South Africa Doesn't Have a Skills Crisis. It Has an Access Crisis.

On Youth Day this year, a story went round the country's group chats for a day: Lungelo, a KwaZulu-Natal-born town planning graduate living in Observatory, Cape Town, stood at a traffic intersection holding a handwritten poster of his own qualifications and his phone number. He had already completed an eight-month municipal internship. He was still looking, and a sign at a red light was the next thing left to try. News24's Drum and Briefly.co.za both covered it in June 2026.

He is a real person, reported in real outlets, not a composite built for a lede. That distinction matters, because the number behind him is real too. Stats SA's Quarterly Labour Force Survey for the first quarter of 2026 put unemployment among people with a bachelor's degree or higher at 10.3%. The national rate was 32.7%. Among people without matric, it was 51.6%.

Read the concession first, because burying it is how this argument loses a fair reader: 10.3% is nowhere near 32.7%. Education still works, by a wide margin. Anyone citing graduate joblessness without naming the comparison rate is playing the same selective game this piece is about to accuse the skills lobby of.

But 10.3% is not zero, and for a cohort that did exactly what it was told, it is higher than the promise implied. That gap, between what education was sold as and what it delivers, is what the rest of this piece is about.

Unemployment rate by education level, Q1 2026

The framing that does not fit

A skills gap does not produce one in ten qualified people sitting idle. A gate does.

The four gates

So if not skills, what?

Geography. Apartheid planning put poor people far from where the jobs went, and the map has barely shifted since. Harvard's Growth Lab, working with South African labour data, put direct commuting costs at around 17% of net wage income nationally, rising to an average of 57% once time is priced in. For the lowest income quintile the transport share of labour income runs above 35%, against under 10% for the highest quintile. Where the median commute in an area exceeds an hour, unemployment sits 21 percentage points higher in cities and 25 points higher in rural areas, compared with areas where the median commute is under 15 minutes. That is not friction. That is a second rent, paid in time and cash, before the first wage arrives.

Share of labour income spent on commuting, by income quintile

Networks. Skip the "most jobs are never advertised" line. It circulates constantly and nobody attaches a method to it, so it does not belong in a piece that holds employers to the same standard. What does hold up: a 2026 British Journal of Sociology study out of UCT on corporate graduate hiring found employers screening for who "feels relatable," a judgement the researchers describe as an institutionalised mechanism that reproduces existing racial and class patterns even inside formally diverse hiring processes. An older UKZN study of private recruitment consultants found much the same thing from the supply side: consultants justifying client requests that exclude on race while denying that is what they were doing. Two different methods, a couple of decades apart, converging on one finding: who gets considered runs on fit with an existing network, and that network was not built by accident.

Credentials. Pnet's own vacancy data shows 54% of job adverts over the past year listed a bachelor's degree or higher as a requirement, while only 36% of applicants met it. Run the same data the other way and it gets stranger: in roles like call-centre and customer support, where 83% of adverts state no degree is required, 15% of applicants have one anyway. Degree-holders are pushed down into jobs that never needed a degree, filling seats a non-graduate could have had, while the vacancies actually demanding degrees stay unfilled. The credential is not measuring the job. It is measuring who gets a hearing.

The credential gap in job advertisements

Capital. Harambee's work with young job-seekers documents the mechanism directly: someone who needs two taxis to reach a job, in the month before the first salary lands, is measurably less likely to keep it, because the fare costs more than the job pays before payday arrives. Searching is not free. Data, transport to interviews, a phone that holds charge, clothes that read as professional: all of it is an expense, and for someone with no income, expenses get cut before they get paid.

Four gates. None of them opens with a training certificate.

Treating a distribution problem as a training problem lets the real barriers stand.

The networks gate is not colour-blind

This is the part that needs stating plainly rather than gestured at, because gesturing at it is worse than either saying it or leaving it out.

The Department of Employment and Labour's 26th Commission for Employment Equity report, released in July 2026, found White representation at top management sitting around seven times the White share of the economically active population, a pattern that has barely moved across successive annual reports. That is not a pipeline problem waiting for this year's graduates to fix. It is a quarter century of successive annual reports showing the same flat pattern, sitting on top of hiring processes that, per the studies cited above, run substantially on network fit.

Here is the distinction the brief for this piece insists on and the one worth holding onto: this is not an argument that a racial group causes unemployment. It is an argument that the network gate, one of the four described above, was built under apartheid and has not been meaningfully rebuilt since, and that a hiring process which rewards "relatability" to an existing network will keep reproducing whoever was already in it, mechanically, without anyone at any single hiring desk intending to discriminate. Explaining the mechanism is not the same claim as blaming the people who currently sit inside it.

Where the skills story is right

The counterargument deserves its full weight, so here it is, all of it, in one place.

Skills shortages are not a myth. Companies reporting difficulty hiring artisans rose from 10% in 2024 to 22% in 2025, and reported ICT specialist shortages rose from 10% two years ago to 22% now, per critical-skills survey coverage from late 2025. South Africa produced 19,536 qualified artisans in 2022/23 against a National Development Plan target of 30,000 a year by 2030. Employers report that only 42% of IT applicants meet stated requirements as demand surges. Those are real shortages with real consequences, and they need a real answer, which is more and better technical training, aimed at the specific trades where it is genuinely short.

The error in the skills story is scale, not existence. Those shortages are sectoral and specific: electricians, millwrights, data engineers, a defined list of named occupations. Stretching a defined list into a general account of an unemployment crisis affecting a third of the labour force is where the reasoning breaks.

There is a second version of the employer complaint too, and it deserves to be taken on its own terms rather than dismissed. When employers say they cannot find people, they are reporting their experience accurately. The question is what "cannot find" is doing in that sentence. Cannot find at the wage offered? Cannot find through the referral networks they actually recruit from? Cannot find among the applicants who survive a degree filter the job never needed? Each of those is an access problem wearing a skills costume, and none of them is the employer lying about what they see from their side of the desk.

Why the diagnosis matters

The diagnosis picks the cure, which is why this is not a semantic argument.

Diagnose skills and you fund training. The National Skills Fund's most recent audit drew a second consecutive disclaimer opinion from the Auditor-General, with R637.6 million in grant expenditure unable to be substantiated due to missing records, and Services SETA reporting a 49% decline in meeting its own administration targets. That is not a modest shortfall in an otherwise working system.

Diagnose access and the evaluated alternatives look different. A South African field experiment found that giving job-seekers a structured reference letter raised callback rates by 60%, with the effect concentrated among women applicants, at the cost of one page of paper and a form. It will not fix a crisis of this size on its own, and nobody evaluating it claims otherwise. But it is a measured effect from a real trial, which is more than the training system above can currently show for considerably more money.

Skills are necessary. They have never been sufficient. Funding the necessary while ignoring the rest produces precisely what the numbers above show: one in ten qualified people still waiting, with the certificate filed and the gate unmoved.

So when the next official release repeats the word "skills," ask the question that number cannot answer on its own: skills for which nine of the ten jobs already found, or access for the tenth?