In 2024, Microsoft and G42 announced a $1 billion data-centre initiative in Kenya. It later emerged that the facility would draw on nearly half the country's electricity grid -  a dependency the government reportedly discovered only after the letter of intent was signed. The deal has since stalled.

That single fact is worth sitting with, because it's not really a story about one contract. It's a preview of a decision hundreds of African governments are about to make, at scale, over the next two years.

AI is being sold into critical public services, health records, education platforms, social protection systems, at the same moment official development aid is retreating. According to the OECD, health-sector aid to the region is projected to fall by up to 60% from its recent peak. 

Into that gap has stepped a wave of commercial AI investment, pitched as the natural successor to the aid that's disappearing. The commercial logic is straightforward enough. The political framing is not: adoption now, or fall behind for good.

Call that framing what it is: manufactured urgency. Not the ordinary pressure of a genuinely time-limited opportunity, but a decision environment built, deliberately, so that waiting looks more expensive than committing. 

You can usually tell the difference. Genuine urgency is backed by evidence specific to your context. Manufactured urgency is asserted before any evidence exists, told the same way in every market, and produced by the same people who profit if you move fast.

Patterns persists

None of this is unfamiliar to African finance ministries. The structural adjustment programmes of the 1980s and 90s were signed by governments under fiscal pressure, on terms nobody had time to properly evaluate, with consequences that outlasted the officials who signed them. By 2023 the continent's external debt had passed $1.15 trillion. AI contracts for critical services carry the same shape of risk: long-term, hard to unwind, negotiated by a handful of overstretched advisors against vendor teams built for exactly this kind of deal.

The response isn't to reject AI. Most governments shouldn't, and won't. The response is to notice who currently has to prove what. Right now, a government has to justify caution and has to explain why it's "falling behind." Almost nothing requires the vendor to justify the impact of the technology and the legitimacy of the deal.

A step forward

That can be reversed. A government that simply asks: show me the real cost over ten years, not year one; show me this actually works in a context like mine, not a pilot somewhere else; tell me honestly what water, power and land this needs; change the negotiation before a single clause is discussed. Waiting for that evidence stops looking like falling behind. It starts looking like the only defensible thing a finance ministry could do.

That's a small shift in principle and a large one in practice and it's also, we think, an integral part of the response that's still missing. 

The continent already has real, capable organisations doing serious work on AI governance and safety. What's harder to find is the economics: the cost models, the benchmarks, the evidence standards that let a government actually weigh a deal, not just govern one. That's why we’re creating the Centre for Applied AI Economics: a shared, independent home for exactly this kind of evidence.

If this matches a gap you're seeing, or you've already tried something like it, we'd like to hear from you. Reach out to our Frontier Technologies team, or me directly, as we shape this together.