In May 2024, Microsoft and Abu Dhabi-based G42 announced plans for a $1 billion data centre campus in Kenya, to be powered by geothermal energy. Kenya already generates roughly 40% of its electricity from geothermal sources, concentrated in the Olkaria fields along the Rift Valley, making it one of the few African markets where large-scale, renewables-powered computing is feasible today. Cassava Technologies and Nvidia have also disclosed intentions to deploy 12,000 specialised chips across five African countries. Governments from Rwanda to Nigeria are publishing national strategies designed to turn artificial intelligence (AI) into a new driver of economic growth.

The obstacle is fundamental: approximately half of sub-Saharan Africa still lacks reliable electricity. Those shortfalls directly threaten the continent's AI ambitions. African nations are competing to attract the infrastructure required to run some of the world's most advanced technologies even as millions of homes, schools, and businesses remain without dependable power or affordable internet access.

The IMF's Two Scenarios

In a report released on Tuesday, the International Monetary Fund (IMF) estimates that AI could increase sub-Saharan Africa's economic output by as much as 4% over the next decade — but only if countries invest substantially in electricity, digital infrastructure, and skills. Under current conditions, the gain could be as little as 0.2%.

Martin Schindler, the IMF paper's lead author, described the smaller figure as little more than "a rounding error."

While governments and workers in the United States and Europe are focused on how many jobs AI might eliminate, Africa confronts nearly the opposite problem: the technology may not spread widely enough to generate meaningful economic gains.

Most workers in the region are employed in agriculture, informal retail, and manual services — sectors where today's AI systems are less likely to directly replace them. However, that does not make them immune. African businesses could lose customers and contracts as competitors elsewhere deploy AI to write software, forecast demand, manage inventory, and deliver services more efficiently. A worker does not need to be replaced by AI to be disadvantaged by it; their competitor only needs to become more productive.

"For sub-Saharan Africa, the central concern is not the risk of technological disruption, but whether countries will be able to adopt, adapt and scale AI quickly enough to capture its benefits and avoid falling further behind," the IMF paper states.

Infrastructure Over Innovation

Africa's AI challenge is therefore less about producing a rival to today's leading AI models than about deploying practical tools on farms, in classrooms, in clinics, and in small businesses. The continent's place in the global AI race may ultimately depend on far less glamorous investments: transmission lines, fibre-optic networks, affordable smartphones, and workers with the skills to use the technology effectively.

Many African states are too small to build complete AI ecosystems independently. Greater regional cooperation could enable them to share infrastructure, align regulatory frameworks, and create larger markets for AI products. The African Continental Free Trade Area, which came into force in 2021 and covers 54 countries, offers an existing framework for cross-border digital trade, though implementation of its digital protocol remains at an early stage. Yet efforts to harmonise digital policies have frequently moved more slowly than the technologies they are intended to govern.

The Adoption Problem

Only 38% of Africans used the internet in 2024, compared with 68% of the global population. Many more live within mobile broadband coverage but remain offline because smartphones and mobile data are simply too expensive.

That gap between coverage and actual internet use has become an AI gap. A chatbot may technically be available across Africa, but availability means little to a market trader who cannot afford a smartphone, a farmer whose device cannot run the application, or a teacher who pays for every megabyte of data.

The IMF's optimistic scenario assumes that AI will spread far beyond banks, telecom companies, large retailers, and well-funded startups. To generate meaningful economic gains, the technology must reach the small businesses and informal workers who account for a large share of employment across the continent.

That will require products designed around local realities rather than imported assumptions. Africa's most consequential AI applications are unlikely to be standalone platforms that demand constant broadband connectivity and dollar-denominated subscriptions. Instead, they may take the form of voice assistants that understand local languages, tools embedded within WhatsApp, or systems that function reliably with low-cost devices and intermittent internet connections.

Mobile money followed a similar trajectory. M-Pesa, launched in Kenya by Safaricom in 2007, succeeded not by replicating Western banking infrastructure but by building on the mobile phones and agent networks that people already relied on. AI will need its own version of that adaptation.

A Data-Centre Economy

Africa has approximately 160 data centres, according to the IMF, with almost half concentrated in South Africa, Nigeria, and Kenya. By comparison, regions such as Western Europe and North America host thousands of facilities each. Those three African markets are also attracting a significant share of the continent's cloud computing and AI investment.

This concentration raises the prospect of a two-speed AI economy. Businesses in Johannesburg, Lagos, and Nairobi could enjoy faster and cheaper access to computing capacity, while companies in smaller or landlocked countries may continue relying on infrastructure hosted abroad, adding cost and delay.

Even countries that successfully attract data centres are not guaranteed broad economic gains. The facilities require large capital investments but employ relatively few people once construction is complete. Their value depends on whether local startups, universities, and public agencies can afford the computing power inside them. Without that local uptake, Africa could end up hosting foreign-owned servers powered by African electricity while most of the commercial value is captured elsewhere.

The Old Problems

Governments have responded to the AI boom with strategies, task forces, and promises to train thousands of workers. These plans are necessary, particularly as the technology raises questions about privacy, cybersecurity, misinformation, and control over public data. But a national strategy cannot compensate for a school without electricity or a government ministry whose records remain on paper.

The IMF's findings suggest that AI policy cannot be left to technology ministries alone. Energy regulators, education departments, competition authorities, and public-procurement agencies may have more influence over adoption than newly created AI councils.

African countries must also decide which part of the global AI race they can realistically contest. Training the largest foundation models requires billions of dollars, advanced chips, and vast amounts of electricity — resources few countries on the continent can marshal. A more practical strategy would focus on adapting existing models to African languages and industries, developing useful local datasets, and giving researchers and businesses affordable access to computing capacity.

The valuable skill may not be creating the world's most powerful model. It may be making an existing one work for a Kenyan farmer, a Nigerian manufacturer, or a Rwandan clinic.

The IMF's 4% estimate describes what Africa could gain. Its 0.2% estimate says more about where the continent is currently heading. Africa's AI boom is already underway in corporate offices, technology hubs, and government conference rooms. Whether it reaches the rest of the economy will depend on a more ordinary question: can people switch on a device, connect to the internet, and afford to keep using it?

For now, the answer remains uncertain.