Why the Diagnosis Holds

Manamela points out that AI makes the university’s classic work more urgent. Students need to question arguments, test evidence, and take responsibility for what they claim to know. AI can spit out text without checking if it’s true, so critical thinking becomes even more important.

The job market backs this up. In the second quarter of 2026, 12.4 % of graduates were unemployed, while 39.2 % of people without a matric certificate were out of work. Youth unemployment hit 47.4 %. Although a degree still offers protection, the advantage is shrinking—unemployment fell for those with less than matric but rose for matriculants and graduates in the first quarter of the same year.

Building the Foundations

Policy

Cabinet approved the draft National AI Policy on 25 March 2026. Public comments are open until 10 June, and full rollout is expected in 2027 or 2028. The policy has six pillars, one of which focuses on talent and capacity building. The 86‑page draft also calls for seven new oversight bodies. The risk is that these governing structures could appear before any funded programmes are ready to support them.

Language Data

Manamela asks whose languages AI actually represents. None of the top 34 languages used online is African. Progress is happening, but it is mostly driven by community projects and donors. The Africa Next Voices initiative has recorded 9,000 hours of speech in 18 languages, including isiZulu and Tshivenda. The Masakhane hub—backed by Google.org, the FCDO, IDRC and the Gates Foundation—is funding data sets for 50 languages. What’s missing is a steady public funding line to keep these efforts going.

Compute

Computing power is arriving mainly through private companies. Stratos Lab is installing 400 Nvidia B300 GPUs in a Centurion data centre, and Cassava has taken delivery of 3,000 GPUs in South Africa. Still, Africa holds less than 1 % of global data‑centre capacity. Researchers feel this dependence too: Lelapa AI used Microsoft’s AI for Good Lab credits to train its InkubaLM model. For universities, the terms of access matter as much as the raw availability of GPUs.

What Universities Should Decide

Manamela urges universities to move from talk to action. Three choices stand out:

Assessment that Tests Judgement

Oral defences and supervised problem‑solving make it harder to outsource thinking than take‑home essays do. These formats force students to show they can reason on the spot.

Joint Programmes Pairing Computing with Humanities and Social Sciences

Questions about bias, power, labour, law and human dignity cannot be answered by computing power alone. Degrees that blend tech with the humanities give students the tools to tackle those issues.

Negotiated Compute Access and Data Stewardship

Researchers should not be stuck renting computing power on someone else’s terms. Universities can act as stewards for language data sets and negotiate fair access to GPUs for their own scholars.

He also notes that lifelong learning will become the norm. Qualifications must stay deep and rigorous, while modular, stackable credentials can help people up‑skill throughout their careers.

The Real Measure

A speech is only the start. Four concrete signs will show whether the ideas turn into real capacity:

Final AI Policy

The completed, approved National AI Policy must be in place and backed by funding.

Funded Compute Access for Researchers

Universities and labs need guaranteed, affordable GPU time that isn’t tied to restrictive corporate contracts.

Sustained Support for African‑Language Data

Long‑term public financing for projects like Africa Next Voices and Masakhane is essential.

Graduate Unemployment Trends

Future labour‑force surveys should show graduate joblessness staying low or falling, not climbing.

Conclusion

The diagnosis is clear: AI makes critical thinking more vital than ever, but the promise of AI can only be realized if universities act now. Smart assessment, interdisciplinary programmes, and fair control over computing resources are the levers that can turn policy into practice. If we watch the policy, funding, data support, and graduate job numbers, we’ll know whether the vision is becoming reality.