South Africa’s Budget Has an AI Problem It Has Not Yet Counted
Personal income tax funds nearly 40% of revenue, and the earners who pay most of it work in AI’s direct path.
Sign up for my Substack daily AI newsletter here.
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
When South Africa's Finance Minister presents his medium-term budget on 21 October, the headline is likely to be good news. Economists expect tax collections to beat forecasts by about R50 billion in the first half of the year, with personal income tax and VAT holding up well. Nothing in those numbers will mention artificial intelligence, and that is the problem. Our tax system was built on the assumption that most income is earned by people in jobs. AI is beginning to test that assumption, slowly and unevenly, but in exactly the part of the economy that funds the state. This article argues that South Africa should start counting that risk now, while the revenue is still flowing, and that the answer lies less in taxing robots than in taxing the AI we import.
CONTEXT AND BACKGROUND
The South African state is, in effect, a partner in its workers’ salaries. Of the R2.01 trillion SARS collected in 2025/26, R767 billion came from pay-as-you-earn tax deducted from wages, about 38% of the total. Add the VAT paid out of wage-funded spending, and the budget rests heavily on people being employed and paid.
That foundation is now being questioned at the highest level. At the IMF and World Bank meetings recently, IMF Managing Director Kristalina Georgieva warned that “AI may raise growth, but it could also reduce labour income, increase inequality, and erode the tax base”, and urged governments to shift taxation away from labour towards sources such as land or investment. She described AI as “a slow-moving tsunami” and noted that hiring of young people in AI-exposed fields has already slowed.
INSIGHT AND ANALYSIS
The exposure may be sharper here than in most economies. In a country with mass unemployment, our salaried tax base is narrow, and much of it is earned in professional, financial and administrative work, the cognitive and clerical tasks generative AI performs best. There is little hard evidence yet that these jobs are shrinking in South Africa, and AI may first raise productivity and pay rather than cut headcount. The risk lies in what happens if that changes faster than the budget can adjust.
The tax code also tilts the choice. Bill Gates points out that employers pay payroll taxes on people while machines can usually be written off at once, which rewards replacing workers, and he proposes taxing robots and AI tokens to fund retraining. The same report records the critics: Larry Summers called the idea protectionism against progress, and economist Robert Seamans warned that it would slow investment, growth and wages.
I have previously written about this, arguing that AI is reducing the hiring of junior workers in roles built on codified knowledge, which weakens the pipeline that produces tomorrow’s experienced professionals. The fiscal version of that point is simple: a graduate who is never hired never pays PAYE.
For Africa, the most practical lesson comes from Kenya. Its new significant economic presence tax regulations expressly cover AI services sold by foreign firms to Kenyan users, at an effective rate of 3% of gross turnover. Taxing the consumption of imported AI suits a continent that buys most of its AI from abroad far better than taxing local investment.
IMPLICATIONS
The strongest counter-argument is that the erosion is predicted, not measured. ITIF, an industry-friendly American think tank, notes that labour’s share of US income has stayed between 57% and 64% since the 1950s, and argues that revenue would fall only after an extreme and sustained decline. South Africa’s current overrun supports the sceptics. Unilateral measures carry risks too: even the British proponents of a token tax concede that a national version “potentially faces competitiveness issues” without international cooperation.
Uncertainty is a reason to plan, not to wait. National Treasury should model AI scenarios for PAYE and VAT and publish them alongside its usual forecasts. SARS, which already uses AI to protect collections, should be ready to show how foreign AI subscriptions are taxed. Boards should recognise that every AI business case built on headcount savings has a fiscal counterpart, and that policy attention will follow the savings.
CLOSING TAKEAWAY
South Africa does not need a robot tax, and it should not panic about a revenue collapse that has not happened. It does need to stop treating AI as a technology question that the Treasury can ignore. Our budget depends on salaries, and AI is changing who earns them and where the value goes. The sensible response is to measure the risk, tax imported AI services fairly, and protect the junior jobs that become tomorrow’s taxpayers. A fiscus that counts the problem early can adapt gradually. One that waits will have to adapt in a crisis.
Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net
Comments