Language becomes a
new infrastructure

“As soon as one telco or bank offers the opportunity for Africans to engage in their preferred languages, the public will demand that language choice becomes service-agnostic.”

That comment by Pelonomi Moiloa, CEO of pan-African artificial intelligence research lab Lelapa AI, is not a prediction. It is a warning. And it captures one of the most overlooked, yet pivotal, technology shifts already reshaping Africa’s digital future: language is becoming a new infrastructure.

While it is almost obvious that the next frontier of AI will be dictated by Silicon Valley’s obsession with scale or China’s data-intensive AI regime, one version of the future is taking shape in African languages, in African contexts, and by African innovators.

The most consequential AI trends of the coming years are not about size, but specificity. They won’t be driven by who has the largest models, but by who builds the smartest, leanest systems tailored to real-world needs.

The most telling clue lies in the global turn toward small language models (SLMs) – compact and efficient generative AI that doesn’t demand planet-warming levels of power or mountains of data.

“Our work at Lelapa AI has always focused on efficiency, building leaner, high-performing AI solutions optimised for African contexts,” says Jade Abbott, Lelapa AI chief technical officer. “It’s exciting to see the rest of the world is now realising what we’ve known all along that smaller, smarter models are the future.”

Lelapa’s InkubaLM, launched in August 2024, is a case study in where AI is going. Africa’s first multilingual SLM, it supports under-resourced languages like Hausa, sisZulu, Swahili and isiXhosa. At a time when most AI tools remain locked into English or other dominant languages, InkubaLM points toward a future where language diversity is not a limitation but a catalyst for innovation.

What this means for businesses is straightforward: AI must speak the languages of its users – not only technically, but also culturally. More than 60% of people say they prefer consuming content in their home languages. In the future, says Moiloa, companies that ignore this reality will lose relevance.

“This shift will redefine customer expectations across industries,” she says.

The trend is clear: future-facing businesses in telecoms, banking, and public services will embed language AI at their core. Those that don’t, may find themselves out of step with their consumer base.

But enabling language inclusion at scale requires more than ambition. It needs serious infrastructure. Investment in Africa’s growing cloud infrastructure – capable of hosting and running AI models locally – is a strategic step forward. Until now, most African AI applications have run on servers overseas, increasing costs and creating latency.

“Hosting AI models locally, rather than relying on overseas servers, is key,” says Abbott. “Reducing latency and improving response times will be crucial for industries that require real-time AI interactions, such as customer support and conversational AI applications.”

That infrastructure push goes hand-in-hand with another critical trend: ethical and inclusive AI development. The old model – scraping data from anywhere and answering to no one – is giving way to more accountable practices. Lelapa AI places community ownership and linguistic fairness at the heart of its model development, on the basis that representation in AI is not a nice-to-have, but a necessity.

Bias in machine learning systems is well documented. But bias in language AI cuts even deeper when it excludes entire linguistic communities. Moiloa sees African language not only as a technical challenge, but also a moral one.

“Making AI work for the people who need it most means solving real problems,” she says.

That could mean enabling live customer support in African languages or allowing small businesses to use voice-based platforms. The goal, ultimately, is not to chase global trends, but to lead with local purpose.

* Arthur Goldstuck is CEO of World Wide Worx and editor-in-chief of Gadget.co.za. Follow him on Bluesky on @art2gee.bsky.social.