How Africa’s AI opportunity depends on power, policy and local languages

Semafor, citing the World Bank, reported that Africa has 18% of the world’s population and 0.6% of global data-center capacity, with 5% of that African capacity equipped for advanced AI workloads. The World Bank’s Africa Economic Update discusses infrastructure constraints, but the precise capacity figures were not independently confirmed in the material reviewed for this analysis. They are reported estimates, not a settled measure of readiness. (Semafor’s report)

The World Bank’s emphasis on practical adoption reflects the difference between using or tailoring existing tools and building frontier models from scratch. Semafor reports that the bank views the latter as an expensive use of limited resources. Adapting existing applications may offer a more immediate route to uses in productivity and service delivery, areas the bank identifies as potential benefits. But the approach still depends on infrastructure, suitable data, and people with the skills to deploy and maintain tools.

That makes public policy consequential, without making it a shortcut to economic gains. The World Bank identifies institutions and governance as part of the enabling conditions and calls for policy efforts to scale practical applications and broaden access to foundational resources. Decisions about infrastructure, skills, data, and oversight can shape whether adoption is feasible and whether its results can be assessed.

The material reviewed does not establish a single continent-wide regulatory plan, or show that policy changes alone would produce the projected gains. Nor does it show that one approach will suit every country. Policymakers face the work of setting priorities around existing capacity and needs, while the evidence needed to judge specific interventions is still limited.

Local-language tools need more than a pilot

Language is one practical test of whether AI services can reach people on their own terms. The reporting describes African languages as poorly represented in AI systems. If a tool performs inadequately in the language people use, its availability online does not by itself make it useful to them.

Togo’s Voix du Togo initiative illustrates one approach: developing language resources alongside a platform intended to help people access public services in local languages. The government and Togo AI Lab are testing an initial phase in Kabiyè, Ewé, and Moba. Expansion to about 50 Togolese languages is a later goal, not a current capability across all those languages.

Togo First reported that the platform had a public test in Lomé on Sept. 28. Sika Finance reported that a national local-language challenge launched on Oct. 2, inviting citizens to contribute language materials. These are signs of an initiative in development; they do not establish that the platform already delivers reliable public services, either in the pilot languages or at the intended wider scale.

The information available also leaves practical governance questions unanswered. It does not describe how contributors’ permission will be handled, how submitted materials will be checked, or who will maintain them. Gathering language material is only one part of building a dependable service: the platform’s performance and the way it handles people’s contributions matter too.

That is where policy and language meet. Public bodies and project operators will need ways to assess whether a service works for speakers of each language and to explain how contributed material is used. Without that evidence, a pilot can show that work is underway, but cannot establish broad access or reliable outcomes.

The GDP projection is not evidence of who benefits

The African Development Bank, drawing on work by consulting firm Bazara Tech, estimates that inclusive AI deployment could add up to $1 trillion to Africa’s GDP by 2035. Its detailed analysis derives the figure from an estimate of AI’s potential global economic contribution and an assumed African share. The bank also identifies data, computing, skills, trust, and capital as conditions for realizing gains.

The estimate describes a possible outcome under assumptions; it is not a measured increase in GDP or a guarantee. The sources reviewed do not establish which countries might capture the projected value, what share would reach local businesses, or whether workers and users of public services would benefit. They also do not provide country-specific evidence that the necessary conditions will be in place.

That gap matters for policy. A regional growth estimate cannot show whether a particular service works, whether a business has improved productivity, or whether people can use an application over a dependable connection. It does not settle who pays for the required infrastructure, who controls data used to adapt systems, or who answers when a service fails. Those questions require evidence from deployments, not conclusions drawn from a forecast.

A stronger test of the opportunity would pair investment in infrastructure and skills with public reporting on what deployed tools do: which services work, in which languages, for whom, and with what evidence of benefit. The World Bank’s focus on practical adoption and Togo’s pilot point to the same unresolved issue: whether tools can meet local needs under conditions people can actually use. Until performance, data practices, and the distribution of benefits are documented, Africa’s AI opportunity remains plausible but conditional.