When 38 million Indian farmers received accurate AI-powered monsoon forecasts last year, the result looked like exactly the kind of story development economists have been waiting for. The predictions, built by an international team blending models from Google and the European Centre for Medium-Range Weather Forecasts, gave farmers four weeks' notice and flagged an unusual late start to the rainy season. It is, in the words of Rose Mutiso, executive director of the African Tech Futures Lab, her favourite AI example. So when the Financial Times asked her whether artificial intelligence could help poor countries 'leapfrog' into prosperity, one might have expected enthusiasm. Instead, she offered a careful warning.
Leapfrogging is the idea that a developing country can bypass an older technology and jump directly to a newer one — skipping landlines for mobile phones, for instance, or skipping bank branches for mobile money. In the mid-1990s and 2000s, Africa did exactly this, becoming a global leader in mobile adoption without ever wiring much of the continent for telephones. Many now hope AI will follow the same pattern. Mutiso disagrees. The crucial difference, she argues, is that mobile phones bypassed legacy infrastructure, while AI is one of the most infrastructure-intensive technologies ever invented. To train and run AI models, a country needs vast amounts of electricity, computing power, data, and trained institutions. Africa, she notes, currently hosts roughly 1% of global data centre capacity.
Mutiso is also wary of treating every problem as a nail for the AI hammer. The Indian monsoon model only worked because decades of groundwork were already in place: a century of rainfall data from the UK Met Office to calibrate against, a global climate data set, and the institutional capacity to localise the model to Indian conditions. Real success, she argues, is not a 'just-add-AI' story. Instead of asking what to apply AI to, she suggests, governments should think in terms of sequencing: deciding which bottlenecks — energy, connectivity, data governance, regulation — to solve, and in what order.
The labour story is equally uncomfortable. Kenya has become a global hub for AI-related outsourcing, with workers performing data labelling, training, and reinforcement learning for foreign tech companies. These jobs, Mutiso notes, often involve difficult content and weak protections, echoing earlier controversies around essay-writing mills that served students in the UK and elsewhere. Yet she is reluctant to dismiss the opportunities outright. Her position is that both governments and tech firms must build proper frameworks — and that countries hosting tech firms have real leverage, if they choose to use it.
Her deeper worry is structural. For decades, the path out of poverty for large parts of Asia ran through manufacturing: factories migrated from Japan to South Korea, then to China, then to South-east Asia and South Asia. Africa was supposed to be next. But automation and 'reshoring' — companies bringing factories back to richer countries — are beginning to close that door. About 80% of Africans, Mutiso estimates, work in the informal sector, which means most are insulated from the first wave of AI disruption hitting white-collar work. The long-term danger is different and bigger: that the traditional ladder out of poverty is being pulled up before African economies can climb it. AI, in this view, is neither the savior nor the villain. It is a powerful technology that will reward countries that have already built the boring stuff — power grids, schools, institutions, rules — and quietly punish those that haven't.
Last year, 38 million Indian farmers got AI-powered monsoon forecasts that actually worked. So why is one of Africa's sharpest tech thinkers warning that AI won't be the magic shortcut everyone keeps promising?
The Financial Times sat down with Rose Mutiso, executive director of the African Tech Futures Lab, to ask whether AI can help developing countries 'leapfrog' — skip past older technologies and jump straight to the cutting edge, the way Africa skipped landlines and went straight to mobile phones.
Mutiso's answer is a careful no. She points to a real AI success — 38 million Indian farmers getting accurate monsoon forecasts — but argues that AI is fundamentally different from mobile phones. It doesn't bypass infrastructure; it demands enormous amounts of it: compute power, electricity, data centres, and trained institutions. Africa currently hosts roughly 1% of global data centre capacity.
The leapfrog metaphor only works when the new tech genuinely replaces the old plumbing. AI doesn't replace plumbing — it IS plumbing, just a new and hungrier kind.
If you're a teenager thinking about a career in tech, policy, or international development, this is the debate that will define the next 20 years. Mutiso is worried that the traditional development pathway — manufacturing jobs migrating from China to South-east Asia to South Asia and eventually to Africa — is being eroded by AI and automation before African countries can catch the wave. That has huge implications for global migration, inequality, and where the next billion middle-class jobs come from.
Two decades of mobile-based 'revolutions' in poor-country agriculture produced more pilot projects than lasting change — a warning that hyped technologies layered onto weak systems tend to disappoint. The questions to watch: will African governments build domestic data centres and power capacity, or remain dependent on cloud infrastructure in Europe and the US? Will policymakers regulate the labour conditions of data labellers, or leave that to the tech companies themselves? And will the next phase of globalisation produce a new development model — or simply leave large parts of the world behind?