Six African telcos are teaching AI to speak the continent's languages — and a working Swahili model proves it's possible
A new GSMA report details how Airtel, MTN, Orange, Vodacom, Axian and Ethio Telecom are pooling data, compute and talent to build AI language models for African languages — with the first open Swahili reasoning model already live.
More than 2,000 languages are spoken across Africa — over 30 percent of every tongue on Earth. Almost none of them can hold a conversation with a modern AI system. That gap, and a coordinated effort by six of the continent's biggest mobile operators to close it, is the throughline of the GSMA's Mobile Economy Africa 2026 report, published this month.
The report documents how a coalition known as the G6 — Airtel, MTN, Orange, Vodacom, Axian Telecom and Ethio Telecom — has partnered with the mobile industry body, alongside researchers, startups and civil-society groups, to build large language models trained on African languages rather than adapted, awkwardly, from English. The initiative was first unveiled at Mobile World Congress Kigali in October 2025 under a deliberately pointed banner: "AI language models in Africa, by Africa, for Africa."
The framing matters because the alternative is already visible. Today's leading models are trained overwhelmingly on English and a handful of other high-resource languages. For hundreds of millions of Africans, that means the most powerful general-purpose technology of the decade simply does not work in the language they think, trade and pray in — a barrier that compounds every other digital divide the continent is trying to cross.
From slideware to a model that actually runs
What lifts this above the usual coalition announcement is that it has shipped something. At MWC Barcelona in March 2026, developers demonstrated the first open Swahili reasoning model, built with MeetKai Zambia. It can browse online content, translate it into Swahili, interpret a query and generate a contextual answer — the full loop, not a translation gimmick.
Swahili is a shrewd place to start: more than 100 million people across East and Central Africa use it, giving any working model an immediate, testable market. The detail worth noticing is where the build came from. Zambia is not a Swahili-speaking country. A team there produced an open model for a language spoken mostly in Kenya and Tanzania — which is precisely the point. The GSMA calls the result a replicable template, and a template is a very different asset from a one-off. If the same pipeline can be pointed at Hausa, Amharic, Yoruba or Wolof, the marginal cost of each new language falls with every model that ships.
The four things you can't fake
The coalition organises its work around four constraints it says no single company can solve alone: data, compute, talent and policy. Each now has a concrete anchor. On data, Google's WAXAL project — three years in the making, released in February 2026 with Makerere University, the University of Ghana and Addis Ababa University — opened a dataset of speech across 27 sub-Saharan languages spoken by more than 100 million people. On compute, Cassava Technologies has opened an AI factory in South Africa that rents GPU capacity to developers and operators. On talent, MTN's zero-rated Skills Academy now runs across 16 markets. On policy and safety, a GSMA–Zindi challenge produced a reusable benchmark for testing AI behaviour across 2,000-plus languages.
Why the operators, and why now
The telcos are not doing this out of charity, and the report is candid about the economics. Mobile technology contributed roughly $240 billion to Africa's economy in 2025, a figure the GSMA projects will reach $290 billion by 2030. Every African who cannot use a digital service because it does not speak their language is a customer an operator cannot fully monetise. Language-capable AI — especially voice interfaces for users who are not literate in English or French — is, in blunt commercial terms, a way to enlarge the addressable market.
That logic is beginning to change what operators buy. In April 2026, MTN joined a $45 million round for ODC, an AI-telecoms startup building network technology for Africa's patchwork of vendors and unreliable power. The move signals a shift from renting cloud capacity off global providers toward owning a slice of the technology layer itself. Whether the G6 can convert a strong Swahili demo and a tidy four-pillar framework into models that ordinary Kenyans, Nigerians and Ethiopians actually use every day is the open question. But for the first time the pieces — a live model, an open dataset, local compute and a paying reason to care — are sitting on the same table.
