If AI really is about to reshape the global economy, the question for South Africa is not whether the technology arrives but whether the country owns any part of it.

As argued in a previous column, the abundance its boosters promise rests on a basic income the South African budget cannot remotely fund. That leaves policy, not the technology, to decide whether the country catches any of the upside or merely pays for it.

The scale of the gap is stark. The US and China control about 90% of the computing power for frontier AI and, according to research firm Recorded Future, own all 50 of the top-ranked foundation models.

The privately led Stargate project in the US alone has earmarked US$500-billion (R8.1-trillion) over four years. India, South Korea and Japan are buying sovereign compute by the hundred thousand graphics chips. South Africa, by contrast, published a draft national AI policy only in April 2026 and withdrew it within weeks after it was caught citing fabricated, AI-generated sources, leaving the country in a holding pattern while Kenya, Egypt and Rwanda pressed ahead.

Even before that embarrassment, critics argued it fixated on governance while neglecting compute, energy and talent. Microsoft’s R5.4-billion local commitment helps, but the gap is widening.

Policy choices – on whether to build compute, train people, tax capital and broaden the grant base – will sort winners from losers far more than the technology itself.

Unglamorous foundation

The work starts with the unglamorous foundation. Critics of the draft policy are right that compute, energy and capital, not another governance charter, are the binding constraints. South Africa needs to treat data centres, the chips inside them and the reliable power they draw as critical national infrastructure, and to push through the energy reform and grid investment without which no serious AI economy is possible. The country already hosts cloud regions for Google, AWS, Oracle and Microsoft; the trick is to turn that foothold into real local capacity rather than let it become a transit lounge for foreign platforms.

The more difficult task is fiscal, and it has to happen before the wave rather than after it. If AI keeps shifting income from wages to capital, a state that taxes mostly wages will watch its base erode at exactly the moment demand for grants climbs. The answer is to start taxing capital, digital services and AI-driven profits now, and to build the broad, stable financing any future basic income support will need, ideally alongside other developing economies so that capital cannot simply pack up and leave. Get this right early and a grant is sustainable; get it wrong and it collapses in its first downturn.

People are the other half of the equation. Talent retention is one of the draft policy’s glaring weaknesses. Egypt graduated 1 300 AI specialists in a single 2025 cohort. South Africa is losing coding talent and needs a pipeline of its own, through universities, technical colleges and schools, with real reasons for those graduates to stay. An AI economy with no engineers is a customer, not a competitor.

It helps, too, to negotiate from strength rather than drift into dependence. With Chinese models already cheaper and spreading through the Global South, South Africa has real leverage as the continent’s gateway, provided it bargains as a bloc, with the African Union and its Brics partners, for affordable access to models and hardware on terms it helps set. Even a modest sovereign or open-source capability is insurance against being priced out by a single supplier.

Finally, people will need a safety net while the economy retools. A targeted, well-run basic income at or near the food poverty line is not the billionaires’ utopia, but it is an achievable floor, one that cushions displacement and, on the Kenyan evidence, stimulates local enterprise rather than idleness. Paired with retraining and support for small business, that is the realistic South African version of abundance: not freedom from work, but protection from ruin.

It is worth conceding what the optimists get right. If AI drives down the cost of production the way its champions predict, the price of food, transport and electricity could fall with it, and a modest grant would stretch far further than it does today. A floor pitched at the food poverty line might, in that world, buy a decent life rather than bare survival, which makes a targeted basic income more tenable over time, not less. But that hope rests on two conditions South Africa cannot take for granted: that the price falls actually reach the goods poorer households buy, and that the productivity gains are passed on to consumers rather than captured by whoever owns the machines. Both lead back to the same question of ownership and tax that runs through everything above.

Read: The AI utopia South Africa can’t afford

None of this is a technological gift. It is statecraft: deliberate, early and unglamorous. Get it right, and South Africa catches a fraction of the upside and builds a floor under those the transition displaces. Get it wrong, and it imports its intelligence, its robots and its inequality, and pays for all three. – © 2026 NewsCentral Media