70% of Nigerians Use AI. Almost None of It Is Built Here.

More than 70 percent of Nigerians have used a generative AI tool. Ninety-three percent of organisations surveyed have started deploying AI, and close to a third are already running it in advanced business operations.

By any reasonable measure, that is fast adoption. It is also, according to a report published today, exactly the problem.

The study, titled Adoption of Artificial Intelligence in Nigeria: A Macro and Micro Economic Review, was authored by Debola Ibiyode, founder of the AI Empowerment Foundation and chief executive of CarbonAI. It was conducted by OLGNova in collaboration with AI in Action Now. Its central argument is that Nigeria is consuming artificial intelligence considerably faster than it is building the institutions, infrastructure and datasets required to own any of it.

Nigeria is not failing at AI. It is adopting AI faster than it is building the institutional, infrastructural, and human foundations needed to sustain and own that adoption.

What AI Is Already Doing in Nigeria

The report is not a warning dressed up as analysis. It documents real deployment across sectors, and the specifics are more advanced than the general conversation suggests.

In government, the Federal Civil Service has deployed Service Wise GPT. AI-powered traffic management pilots in Lagos and Abuja have cut peak-hour travel times by roughly 20 percent, which in Lagos terms is a meaningful return of hours to people's lives.

In healthcare, diagnostic platforms are helping doctors read radiological images and identify birth asphyxia in newborns. AI chatbots are being used to extend mental health support in a country with a severe shortage of psychiatrists. In agriculture, advisory platforms are lifting crop yields by 20 to 30 percent while lowering production costs, and route optimisation is reducing food waste in supply chains.

Nigerian research output is also substantial. Local scholars have produced more than 11,600 AI-related academic publications, with the University of Ibadan leading the country.

This is not a country waiting for AI to arrive. It has arrived, and it is working.

The Ownership Gap Is Where the Value Leaks

Here is the uncomfortable part. Nearly all of that activity runs on foreign-built models, foreign cloud infrastructure and foreign software platforms. Nigeria contributes very little to foundational model development, indigenous datasets, or domestic computing capacity.

That arrangement has three consequences the report spells out clearly.

Nigerian data improves systems Nigeria does not own. Every interaction with a foreign model is a small contribution to a product whose value accrues elsewhere. The country is, in effect, paying to train someone else's asset.

Critical services sit behind decisions made abroad. Licensing terms can change. Services can be withdrawn or repriced. A hospital diagnostic tool or a civil service assistant built entirely on an external platform is one commercial decision away from disruption, and nobody in Abuja gets a vote.

And the models themselves do not speak Nigeria. Leading systems are trained overwhelmingly on English-language data, which leaves them weak in Hausa, Yoruba and Igbo. For a country where a majority of people conduct daily life in a language these models handle poorly, that is not a minor gap. It is a ceiling on who AI can actually serve.

Nine Million Jobs Out, Eleven Million In

The report projects that automation could displace around nine million routine jobs by 2030, concentrated in banking, clerical work and public administration. It also estimates that AI could create roughly eleven million technology-enabled roles over the same period.

A net gain of two million jobs reads like good news. Read the condition attached to it and it becomes something else. That upside is contingent on heavy investment in digital skills and workforce reskilling. Without it, the nine million still go, and the eleven million either do not materialise or get filled by people who were never in the displaced group.

Job displacement and job creation are not the same population. A bank clerk in Kano does not become a machine learning engineer because a projection says the roles exist. The transition is the policy, and it is the part nobody has costed.

What Sovereign AI Would Actually Require

The report's recommendation is what it calls sovereign AI: local infrastructure, indigenous language datasets, innovation hubs, and affordable technology financing. It also asks universities to train students to build models rather than simply operate them.

The direction is right. The honest question is what it costs, and here the report is more optimistic than the underlying economics support.

Training a competitive foundation model requires compute at a scale that consumes electricity Nigeria does not reliably generate, in data centres it has not yet built, using chips that are export-controlled and expensive. Matching frontier labs is not a realistic national objective, and framing it as one invites the wrong investments.

The achievable version of sovereignty is narrower and more valuable. Indigenous language datasets for Hausa, Yoruba and Igbo are a genuine strategic asset, because no foreign lab has a commercial reason to build them well and whoever does will own that layer. The same applies to Nigerian health records, agricultural data, and payment behaviour, all of which are moats that compute cannot buy.

Sovereignty over the data and the application layer is achievable this decade. Sovereignty over the model layer is not, and pretending otherwise wastes scarce capital.

A Note on the Adoption Numbers

The 70 percent figure sits awkwardly beside other measurements. The 2026 Microsoft AI Diffusion Report ranks Nigeria 110th of 147 countries, with an adoption rate closer to 10 percent.

Both can be true because they measure different things. Having opened ChatGPT once is not the same as AI being embedded across an economy's workflows. The gap between those two numbers is itself the story: broad casual contact, shallow institutional integration. Readers should treat any single adoption statistic, including this one, with appropriate suspicion about what exactly was counted.

What This Means for People Building in Nigeria

For founders, the report is effectively a map of underpriced opportunity. The gaps it identifies as national weaknesses are the same gaps that make defensible businesses: language tooling, sector-specific datasets, and applications built for infrastructure that fails intermittently.

For policymakers, the useful takeaway is to stop measuring adoption and start measuring ownership. How much Nigerian data sits in Nigerian systems. How many production models were trained locally. How many AI companies here sell abroad rather than only at home. Those numbers would tell you something the adoption rate never will.

For everyone else, the framing worth keeping is simple. Adoption is a consumption statistic. Creation is an economic one. Nigeria is currently very good at the first and is running out of time to get serious about the second.

Frequently Asked Questions

What is sovereign AI?

Sovereign AI refers to a country having control over the AI systems its economy depends on, including local computing infrastructure, locally held datasets, and domestically developed models, rather than relying entirely on foreign platforms.

How many Nigerians use AI?

The report found that over 70 percent of Nigerians have interacted with generative AI tools and 93 percent of organisations surveyed have begun deploying AI. Other measures, such as the Microsoft AI Diffusion Report, put economy-wide adoption closer to 10 percent, because they measure depth of integration rather than casual contact.

Will AI cost Nigerians their jobs?

The report projects around nine million routine jobs displaced by 2030 in banking, clerical work and public administration, alongside roughly eleven million new technology-enabled roles. The net gain depends entirely on large-scale investment in reskilling, which has not yet happened.

Why do AI models struggle with Nigerian languages?

Leading models are trained overwhelmingly on English-language text, so their performance in Hausa, Yoruba and Igbo is comparatively weak. Building high-quality indigenous language datasets is one of the clearest opportunities for Nigerian AI companies, because foreign labs have little commercial incentive to do it.

Are you building AI in Nigeria? We want to hear what is working and what is not. Tell us in the comments.