A new generation of artificial intelligence research institutions has taken root across the continent — in Nairobi, Johannesburg, Kigali, Lagos, and beyond. But follow the funding, and a dependency question emerges: who is actually paying for Africa’s AI future?
In a converted office block in Johannesburg’s Braamfontein, a small team at Lelapa AI is building what they call “AI for us, by us” — language models trained on African tongues like Zulu, Sesotho, and isiXhosa. Their Vulavula API is live. Their InkubaLM model, released in late 2024, benchmarks competitively against global models on African-language tasks. The pitch is unmistakably local.
The cheque, however, came from San Francisco.
Lelapa AI’s $2.5 million seed round — confirmed by Crunchbase and multiple independent sources — was led by Mozilla Ventures, with participation from Atlantica Ventures (a Lagos-Nairobi fund backed by the European Investment Bank) and angel investors Jeff Dean and Karim Beguir. Not a dollar of confirmed African private capital or government cash grant sits in the round. Pelonomi Moiloa, Lelapa’s CEO, has framed the stakes: “Language models created elsewhere lack an understanding of the local context. If there is anyone who should benefit from profit generated from a language — a cultural heirloom — it should be the people to whom that language belongs.”
Lelapa is not an outlier. Across six of Africa’s leading AI research institutions, BETAR.africa found total estimated funding exceeding $430 million. African governments and community contributors account for less than nine percent of that total. And every cent of that local share is in-kind: land, buildings, hosted compute, and volunteer hours. No African investor, government cash grant, or diaspora fund has made a confirmed direct cash investment in any of these six institutions.
Four cities, six institutions
The geography of African AI research is tighter than the continent’s scale might suggest. Nairobi hosts Google Research Africa’s research lab — opened in 2016 as Google Brain, rebranded following the DeepMind merger — and Microsoft Research Africa, the oldest corporate AI lab on the continent, operating since 2014. Kigali is home to AIMS Rwanda, the anchor campus of a pan-African network of postgraduate mathematical sciences institutes, and to CMU-Africa, Carnegie Mellon University’s campus on the continent, which has produced over 1,000 graduates since 2012. Johannesburg hosts Lelapa AI. Lagos is home to UNILAG’s AIRLAB and NAIL research units — and, as of October 2025, the first African campus of the OpenAI Academy.
Masakhane — a grassroots research collective focused on African-language natural language processing — operates pan-continentally, with contributors across more than 30 countries.
Together, these institutions represent the institutional core of Africa’s AI research base. What they produce varies considerably.
Google Research Africa publishes through Google’s global research organisation — dozens of peer-reviewed papers annually, from climate modelling to computational biology — but discloses no Africa-specific headcount or budget figures. Microsoft Research Africa (MSR Africa) operates a dedicated lab of roughly 20 researchers in Nairobi; MSR Africa affiliates have published across computer vision, machine learning systems, and health informatics, including notable work on automated retinal screening adapted for low-resource clinical settings.
AIMS is the oldest and largest in scope. Since its founding in South Africa in 2003, the network has expanded to six campuses — South Africa, Senegal, Ghana, Tanzania, Rwanda, and Cameroon — and graduated 3,177 master’s students as of 2023. Thirty-four point six percent of those graduates are women. The AIMS AI for Science Masters, a programme co-funded by Google DeepMind and launched in 2022, trains 40 scholars annually through July 2027.
CMU-Africa in Kigali, capitalised by a $275.7 million commitment from the Mastercard Foundation, is now one of the largest single investments in African graduate engineering education ever made. The institution reports that 85 percent of graduates work on the continent after completing their degrees — the strongest retention figure in the dataset.
Masakhane, founded in 2018, has taken a different path: an open, distributed research model built on volunteer labour and small grants. Its 50-language NLP initiative — building machine translation datasets and speech corpora for languages that global tech companies systematically underprivilege — has produced datasets now used in research globally. The 2025–26 grant round, supported by Google.org, the UK Foreign Commonwealth and Development Office, IDRC Canada, and the Gates Foundation, drew 93 applications from 22 countries.
In Lagos, UNILAG’s AIRLAB and NAIL research units represent a newer entry point — and the OpenAI Academy partnership, announced October 2025, signals growing international interest in West African AI research infrastructure. The partnership remains early-stage; it has not yet produced disclosed funding commitments or research outputs at the scale of the Nairobi or Kigali institutions.
Who is writing the cheques
Disaggregating the funding reveals a clear hierarchy of capital sources. International bilateral government programmes — principally the UK FCDO, IDRC Canada, and partner mechanisms — contribute the largest share of documented funding, driven largely by long-term AIMS programme commitments. International philanthropy (the Gates Foundation, the Mastercard Foundation, the Wellcome Trust, and Mozilla Foundation) accounts for a substantial portion, boosted significantly by CMU-Africa’s $275.7 million Mastercard Foundation capitalisation. Big Tech — Google and Microsoft — contributes an estimated share through lab operations and research partnerships, though both companies treat their Africa lab budgets as proprietary and figures are BETAR.africa estimates based on headcount and comparable lab benchmarks.
That leaves less than nine percent local. Of that, almost all represents host-government contributions to AIMS campuses: land, facilities, and operational hosting by Rwanda, Senegal, South Africa, Ghana, Tanzania, and Cameroon. The Masakhane fraction reflects volunteer hours and in-kind university hosting. African governments have contributed land and buildings. No African government has deployed cash for AI research at any of these institutions. No African sovereign wealth fund, no African private equity firm, no African philanthropic foundation at scale appears in the funding history of any of the six institutions reviewed. The Kigali AI Declaration, signed in April 2025, pledged a $60 billion Africa AI Fund — an aspiration that has not yet produced documented disbursements.
Moustapha Cissé, who founded Google AI Accra before leaving to build KERA Health Platforms in Senegal, identified the structural challenge in a peer-reviewed essay published in Nature in 2018: “If AI is to improve lives and reduce inequalities, we must build expertise beyond the present-day centres of innovation.” Seven years later, that expertise is being built — but the centres of capital remain the same.
Talent: who stays, and who leaves
The brain drain question resists simple answers. AIMS data, drawn from a 2024 study published in the European Mathematical Society Magazine, shows that 78 percent of employed AIMS graduates work within Africa. CMU-Africa’s 85 percent on-continent retention figure is higher still. Both figures, however, apply only to graduates who are working — and exclude the substantial share who pursue international doctoral programmes, of whom the majority are enrolled at universities in Europe or North America.
Whether that constitutes brain drain or brain circulation is contested. Researchers who leave often return; networks built in diaspora flow back as mentorship, grant introductions, and co-authorship. Masakhane’s distributed model is itself partly enabled by the diaspora: contributors in Berlin, Toronto, and London participate in African-language NLP because they maintain institutional ties.
The harder question is infrastructure. African AI researchers increasingly have access to global compute via cloud agreements — Google TPU research credits, AWS grants — but data centre sovereignty remains limited. Rwanda’s Kigali data centre, announced in a partnership with NVIDIA, is a notable step; it is not yet operational at scale. Training a large African-language model requires compute that no African institution currently owns.
The Google Research Africa residency programme, based in Accra and Nairobi, trains promising early-career researchers in a structured 12-to-18-month environment. Google has never disclosed retention or placement statistics. The gap is itself an editorial finding: an institution that trains at African institutional expense — reputational, competitive, relational — and discloses nothing about where the talent goes.
What comes next
The Nairobi AI Forum 2026, held in February, set a headline target of 45 million AI-linked jobs across the continent by 2035. Whether that ambition materialises will depend heavily on whether the current funding architecture evolves — from dependency to partnership, and eventually to something closer to African-capitalised research infrastructure.
For now, the most honest description of Africa’s AI research ecosystem is this: internationally funded, locally operated, and producing genuine research outputs and trained talent at growing scale. The question of who owns that output — whose interests it serves, whose compute it runs on, whose languages it learns — remains structurally unresolved.
The labs are real. The research is real. The money, for now, is mostly not African.
BETAR.africa used independent data from Crunchbase, Tracxn, Semantic Scholar, EMS Magazine (2024 AIMS alumni study), ICIAM institutional profiles, EIB press releases, and AI Expo Africa coverage. Funding estimates for Google Research Africa and Microsoft Research Africa are BETAR.africa estimates based on headcount and comparable corporate research lab benchmarks; both companies declined to disclose Africa-specific figures. CMU-Africa funding confirmed via Mastercard Foundation press release. Visualisations: Data & Visual Journalist, BETAR.africa.
UphivaBETAR
