The world’s digital wealth is being built on a foundation of language, but not our indigenous languages.
The AI revolution, powered by large language models (LLMs), speaks English first.
The LLM model was born in Silicon Valley and nurtured in Beijing. It has supercharged economies there, creating fortunes and consolidating power.
For the originators, it’s innovation. For us, all too often it’s just another imported product we must struggle to use. This anomaly was on my mind on Monday at the SATNAC 2025 conference, while listening to Telkom group chief digital officer Sello Mmakau.
Instead of the usual tech jargon he spoke about a paper from young innovators at the University of Zululand, “Morpheme-aware tokenization and fine-tuning of large language models for isiZulu”.
The project aims to build and fine-tune LLMs for Zulu. Simple yet profound, the project is about teaching AI to think, understand and respond in the nuanced, beautiful complexity of Zulu. Not just translating from English, but originating in Zulu.
This project is not a small academic experiment; it is the seed of a different digital future for South Africa.
The US economy is being reshaped by LLMs that automate tasks and spawn new businesses. China uses its own AI to modernise factories and cement its tech independence. Their growth is tied to their linguistic command of the tech.
This raises the question of how South Africa’s digital economy can truly soar if its most advanced tools cannot speak to millions in their mother tongue. Real growth isn’t just about having the apps; it’s about whether the apps can have a meaningful conversation with a grandmother in Ngcobo, a farmer in Kokstad, or an entrepreneur in KwaNyamazane.
The innovative students from the University of Zululand have understood this clearly. Their paper says strengthening Zulu in tech “directly supports education and telecom access, [and] narrows the digital divide”. They are right. A Zulu LLM could be the bridge we desperately need.
It could allow people to access banking, government services and market information using their voice. It could unlock the economic potential of those who are digitally eager but linguistically left behind.
Mmakau’s insight is refreshing, because he knows a brilliant model is not enough. These students need a foundation.
Mmakau also spoke of “strong data sovereignty frameworks” ― meaning our data must stay here, governed by us, for our benefit. He talked about Telkom’s role in building affordable connectivity and cloud infrastructure to enable, among other things, LLM innovation.
This is about building a complete ecosystem on our own terms. Without this even the best Zulu LLM would just be a clever tool running on someone else’s platform, feeding into someone else’s economy.
At the same event Telkom Group CEO Serame Taukobong stated: “When Africa connects, Africa competes. When Africa competes, Africa creates. And when Africa creates, humanity rises.” The connection he speaks of must be more than installing fibre cables. It must be cultural and linguistic.
The University of Zululand’s project is a bold step towards that kind of creation. It is about competing not by mimicking others but by leveraging what is uniquely and powerfully ours ― our languages, our context, our people.
So, can a Zulu LLM enable growth? It is not just a question of tech. It is a question of choice. Will we settle for being consumers in a digital world shaped by others, or invest in the difficult, foundational work of building our own LLMs?
The paper from the University of Zululand is more than research. It’s an invitation to a challenge: our digital destiny should not be written for us in English or Mandarin. It must be coded, spoken and built in Zulu, Sesotho, Afrikaans and all our other languages.
The rain does not fall on one roof alone. Our digital prosperity shouldn’t either.
• Lourie is founder and editor of TechFinancials.
Would you like to comment on this article?
Sign up (it's quick and free) or sign in now.
Please read our Comment Policy before commenting.