A Different Kind of AI Rollout
What South Africa Is — and Isn’t — Doing, and Whether It Can Close the Race Gap
eyesonsouthafrica
Amsterdam, Septermber 7, 2026 – Saudi Arabia and the UAE moved fast and moved from the top down: a state-mandated AI curriculum for six million schoolchildren, a national enrollment drive putting a premium AI subscription into university students’ hands, ministries and tech giants working from the same script. South Africa’s version of this story looks nothing like that — not because the ambition is missing, but because the country is building on a much more uneven foundation, one where the biggest question isn’t whether AI can improve education, but whether it will do so for everyone or mostly for those already ahead.
What South Africa Actually Has
South Africa is not without its own version of the free-AI-for-students moment. In October 2025, Google extended a 12-month free Google AI Pro subscription — including its Gemini 2.5 Pro model, research and coding tools, and 2TB of cloud storage — to university students aged 18 and over across six African countries: Ghana, Kenya, Nigeria, Rwanda, South Africa, and Zimbabwe. Google’s Sub-Saharan Africa lead, Alex Okosi, framed the goal as an attempt to help students “compete and lead on a global stage.” It is, in scope, a continental cousin of the Gulf’s offer rather than a South Africa-specific initiative, and — like the Saudi program — it targets university students, not the roughly 13 million children in South African public schools.
At the school level, the picture is thinner and further behind. South Africa’s Department of Basic Education has acknowledged it is developing a National AI in Education Framework, referenced most recently in its 2026 digital education update following the eLearning Africa Ministerial Roundtable, where officials stressed that AI must be adapted for African contexts and that foundational literacy and numeracy should remain central rather than be displaced by new tools. A parallel Draft National AI Policy, considered by the Department of Communications and Digital Technologies in 2026, sets out a vision explicitly built around inclusion — strengthening AI education through STEAM curricula and community AI centers, and using the technology to digitize South Africa’s twelve official languages. But both documents remain in draft or framework form. In the meantime, individual schools and private-sector partnerships are filling the gap: consultancy Helm has worked with early-childhood platform SmartStart to deploy an automated assistant in eleven languages that eases administrative pressure on teachers, and with Capitec on a financial-literacy chatbot whose design principles are increasingly being borrowed for classroom use.
Critics argue that gap between ambition and implementation is itself the story. Celeste Labuschagne, a lecturer and learning-framework developer who has written on the subject, has warned that without clear direction from the Department of Basic Education, teachers are left experimenting with AI tools on their own simply because learners are already using them at home — a dynamic she has described as the country “sleepwalking” into policy failure by default rather than by design.
The Question the Gulf Countries Don’t Have to Ask in Quite the Same Way
South Africa’s AI-in-education debate carries a weight that Riyadh’s and Abu Dhabi’s largely don’t: the country’s school system is still, three decades after apartheid ended, one of the most unequal in the world, and that inequality tracks closely with race. The Department of Basic Education’s own funding model sorts schools into five “quintiles” based on the socio-economic status of the community they serve, with quintile 1 schools — overwhelmingly serving Black African communities in townships and rural areas — receiving the least per-pupil funding and quintile 5 schools, disproportionately former “Model C” and independent schools with a more racially mixed but historically white and Indian/Asian pupil base, the most resourced.
That divide shows up starkly in the data researchers use to think about AI readiness. A recent analysis of South Africa’s economic hub, Gauteng province, found that only 39% of Black African households had home internet access, compared with 87% of Indian/Asian households and 86% of white households — a gap the researcher who conducted it attributed to entrenched social inequality along lines of race and income rather than any single policy failure. Rural quintile 1 schools studied elsewhere in the country reported being unable to run remote lessons at all during COVID-era closures, while quintile 4 and 5 schools moved seamlessly online — a contrast one study summarized starkly as proof of “how unequal we are.”
This is precisely the tension a recent national education gathering, or lekgotla, wrestled with directly: AI would clearly play a role in South Africa’s educational future, but — as the discussion concluded — how it was implemented would determine whether it helped or harmed learners, with growing concern about the gap between well-resourced and underprivileged schools widening rather than closing.
So Would It Help Close the Race Gap — or Widen It?
The honest answer, consistent with the global research on AI in classrooms, is: it depends entirely on the distribution model, not the technology itself.
The optimistic case is real and worth taking seriously. Personalized, adaptive AI tutoring is, in principle, exactly the kind of resource that wealthier South African families already buy privately — after-hours tutors, small-group extra lessons, one-on-one homework help — and that under-resourced Black African households in townships and rural areas have historically been priced out of. If a struggling Grade 8 learner in a quintile 1 school in Limpopo could get the same infinitely patient, judgment-free explanation of a maths concept that a fee-paying student in a Johannesburg suburb gets from a private tutor, that is a genuine leveling force. Jeané van Greunen of Helm has argued exactly this, describing AI’s potential to “accelerate progress toward high-quality, inclusive education” if the rollout is approached thoughtfully.
The less comfortable case is that “if” is doing enormous work. AI tools built for smartphones and reliable broadband are of limited use where 61% of Black African households in South Africa’s most developed province still lack home internet, where load-shedding disrupts power supply unevenly across regions, and where data costs remain punishingly high relative to income for the households who would benefit most. Layered onto a school-funding system that already allocates resources along the old lines of the quintile system, an AI rollout with no deliberate equity mechanism attached to it would most likely follow the path every previous education technology wave in South Africa has followed: quintile 4 and 5 schools adopt it fastest and see the earliest gains, quintile 1 and 2 schools adopt it last, if at all, and the gap between the two — a gap already correlated with race — widens rather than narrows.
What separates the optimistic outcome from the pessimistic one is not the sophistication of the AI itself but decisions that are entirely human: whether the eventual National AI in Education Framework comes bundled with real investment in connectivity and devices for quintile 1 and 2 schools, whether offline-capable and low-data tools are prioritized alongside flagship cloud products, and whether teacher training reaches rural and township schools with the same urgency as it reaches well-resourced suburban ones. South Africa has, in other words, exactly the same choice the Gulf states are testing at a different scale: whether AI is deployed as a scaffold that lifts the students furthest behind, or handed out in a way that simply lets the students already ahead move faster still.
This piece draws on 2025–2026 reporting on Google’s African student AI offer, South Africa’s draft national AI policy and education-department frameworks, and academic and journalistic research on the country’s digital divide. Policy documents referenced remain in draft or developmental stages and may change as they are finalized.
