of people. A medical AI deployed in Lagos that reasons well in English and poorly in Yoruba is not a neutral tool. It embeds an assumption about whose knowledge matters.

This is not an argument against using AI. It's an argument for building it, shaping it, and doing that work now rather than in fifteen years when the sediment has hardened.


Three Things That Need to Happen

Infrastructure comes first, but not just power

The conversation about AI in Africa often starts and ends with power and connectivity. These matter — you cannot run inference on unstable electricity. But the infrastructure gap that worries me more is compute access for training and research.

A researcher at a Nigerian university who wants to train a model on local language data has, in practice, almost no path to the GPU resources needed. Cloud credits help at the margins. They don't solve the structural problem. West Africa needs at least one high-performance computing facility with serious GPU capacity, institutionally supported and accessible to academics. One. That's not an outrageous ask.

Education that produces builders, not just users

AI literacy programs in Africa are, by and large, teaching people to use tools that others built. That's valuable — but it's the floor, not the ceiling. The distinction I keep coming back to: literacy vs. fluency.

Literacy is knowing what a large language model is and how to prompt it effectively. Fluency is understanding the training dynamics, the data curation decisions, the evaluation pitfalls — the knowledge that lets you question the tool, adapt it, build something new on top of it.

We need programs that produce fluent practitioners. That means graduate training in ML that is genuinely research-grade, not just applied workshops. It means connecting that training to real problems — African healthcare data, African language corpora, African agricultural systems — so the research has gravity and context.

Policy that is proactive, not reactive

African governments have a long history of regulating technology after its harms have already materialized. The GDPR equivalent arrives years after the data extraction. The platform accountability framework comes after the radicalization pipeline is well-established.

AI governance is happening right now. The EU AI Act, the US executive orders, the emerging ISO standards — the rules are being written. African nations are largely absent from those conversations. That absence is not passive. It has consequences: the defaults being set will apply globally, and they are being set by people optimizing for different risk profiles and different values.

There are smart people in African policy circles working on this. They need more institutional support, more coordination across the continent, and more direct engagement with the international frameworks being built.


Why This Is Personal

I'm doing a PhD in the United States, working on AI for scientific computing. The work is good. The environment is good. I'm grateful for it.

But I think about it constantly: the fact that the path to doing serious AI research, for someone from Nigeria, almost necessarily runs through institutions in the Global North. That the mental models I'm developing, the collaborators I'm meeting, the problems I'm choosing — all of this is being shaped by a context that is not the one I came from.

I don't think that's irreversible. But I think the window where it can be reversed — where African institutions can become genuine generators of AI knowledge rather than consumers of it — is shorter than people realize.

The moment AI becomes infrastructure, the way electricity or the internet did, the opportunity to shape its foundations closes. We are not there yet. We are close.

This is the thing I want to spend some part of my energy on for the rest of my career. Not because it's strategically optimal. Because it matters, and because I'm in a position where I can see both sides of the gap clearly.

That tends to come with a responsibility.