» Unfair Advantage
Global AI Watch

Africa wrote the AI strategies. It forgot to check the power grid

2026-06-10 · Unfair Advantage

In May, Kenya's president suspended a flagship $1 billion data centre backed by Microsoft and the UAE's G42. The blocker wasn't money, politics, or red tape. It was electricity: the facility would have eaten roughly a third of Kenya's entire 3,000 MW national grid. That single decision exposes the gap between Africa's AI ambitions and its physical reality. The African Union has a continental AI strategy; Egypt, Rwanda, Senegal, Nigeria, and Kenya all have national ones. But the whole continent hosts about 200 data centres delivering only 500 MW of compute — roughly 1% of global capacity — and the hyperscaler regions where AI workloads actually run exist in just two countries: South Africa and Egypt. Everywhere else, AI traffic gets routed there, picking up delay, foreign-currency costs, and data-sovereignty friction along the way. The lesson is blunt: you can't run the AI economy on policy documents. You run it on reliable power, fibre, and water for cooling — and that's the part most strategies skip.

Why it matters

If you sell to, source from, or operate in African markets, latency and data residency are about to shape what AI you can actually deploy there. With live hyperscaler regions only in South Africa and Egypt, an app serving Lagos or Nairobi routes its AI calls hundreds of milliseconds away — slow enough to hurt real-time tools and expensive enough to notice. The flip side is opportunity: enterprise integration (AI built into healthcare, logistics, finance workflows) has lower barriers than building data centres, and that's where a small, focused operator can win before the infrastructure giants arrive. Watch the power story. When a country fixes grid reliability, that's the signal a real local AI market is opening.

Network impact

LatencyAI workloads from most of Africa route to South Africa or Egypt, adding hundreds of milliseconds and degrading real-time inference for users elsewhere on the continent.
SecurityCross-border routing of AI traffic raises data-sovereignty and residency exposure — sensitive data leaves its country of origin to be processed, complicating compliance under emerging local rules.
ScalabilityGrid capacity is the hard ceiling. A single hyperscale facility can demand a third of a national grid, so compute can't scale until power does — capacity is projected to grow fourfold by 2030 but starts from a 500 MW base.

What to do

  1. If you deploy AI features for African users, check where your provider's nearest region is — South Africa and Egypt are the only live hyperscaler regions, so measure real latency before you promise speed.
  2. For data touching African customers, map residency rules now; routing inference cross-border may breach local data-sovereignty laws as they tighten.
  3. Target the enterprise-integration layer (AI inside existing workflows), not infrastructure — it has lower barriers and is where a small operator captures value.
  4. Track national grid and data-centre announcements per country; reliable power is the leading indicator that a local AI market is genuinely open for business.
  5. Budget for foreign-currency and connectivity costs when serving markets without local compute — they quietly inflate every AI call.

Sources

« All articles