Blog
Browse published essays and reports. Search by keyword or filter by category.
19 articles
African governments are adopting AI for public services faster than they are passing AI laws. This post asks whether public procurement can fill part of that gap. It traces procurement’s history as a policy tool, including in Africa, and sets out five conditions for it to work. It then draws on evidence from the US and EU on AI procurement and considers what this means for African governments.
Almanac Africa Editor 38 min read
African researchers and builders have already produced useful AI systems with small models, modest data and limited computing power. This article examines examples in agriculture, health, language and environmental monitoring, the methods that made them affordable, and what they suggest for public policy.
Almanac Africa Editor 16 min read
How AI licences decide who benefits from open models, and how African licences such as NOODL, Esethu and N-ATLaS try to make openness work on equitable terms.
Almanac Africa Editor 30 min read
Foundation models now draft and translate, write code, read images and act on a user’s behalf. What a model can reliably do in a particular setting is a separate question from what a benchmark score records. This article sets out the current range of capability, how claims about it are produced and what they establish, and what the term “reasoning model” describes.
Almanac Africa Editor 19 min read
Foundation models perform poorly in most African languages than in English, and safety protections do not reliably carry across languages. A growing body of African-led evaluation now documents these gaps. This brief sets out what that evidence shows, where its limits lie, and the questions it raises for policymakers designing oversight.
Almanac Africa Editor 32 min read
The second article in the Powering AI series explains three terms found in any data centre proposal: load, rack density and Power Usage Effectiveness (PUE). It shows how policymakers can use them to assess what a facility will demand from the grid and from local water supplies, with examples from South Africa, Nigeria and Kenya.
Winfred Mutinda Editor 17 min read
Discussion of African AI is usually measured against the largest frontier models. This opening instalment of The Lightweight Stack sets out the other forms of participation available to the continent today, including small and mid-scale models, adapted open models, and African language datasets and benchmarks, along with the limits of that approach.
Almanac Africa Editor 13 min read
The European Union has shaped global digital rules through laws such as the GDPR and the AI Act, yet it still relies heavily on external providers for cloud services, chips and frontier AI models. This article reviews Europe's regulatory, infrastructure and industrial approaches to digital sovereignty, considers how Africa's institutions, markets and development priorities differ, and sets out seven lessons for African policymakers.
Almanac Africa Editor
At UNGA81, 128 delegations raised artificial intelligence, and African priorities on the AI divide are now reflected in official UN texts. This analysis reviews the key AI developments of the week, the positions African leaders advanced, the state of financing and bilateral agreements, African engagement with Chinese, European and US approaches to AI governance, and the developments to watch before UNGA82.
Almanac Africa Editor 25 min read
Which foundation models sit beneath the AI services used across Africa, how they arrive, and what the public record cannot show. A look at the models built elsewhere, adapted here and trained here, and why the gaps matter for regulators.
Almanac Africa Editor 13 min read
Almanac Africa Editor 11 min read
AI runs on electricity. How training and inference create demand, what data centres need from a grid, and why it matters for Africa's power systems.
Winfred Mutinda Editor 13 min read