Sovereign AI on a Small Budget: Choices for Emerging-Market Governments
Building frontier AI models is out of reach for most governments. Controlling the data, standards, skills and procurement that determine how AI serves citizens is not.
“Sovereign AI” is often presented as a race to build national AI models and data centres. For most emerging-market governments, that race is unaffordable and unnecessary. Sovereignty in AI is less about owning the most powerful models than about controlling the terms on which AI is used in the country.
Kenya’s National Artificial Intelligence Strategy 2025–2030, launched in March 2025, is built around three pillars: digital infrastructure, a data ecosystem, and research and innovation. That framing points to where a limited budget goes furthest.
Govern the data. Clear rules on how public and private data can be shared, protected and used for AI training matter more than any single model. Data that stays fragmented or locked away cannot serve local needs.
Invest in local language and local context. Global models perform worst on the languages, accents and contexts that are least represented in their training data. Funding local datasets, in partnership with universities, gives citizens AI that works for them.
Buy computing, do not necessarily build it. Shared regional capacity and agreements with cloud providers, including terms on where data is stored, can deliver much of the benefit of national data centres at a fraction of the cost.
Use procurement as policy. Governments are among the largest buyers of technology. Requiring transparency, local data storage, audit rights and skills transfer in AI contracts shapes the market more than most regulation.
Grow people. Every other choice depends on public servants who can specify, evaluate and oversee AI systems.
The countries that make these choices deliberately will shape how AI serves their citizens, whatever models they end up using.