In a windowless server room in Lagos, Nigeria, an AI founder watches their runway shrink, squeezed by a stack of constraints: revenue in volatile naira but compute bills in U.S. dollars, scarce local GPU infrastructure, unreliable power, limited access to high-quality local datasets, and talent that often looks abroad for more opportunities.
The narrative of African AI is often framed as “leapfrogging” and “potential.” But the operational reality is a friction between ambitious government policy and the cold, hard economics of infrastructure.
In April 2024, the Nigerian Federal Ministry of Communications, Innovation, and Digital Economy convened 120 experts to draft the National AI Strategy. The plan promises a “compute revolution” and partnerships with local firms such as 21st Century Technologies to build domestic cloud capacity. But for researchers on the ground, that vision remains distant.
“Let me be honest, the vision hasn’t yet translated into real compute capacity for most public universities,” Sakinat Folorunso, an associate professor of AI systems at Olabisi Onabanjo University and a key contributor to the National AI Strategy, told The AI Innovator. “There’s still a very real gap between what’s written in policy documents and what researchers deal with every day.”
In practice, many university labs rely on personal laptops, sporadic cloud credits, and unstable power. Those constraints shape the work itself.
“That gap forces many of us to shrink our ambitions to smaller datasets and lighter models,” Folorunso explained. “Co-creation succeeded socially, but delivery is where we’re still stuck.”
Enterprises buy off-the-shelf tools
Large Nigerian companies have the deep pockets to deploy AI, now going beyond chatbots to agentic AI systems that can execute tasks across workflows.
- BUA Foods, one of the country’s largest consumer goods companies, integrated AI agents into its distribution network. Using the BeatRoute platform, it deployed a scheduling AI agent and an order AI agent to optimize delivery routes and predict stockouts before they happen. The result was a 36% revenue increase to ₦913 billion in the first half of 2025, largely driven by supply chain efficiency.
- MTN Nigeria, a telecom provider, is using graph analytics and federated learning to fight fraud. Traditional fraud detection fails against “SIM farms,” warehouses of SIM cards mimicking human behavior. MTN’s ‘Project Genova’ uses AI to map the social network of transactions, identifying clusters of non-human actors.
But these companies are buying tools off-the-shelf instead of becoming co-creators of the technology. That risks foreign dependency and doesn’t help the country develop its own expertise.
“Right now, most Nigerian banks and telecoms are users of AI, not co-creators of it,” Folorunso said. “They tend to buy ready-made solutions or license black-box systems, focusing on efficiency rather than long-term capacity.”
She argued that until public procurement rewards companies that work with local researchers, the ecosystem will remain fragmented. “If we want local research to matter … firms should be incentivized to build with Nigerians, not just buy from abroad.”
The sovereign data battle
While code is important, the most critical battle centers on data. Global models such as GPT-4 are trained disproportionately on English language and Western data, leading to a cultural alignment gap. But efforts are underway to alleviate the situation.
Intron Health, founded by Dr. Tobi Olatunji, is building sovereign datasets. Standard speech recognition models struggle with heavy African accents, leading to errors in medical transcription. Intron collected 3.5 million audio clips from 18,000 speakers to build “Sahara,” a proprietary model that achieves 92% accuracy on African medical terminology.
This aligns with Folorunso’s work in federated learning. She said this technique of training models on decentralized data without moving the raw files is a geopolitical necessity, not just a privacy tool.
“As global hyperscalers expand across Africa, the real risk isn’t collaboration; it’s dependency,” Folorunso said. “My work in federated learning is about making sure we can collaborate globally without exporting our raw data. It shifts the dynamic from extraction to negotiated collaboration.”
This distinction is vital, as companies such as Microsoft and G42 pledge to invest $1 billion in Kenyan geothermal data centers. Without sovereign controls, African nations risk becoming “data colonies,” renting back intelligence derived from their own citizens.
A coming technical brain drain?
Nigeria also faces an imbalance between talent and opportunity. The government’s flagship 3 Million Technical Talent (3MTT) program aims to train 3 million Nigerians in technical skills. It is an ambitious program designed to position Nigeria as a net exporter of talent. But there are risks.
“Here’s the uncomfortable truth: We’re training talent faster than we’re creating serious local opportunities,” Folorunso said. “Without a strong local ecosystem like deep-tech startups and industry R and D roles, we risk preparing young people mainly for jobs outside Nigeria.”
This leads to the “Japa” phenomenon, or emigration. Folorunso’s initiative, “The Route to AI Learning (TRAIL),” attempts to counter this risk by embedding students in local community projects such as health and agriculture. But she is realistic about the economics.
“People don’t leave because they lack patriotism; they leave because opportunity is thin. Until staying makes economic sense, ‘Japa’ will remain a rational choice,” she said.
Taken together, these forces — compute constraints, reliance on imported systems, data dependency, and talent outflow — define Nigeria’s current AI landscape. Innovation is happening, but often in spite of the environment rather than because of it.
The government has established policy direction and momentum. What remains is the harder task of building the underlying infrastructure: reliable power, shared compute, and incentives that align research, industry and talent.
“Sovereignty isn’t about how big the model is. It’s about where control sits,” Folorunso said. “AI sovereignty won’t be announced at a conference. It will be built quietly, locally, and over time through choices we make now.”
Be First to Comment