Africa and AI: the technological leap not to miss

In brief

For the first time since the steam engine, a technological revolution is not reserved from the outset for countries with money and factories. AI mostly requires method, data and trained people. These are things Africa can build for itself.

The continent has leapfrogged once before: it went straight to mobile phones, then to mobile payments, without ever having a real landline network. A similar leap is possible with AI, provided we build on open models and on the data of our own territories.

My conviction, after years in a field as complex as aviation: let’s forget billion-dollar mega data centers, keep control of our data, and aim for lean AI, applied first to airports, roads, hospitals, schools and public services.

Preamble

In February 2024, I was in Cairo with African airport directors gathered by ACI Africa. In the sessions, we talked about air traffic, security, airport development and financing. I had been invited to give a presentation on AI in general and its possible applications in our field. The audience seemed genuinely interested, and the session was followed by many conversations on the subject. Yet I came away with a bitter feeling that I believe we all shared: we were watching the AI revolution from the bench.

I have forgotten neither that moment nor that sense of frustration. Almost three years later, I am trying to answer it as best I can. What levers do our countries really have? What is wishful thinking, and what can be done tomorrow morning? This text is my answer, necessarily incomplete, to those colleagues.

1. A revolution unlike the others

Until now, every major technological wave has widened the gap between North and South.

Steam, steel and electricity required decades of investment in railways, blast furnaces and power plants. Under colonial rule, Africa mostly served as a reservoir of raw materials for industries built elsewhere.

With microelectronics, the door closed even further. A leading-edge chip plant now costs more than $20 billion. The most advanced lithography machines have only one manufacturer in the world, the Dutch company ASML. There is no point dreaming of breaking in.

AI changes the game. It isn’t free, but its center of gravity has shifted. Highly capable models are now released with open weights. They lag a few months behind the best closed models, which is more than enough for the vast majority of real-world uses. The rest, meaning data, skills and organization, can’t be bought off the shelf. It has to be built.

We have done it before. In 2007, Kenya launched M-Pesa. Millions of people who had never had a bank account started paying, saving and sending money with a simple phone. Nobody waited to go through the bank-branch stage first. That is the spirit of the leap I am advocating for AI: don’t retrace other people’s path, take the shortcut.

2. “With what? We still lack water and roads”

That is the objection I hear most often. Why talk about servers when villages have no running water?

I understand it, but I think it rests on a misunderstanding. Used well, AI is not a rich country’s gadget. It is precisely a way to better manage what is lacking. A few examples:

Cheap sensors and a small predictive model are enough to fine-tune irrigation and anticipate a poor harvest.
In a rural clinic with no radiologist, an X-ray reading assistant helps spot urgent cases and refer them faster.
In ports, airports and at border crossings, better flow forecasting shortens queues without pouring a single extra cubic meter of concrete.

In transport, which I know from the inside, this is even clearer. Anyone who has run an airport or an airline in Africa knows that daily life means power cuts, telecom links that drop, and operational surprises to be handled within the hour. AI won’t make these problems disappear on its own. But with the right resources, it helps you anticipate them instead of suffering them.

The real risk, in my view, is not doing too much. It is falling behind. If we let this train pass, we will remain mere users of tools designed in California or Shenzhen, with their priorities and their blind spots.

3. Forget the mega data centers

Let’s be honest: there are things we won’t be able to copy, and that’s just as well.

A cutting-edge training cluster, with some 16,000 interconnected GPUs, costs around a billion dollars. It draws tens of megawatts around the clock and, depending on its cooling system, huge volumes of water. In countries where power cuts are part of daily life and water is getting scarcer, that model doesn’t hold up.

The right answer is called lean AI. We give up on training our own giant models. Instead, we take open models, often small ones, adapt them to our needs, and run them as close to the field as possible: on a local server, or even directly on equipment installed in a hangar or next to a runway.

In early 2025, DeepSeek showed how far optimization can go. The Chinese company said it had trained a frontier model for a few million dollars. That figure covers only the final training run, not the years of research before it. The lesson still stands: engineering matters as much as the checkbook.

We still need a minimum of computing power to adapt models to our own data. That is where pooling makes sense. A computing center shared between several countries or sectors costs far less than twenty small national projects competing with each other.

Morocco has chosen a more ambitious path. The country has launched studies for a 500-megawatt green data center complex in Dakhla, with a sovereign segment and a segment open to private investors, and another project of the same size is planned in Nouaceur. The argument makes sense: the Kingdom has abundant renewable energy, which few African countries can claim. But questions remain. According to an estimate reported by TelQuel, a 500 MW site could consume nearly 13 billion liters of water a year, and domestic demand won’t be enough to fill it. These centers will therefore have to find their customers abroad. The bet can pay off, as long as the more modest but immediate needs on the ground are not forgotten.

And let’s keep our feet on the ground. Building a billion-dollar computing center next to a runway that doesn’t even have modern lighting would be absurd. First things first.

4. Our data, our business

Jensen Huang, NVIDIA’s CEO, likes to explain that AI has become an industry that produces intelligence, the way others produce energy. If that’s true, the raw material of this industry is data. And some of that data exists nowhere but here: our languages and dialects, customary law, the medical specifics of our regions, or the informal economy, which supports millions of people and appears in no Western database.

Handing all this over to foreign clouds without thinking carries three risks. First, seeing our public data feed, for free, products that will later be sold back to us as subscriptions. Second, hosting sensitive information (borders, air traffic, citizens’ identities) on servers subject to foreign laws, such as the US CLOUD Act. Third, depending on tools that poorly understand our languages and social realities.

I want to be clear-eyed on one point, though. This data is valuable, but rarely ready to use. Much of it still sits on paper, in incompatible formats, or simply in people’s heads. Our languages have very few digitized texts. Before talking about sovereignty, there is real groundwork to do: collecting, cleaning and labeling data. Communities like Masakhane and companies like South Africa’s Lelapa AI have started this work for African languages. They deserve support, and many more like them are needed.

Morocco has started tackling this: its partnership with the French company Mistral AI (in French) includes developing solutions adapted to Darija and Amazigh. It’s a useful step, as long as the data and the skills stay at home.

Should we wait for others to digitize our dialects for us? Certainly not. In practice, sovereignty means keeping control of your data, training engineers locally, and adapting models on infrastructure whose legal framework you control.

5. Where do we really stand?

Recent figures show a continent on the move, but a very uneven one.

On strategy, the world has moved fast: more than 60 countries have developed a national AI strategy, according to the OECD. The African Union adopted its continental strategy in July 2024. But turning it into national action is slow: as of July 2025, only 16 of Africa’s 54 countries had launched their own.

On readiness, the Oxford Insights index, which assesses 195 governments, paints a mixed picture. In the 2025 edition, Egypt leads the continent, ranking 51st worldwide with 57.5 points. In Sub-Saharan Africa, Kenya (65th), South Africa (67th), Mauritius (71st) and Nigeria (72nd) lead the pack. The index’s authors point to two persistent weaknesses in the region: computing infrastructure and AI adoption by public administrations.

On funding, the picture is well known: most AI venture capital goes to the United States and China, and Africa receives only a tiny share. Expecting the market alone to close the gap would be wishful thinking. Targeted public procurement and well-designed public-private partnerships will have to do part of the work.

A ranking doesn’t tell the whole story, but this one makes a point: falling behind is not inevitable. Egypt went from 111th place in 2019 to 51st in 2025.

Focus on Morocco

For a long time, Morocco marked time in these rankings. Oxford Insights placed it 88th in 2023, then 101st in 2024 with 41.78 points, held back by weak scores in governance and technology. In 2025, it climbed back 14 places, progress the government attributes to a stronger legal framework and skills development.

This rebound reflects a policy that has picked up speed. The Digital Morocco 2030 strategy, launched in 2024, has a budget of 11 billion dirhams (about $1.2 billion) for 2024-2026, covering AI and fiber optics in particular. Following the National AI Conference in July 2025, the government unveiled the Maroc IA 2030 roadmap in January 2026. Its targets are concrete: 100 billion dirhams (over $10 billion) added to GDP, 50,000 jobs and 200,000 trained and certified professionals by 2030.

The backbone of this strategy is the network of Jazari Institutes, set to cover all twelve regions of the Kingdom, with branches dedicated to smart cities, education and industry. On the infrastructure side, Mohammed VI Polytechnic University has hosted a first sovereign data center since January 2025, providing cloud services to public and private organizations, and a 50 MW center is being prepared in Rabat.

The framework is in place. What remains is the essential part: delivering projects on the ground that make a real difference to people’s lives. The Kingdom’s airports, ports and hospitals will be the best judges.

6. A concrete case: the airport

Let’s take a field I know well. At an airport, three projects can pay off quickly without heavy infrastructure.

Foreign object debris (FOD) detection. Cameras and a vision model installed along the runway can spot a suspicious object within seconds, day or night. A single piece of debris ingested by an engine can cost hundreds of thousands of dollars, not to mention the possible consequences for operational safety. The same AI vision principle can help fight bird strike hazards, given the risk birds pose to aircraft movements at African airports.

Turnaround time forecasting. A model trained on historical rotations predicts delays before they spread through the schedule. It also helps allocate parking stands and ground crews.

Equipment maintenance. Baggage belts, boarding bridges, generators: their failures follow patterns that sensor data can anticipate. I discuss this in more detail in my article on predictive maintenance.

None of these projects requires a data center. A local server, clean data and a small team able to deploy a model are enough to get started.

7. Risks not to underestimate

Moving fast doesn’t mean moving without safeguards.

The more we connect and automate critical infrastructure (airports, power grids, railway signaling), the more it becomes a target. Cybersecurity must come before automation, not after it.

Deepfakes and information manipulation can do serious damage in societies where trust is already fragile. Election periods are especially exposed.

Finally, no system that affects safety or citizens’ rights should be deployed unless its decisions can be explained and traced. In aviation, we have a strong culture of traceability and human factors. It should serve as a model: the machine proposes, the human decides.

8. Where to start

To make this leap without spreading resources too thin, here is the approach I recommend to decision-makers and infrastructure managers alike.

Priority	In practice
Pool resources	Bring together airports, ports and rail operators in the same region into data consortia, to train shared models without anyone giving up their intellectual property.
Choose cost-effective use cases	Skip showcase projects. Start with runway safety, debris detection, punctuality and maintenance.
Train “deployers”	Quickly train developers and engineers able to put a tool into service within three months.
Stay free	Reject contracts that lock you into a single vendor, and favor open technologies that can be adapted locally.

Final thoughts

AI offers developing countries a rare opportunity: not to repeat the history of previous industrial revolutions. Knowledge is accessible. The data is ours. What is still missing is the decision to make it a priority, and the discipline to execute.

The airport directors I met in Cairo have no reason to stay on the bench. The game is only just beginning.

What about you: in your sector, what would be the first AI project to launch? Let me know in the comments.

Zouhair Mohammed El Aoufir, founder of Flying Tek, former executive in air transport and airport infrastructure.

Africa and the AI Train: The First Technological Leap Without an Insurmountable Barrier