Ask the people building AI where it all leads and the answer is paradise. Sam Altman, OpenAI’s CEO, argues in his essay, Moore’s Law for Everything, that AI will drive the cost of goods and services towards zero, because labour is the largest input in most supply chains.

Build a house with robots and onsite solar, he writes, and the cost falls to “close to just the cost to rent the robots”. He now says he wants “universal extreme wealth for everybody”. Demis Hassabis of Google DeepMind talks of :radical abundance” and a “new renaissance”. Elon Musk goes further, predicting “universal high income” and a world with “no shortage of goods or services”.

Look past the hype and the mechanism is surprisingly mundane: if humans no longer earn wages, something has to replace the wage. That something is almost always a universal basic income (UBI), a no-strings monthly payment to every citizen. This entire utopia hinges on whether the state can fund it. So can ours?

Let’s start with what we have. The 2026 national budget put consolidated spending at about R2.67-trillion against roughly R2.35-trillion in revenue, with gross debt peaking near 79% of GDP. Social grants already absorb more than a quarter of a trillion rand and reach about 26.5 million people. The social relief of distress grant pays R370/month to roughly 8.5 million people and costs around R35-billion/year, and national treasury still frets about affording even that.

Now scale it up. Stats SA’s rebased 2025 food poverty line, the level of bare survival, is R855/month. Its upper-bound poverty line, closer to a dignified minimum, is R2 846. Pay every one of South Africa’s roughly 43 million adults even the current R370 and the bill is about R191-billion/year, or about 8% of all revenue, on top of the existing grant budget and an existing deficit. Pay them the upper-bound R2 846, the closest thing to an income you could live on without working, and the cost is about R1.47-trillion, nearly two-thirds of everything government collects. This is not a tight call. It is impossible.

The honest answer

So the honest answer to “what UBI can South Africa afford today?” is: roughly R370/month, and only because it is targeted at the poorest rather than being universal. Even the government’s proposed basic income support, aimed at 10.7 million working-age adults below the tax threshold, would cost more than R100-billion/year at the food poverty line. With serious tax reform and a decade of growth, the realistic ceiling by 2035 is a targeted basic income somewhere near that food poverty line for the working-age poor. That is relief from destitution. It is nothing like freedom from work.

None of this is an argument against cash. The evidence for basic income is genuinely encouraging. GiveDirectly’s 12-year Kenyan study, the largest ever run, found lasting gains in income, food security and enterprise, with money spilling into local economies and creating work even for people who received nothing, and no sign that recipients stopped working. Finland’s landmark trial improved mental health without denting employment. The lesson is that modest, targeted cash can work well.

Read: US government puts GPT-5.6 behind closed doors

The leap from that to a liveable, universal income is a different beast, and it demands rebuilt tax systems. Proponents argue a grant pays for part of itself by lifting spending and tax revenue, which is plausible. But funding the rest means taxing the thing AI enriches, which is capital, not wages. Altman himself concedes the technology is “killing the labour-capital balance” and that “nobody knows what to do about it”. For a developing country with a thin tax base, a large informal economy and highly mobile capital, taxing AI profits that accrue mostly to a handful of firms in California is a central, unsolved problem.

The Future of Life Institute argues the real danger is not killer robots but concentration. AI, it warns, is “on course to concentrate power within a small number of groups”, creating what it calls a “second species” of intelligence that could “disempower huge swathes of humanity”. Seen this way, a basic income bankrolled by a few foreign AI giants is not emancipation but dependence. It turns citizens into pensioners of someone else’s machine, and a payment that can be granted can also be withdrawn.

That exposes the most difficult question: if Washington and Beijing decide who gets top-tier models and cheap humanoid robot labour, abundance becomes an export they price and ration. A country that imports its intelligence and its robots, and cannot tax the profits they generate, does not get the utopia. It gets the bill. Chinese models are already being entrenched across the Global South, and that, as much as anything, may decide whose ecosystem South Africa ends up living inside.

The endgame

The abundance the billionaires describe may well arrive for whoever owns the machines. Whether it reaches an ordinary South African depends on arithmetic the evangelists rarely mention: a budget that cannot fund a living income, a tax system not built to capture AI’s gains and compute we mostly rent from abroad.

Utopia, it turns out, has a balance sheet. Ours does not yet add up, and closing that gap will be a political project, not a technological gift.

In a second column on this topic, the author will look at what South Africa would have to do to catch even a fraction of the AI upside.