African central banks weigh the cost of an AI-shaped financial order

When Kenya's President William Ruto chaired the African Heads of State Summit on Artificial Intelligence in Nairobi in March, the room produced a declaration that was shorter on ambition than on anxiety. African leaders want what the technology promises, the declaration said: cheaper credit, faster diagnostics, more accurate climate forecasts, and, eventually, productive work for the roughly 12 million young Africans who enter the labour market each year. They also acknowledged, more plainly than such gatherings usually do, that the platforms, chips, and foundation models underwriting those gains are not African. The declaration named that gap as the central problem.

The numbers behind the worry are not new but have hardened in the past 18 months. According to the World Bank, Africa holds about 1 percent of the world's data centre capacity, despite carrying roughly 17 percent of the world's population. Most African sovereigns depend on foreign-owned hyperscalers for cloud storage, and almost all frontier-model deployments are accessed through APIs that route inference and, frequently, training data, through facilities in Virginia, Frankfurt, or Singapore. The 2024 African Union Continental AI Strategy, ratified in Addis Ababa, identified the imbalance as a sovereignty question. African governments, the strategy argued, must decide whether to build capacity, regulate the foreign platforms operating on the continent, or both.

African central banks have been the most concrete actors so far. The Bank of Ghana, the South African Reserve Bank, the Bank of Botswana, the Central Bank of Nigeria, and the Banque Centrale des États de l'Afrique de l'Ouest have all, since early 2025, published consultation papers or supervisory letters addressing AI-driven credit scoring, automated market-making, and the use of machine learning in anti-money-laundering workflows. The papers share a common vocabulary. They speak of model risk, third-party concentration, and operational resilience, and they almost always end with the same request: local data infrastructure, the technical capacity to audit algorithmic systems, and a seat at the table when foreign platforms are licensed.

A first attempt at coordination is underway. The African Monetary Cooperation Programme, an initiative of the African Union, has convened a working group on AI in finance, scheduled to deliver draft model rules in mid-2026. According to a senior official at the African Union Commission, the working group is drawing on the European Union's AI Act, Brazil's draft AI bill, and, more cautiously, the United States' Executive Order 14110, which the Trump administration revoked in February 2025. The African approach, the official said, will emphasise tiered risk classification, public-sector data trusts, and a presumption that core financial infrastructure should be held by African-domiciled entities.

The question is whether domestic capacity can be built quickly enough. The AU's strategy estimates that the continent will need to train roughly 100,000 AI specialists by 2030 to meet even half of its public-sector demand. The current output of African universities in computer science and applied mathematics is well below 10,000 a year. The shortfall is most acute in French-speaking and Portuguese-speaking Africa, where the European Union's language and infrastructure footprint is heavier, and where joint-research agreements with European institutions tend to direct both talent and data northward.

Private capital is partly filling the gap. In 2025, African AI start-ups attracted roughly $1.2 billion in venture funding, according to a recent industry survey, with South Africa, Egypt, Nigeria, and Kenya accounting for more than four-fifths of the total. The funding is concentrated in three areas: fintech credit scoring, agricultural advisory services delivered by SMS, and language-model adaptations for low-resource African languages. Most of these firms license foreign foundation models rather than building their own. The firms' pitch to investors is the same: cheaper, faster, more local than what the American platforms offer.

Multilateral lenders are now restructuring around the same premise. The African Development Bank, working with the Afreximbank and the Development Bank of Southern Africa, has proposed a $2.5 billion continental AI facility, with a first tranche expected to be approved by the AfDB board in late 2026. The facility is intended to fund both compute infrastructure and the integration of AI tools into the operations of African public-development banks. Whether it succeeds will depend, as so much in African financial policy does, on whether the larger foreign platforms that hold the underlying models choose to extend their presence on the continent, or to retreat behind the regulatory walls now rising in Brussels, Washington, and Beijing.

The trajectory that African policymakers are most alert to is the one now under way in other emerging-market regions. Indian, Brazilian, and Indonesian authorities have, in the past 18 months, taken steps to require that AI systems used in regulated finance undergo local review and, in some cases, store training data domestically. These steps are not aimed at exclusion. They are aimed, in the plain language of a 2025 Reserve Bank of India circular, at ensuring that the foreign platforms do not become systemically important in ways that African regulators cannot supervise, audit, or, if necessary, switch off.

That is the position African regulators are moving toward. It is, in the language of the AU strategy, a question of financial sovereignty in a moment when the infrastructure of money is increasingly software. The continent's leaders are aware that the build-out will take years, that it will require fiscal and diplomatic commitments they have not yet made, and that it will run against the preferences of the foreign platforms on which African finance now depends. The Nairobi declaration's restraint was, on that reading, a kind of acknowledgement: that the AI moment has arrived in Africa, and that the cost of missing it has begun to be counted.