Four percent, or a rounding error: the IMF's verdict on Africa and AI
A new IMF study puts a number on what everyone in African tech already suspected: without power, broadband and skills, AI adds almost nothing to the continent's growth. With them, it adds a decade's worth of ambition.
Four percent. That is the size of the prize the International Monetary Fund says artificial intelligence could add to sub-Saharan Africa's economic output over the next ten years, if governments get the fundamentals right. Get them wrong, and the number the IMF's own economists use is 0.2 percent — a figure lead author Martin Schindler, in comments reported by TechCabal, called little more than a rounding error.
The report, Unlocking the Potential: AI in Sub-Saharan Africa, published July 21 and confirmed independently by Reuters, the IMF's own newsroom and TechCabal, is not a story about whether Africa will get AI. Data centres are already rising in Johannesburg, Lagos and Nairobi; Microsoft and the Emirati firm G42 opened a geothermal-powered, roughly $1 billion campus in Kenya in 2024, and Cassava Technologies has said it will deploy 12,000 Nvidia chips across five countries. The report's real subject is whether that AI will do anything for the roughly 1.2 billion people who live outside the buildings housing it.
The gap between 4 percent and 0.2 percent is not a modelling quirk. It is, in the IMF's telling, the gap between a continent that fixes its oldest infrastructure problems and one that does not. Around half of sub-Saharan Africa's population still lacks reliable electricity; of the roughly 670 million people worldwide without any access at all, 85 percent live in the region. Only 38 percent of Africans used the internet in 2024, against a global average of 68 percent — and millions of those offline already live inside mobile broadband coverage, kept out only by the cost of a smartphone or a data bundle. Training and running AI models requires power that runs continuously and connections that hold steady. Neither is a given across most of the continent.
What makes this report worth reading past its headline number is the asymmetry it identifies between Africa's risk from AI and the risk richer economies worry about. In Washington, London and Brussels, the anxiety is job displacement — AI doing work people used to do. In sub-Saharan Africa, where most employment sits in agriculture, informal retail and manual services less exposed to direct automation, the IMF argues the greater danger is irrelevance: a local manufacturer or trader who never touches an AI tool but loses customers to a better-resourced competitor abroad who does. As the report puts it, a worker does not need to be replaced by AI to be disadvantaged by it; a competitor only needs to become more productive.
That unevenness shows up geographically, too. Africa has roughly 160 data centres, representing about 5.5 percent of global computing capacity — and nearly half of them sit in just three countries: South Africa, Nigeria and Kenya. Businesses and researchers in Johannesburg or Lagos already have a shorter, cheaper path to the computing power AI requires than counterparts in Malawi or Chad, who must lease capacity hosted abroad at added cost and latency. The report's sharper warning is about who captures the value once that capacity exists: without deliberate policy, the IMF cautions, the continent could end up hosting foreign-owned servers, running on African electricity, while most of the commercial return flows elsewhere. A parallel UNDP estimate cited in the report puts a number on the resulting brain drain: an AI researcher in a G7 country can iterate on a model roughly every 30 minutes; one in Africa, working with the same idea but without comparable compute access, may wait up to six days for a single training run to finish. That gap, more than any policy document, is what pushes talent abroad.
The IMF's prescription is deliberately modest, and more useful for it. Rather than urging African governments to compete for frontier foundation models — a contest that requires the kind of capital and chip access few can muster — it points to adapting existing models to local languages, building genuinely local datasets, and giving researchers and small businesses affordable access to computing power that already exists. The report's own framing captures the shift: the valuable skill may not be building the world's most powerful model, but making an existing one work for a Kenyan farmer, a Nigerian manufacturer or a Rwandan clinic. TechCabal's independent reporting on the same document draws the analogy to mobile money, which won not by replicating Western banking but by building directly on the phone and agent networks Africans already had.
The report's most pointed line, though, is aimed at governments rather than technologists. A national AI strategy, however well drafted, cannot compensate for a school with no electricity or a ministry whose records still live on paper. That argues for a different distribution of responsibility than the one currently in place across much of the continent — energy regulators, education ministries, and procurement offices doing more of the work than the technology ministries usually given the AI brief. Whether that redistribution happens is, per the IMF's own framing, the difference between a rounding error and a decade of real growth.
