On February 2, Google released WAXAL — 11,000 hours of open speech data covering 21 African languages, collected over three years with African university partners, owned by those African partners under an open licence, and published to Hugging Face for any developer in the world to use. On March 6, Google rolled out AI Overviews and AI Mode in 13 African languages inside Google Search. These are not the same move. One hands African researchers raw material to build language AI themselves. The other brings Google’s AI product directly to African-language users. Both are significant. Neither is sufficient. And the gap between what happened in February and March and what remains undone is a precise measurement of where AI localisation in Africa actually stands.
What Google’s AI Search Expansion Actually Does
AI Overviews places an AI-generated summary at the top of search results — a condensed answer to a query, accompanied by source links. AI Mode enables a more conversational experience: users can ask follow-up questions in text, voice, or by sharing an image, and receive iterative responses in their language. Both features are now available to users searching in Akan, Amharic, Hausa, Kinyarwanda, Afaan Oromoo, Somali, Kiswahili, Wolof, Yorùbá, Afrikaans, Sesotho, Setswana, and isiZulu — accessible via the Google mobile app or a mobile browser, with no additional hardware or download required.
The practical change for users in those language communities is material. A subsistence farmer in Oromia — where Afaan Oromoo is spoken by an estimated 40 million people — can now ask a health or agricultural question in their first language and receive an AI-synthesised answer sourced from the web. A Hausa-speaking student in Kano or Niamey can use AI Mode to research a topic conversationally, without being forced through English. A Kiswahili speaker in Dar es Salaam querying a legal or financial question gets an AI-generated summary in the language they think in, not the language Google’s systems were originally optimised for.
The languages were selected based on actual search volume across sub-Saharan Africa — Hausa, Kiswahili, Yorùbá, Amharic, and Afaan Oromoo among them represent the continent’s highest-frequency search languages. That selection methodology is transparent and defensible. It also means languages with smaller digital footprints — which often correlates directly with the communities that have least access to digital infrastructure — get no upgrade.
WAXAL: A Different Philosophy, Same Company
The WAXAL dataset, released a month earlier, represents a distinctly different approach to African language AI investment. WAXAL is not a product Google is building for African users. It is infrastructure African developers can use to build products themselves. The 11,000-hour speech corpus — covering 21 languages including Acholi, Luganda, Hausa, and Yoruba — was collected by Makerere University in Uganda, the University of Ghana, and Digital Umuganda in Rwanda, among others. The critical governance term is ownership: the data belongs to those African institutions, not to Google. It is published under an open licence on Hugging Face.
The dataset includes approximately 1,250 hours of transcribed speech for automatic speech recognition training and more than 20 hours of studio-quality recordings for text-to-speech synthesis. For a researcher or startup wanting to build a voice assistant, a health information system, or a fintech onboarding flow in Luganda or Acholi, WAXAL provides the training data that has historically been the primary barrier to building these tools. That barrier had been exploited by commercial data vendors who charged African developers for access to African language data collected from African communities — a dynamic that WAXAL’s community-ownership model explicitly disrupts.
BETAR reported on WAXAL’s sovereignty implications in February. The March search expansion is a related but separate event. Together, they are the most substantive African language AI investment from any technology company since Masakhane’s founding in 2018.
The Gap Between 13 and 2,000
Africa has more than 2,000 languages. Thirteen now have AI Overviews. Twenty-one have open speech data. The arithmetic is stark, but the distribution is the more important analysis.
The 13 languages in Google’s AI search expansion are the continent’s highest-traffic digital languages — populations with existing smartphone penetration, data connectivity, and established online communities. Hausa at 90 million speakers, Kiswahili at an estimated 200 million, Yorùbá at 50 million. These are not marginal populations. Including them is meaningful. But the languages they reach are already, by definition, the ones with the most existing representation in AI training data. The communities furthest from AI-generated information access — speakers of lower-digital-footprint languages in rural areas — are precisely those whose language search volumes did not clear the threshold for inclusion in Wave One.
Igbo, with 45 million speakers across southeastern Nigeria, is not in the 13. North African Arabic — across Morocco, Algeria, Egypt, Libya, Tunisia — is not in the 13, despite North Africa representing roughly a third of the continent’s population and internet users. Lingala, lingua franca across the DRC and Congo-Brazzaville, is absent. The absence of these languages is not an editorial failing by Google — it is a reflection of a genuine constraint. Building reliable AI Overviews in a language requires enough high-quality training data in that language to produce accurate, trustworthy summaries. Languages with limited digital text corpora produce AI hallucinations at rates that make deployment harmful rather than helpful. The WAXAL project is, in part, the infrastructure investment that makes a future expansion to those languages technically possible.
What the Market Signal Means
Google does not roll out AI product features without commercial intent. AI Overviews in African languages means Google Search is preparing to surface more advertising inventory in those language markets — and it signals to the advertising and e-commerce sector that African-language digital audiences are large enough to justify AI product investment. That is a market validation signal that African language tech startups, media companies, and advertisers should read carefully.
For African fintech specifically, the AI Mode voice and text query capability in Hausa, Kiswahili, and Yorùbá opens a user-experience frontier that has been difficult to access previously. Loan application guidance, insurance product explanation, mobile money troubleshooting — all of these customer interaction flows become AI-addressable in the user’s first language at a Google Search integration level, rather than requiring the fintech to build the language capability into its own app. The effective cost of African-language customer support for digitally-distributed financial products drops when Google’s AI is capable of handling first-contact queries in Hausa or Kiswahili.
Two Models, One Continent
The February–March Google sequence presents African governments, researchers, and startup founders with a clear-eyed view of what tech company African language investment looks like in 2026. WAXAL is Google investing in African language infrastructure it does not own — a genuine capacity-building move that reflects meaningful community consultation. AI Overviews is Google building African language into its own product stack, where the value capture stays with Google and the benefit to users is real but the sovereignty question is unresolved.
Neither model is wrong. Both are happening simultaneously. The productive policy and business question is not which model to prefer — it is how African institutions, researchers, and companies use the open infrastructure from WAXAL to build language AI that operates outside Google’s product ecosystem, while also capturing the near-term user access benefits of a Google AI search that now speaks Amharic and Kiswahili.
The gap between 13 and 2,000 will not close through goodwill. It will close through the sustained production of training data in the remaining languages — which is exactly what WAXAL’s community-ownership model is designed to enable. The timeline depends on whether Africa’s research institutions and tech startups treat the February 2 dataset release as the end of the story or the beginning of the infrastructure programme it was meant to start.
