Interview with Matija Vidmar: The Future Of AI For Africa
Matija Vidmar is an AI Solutions Architect and Software Engineer based in Trieste, Italy. As a freelance consultant through Evolbot, he helps SMEs adopt practical AI by analyzing business processes, identifying high-value use cases, and building clear adoption roadmaps.
He designs and delivers AI training programs for teams—covering tools such as Claude, Claude Code, and collaborative workflows—that meet the EU AI Act’s Article 4 AI-literacy requirements and can be funded through interprofessional funds. His technical work centers on custom solutions: AI agents, retrieval-augmented generation (RAG) systems, LLM integrations (Anthropic Claude and OpenAI), and function calling.
Matija single-handedly designed and built the Evolbot platform (evolbot.com), a multi-tenant SaaS for AI assistants featuring per-tenant RAG (document, website, and database ingestion), tool/action calling to external systems, Stripe billing, and multilingual support, developed in Python/FastAPI.
He is the author of the newsletter The AI Architect (1,300+ subscribers) and the course “From User to Orchestrator.”
Good evening Matija Vidmar. I am George Lucky Okello from Kisumu, Kenya currently interning at United Nations Association Boulder Colorado Chapter.
You are well versed in AI that is why I'm seeking your opinion on AI and it's impact on international development, developing world and prospects for Africa
How can AI help developing countries?
The honest answer is: mostly by making scarce expertise go further, not by inventing new industries. A developing economy is short of doctors, agronomists, teachers and loan officers — AI is very good at giving a first-pass version of those roles to people who currently get nothing at all.
A concrete Kenyan example: Jacaranda Health's PROMPTS service texts pregnant and postpartum mothers and uses natural-language processing to triage incoming questions, auto-answering roughly 86% and escalating the rest to human clinicians. A cluster-randomised trial with 6,139 women across 40 facilities in 8 counties found real improvements — a 7.4 percentage-point rise in women getting the recommended postnatal visit — but the overall effect sizes were small (0.06–0.09 standard deviations), and there was no measurable effect on seeking care for danger signs.
I'd hold onto both halves of that. AI produced a genuine, measured gain in a country where over two million women have now enrolled — and it was modest, not transformative. Most honest evidence in this field looks like that. Be sceptical of anyone whose examples don't come with an effect size.
2. Can AI help african businesses compete globally and how?
Yes, and there's already proof. InstaDeep was founded in Tunis in 2014 and sold to BioNTech in January 2023 for around $680 million — the largest African startup exit to date. That happened because the founders competed on machine-learning research quality, not on labour costs.
That's the pattern worth copying. The old export model was "we're cheaper" — and that's precisely the model AI erodes fastest (see question 4). The durable model is "we have something you can't get elsewhere": data, context, distribution, or a hard technical problem solved well. African firms hold assets nobody else has — agricultural and health data from environments no US or Chinese model has ever seen, and languages that essentially don't exist in today's systems.
The near-term commercial opportunity is less glamorous and probably bigger: an African SME using AI for customer support, translation, bookkeeping and export documentation now operates with a back office that used to require a dozen people. Under AfCFTA, that's the difference between selling in one country and selling in twenty.
3.What can african countries do to avoid being left behind in AI development?
Four things, in rough order of return on effort:
- Electricity and connectivity before algorithms. No AI strategy survives contact with an unreliable grid.
- Data, which is cheaper than compute and more defensible. Curated agricultural, health, financial and language datasets are the input every model needs and almost no lab has. Owning them is leverage; giving them away for free is not.
- Skills at the applied end, not just the research end. The scarce person isn't the one who can train a model — it's the one who can take an existing model and make it work inside a hospital, a bank, or a ministry.
- Governance that arrives early enough to matter. The AU adopted a Continental AI Strategy in July 2024, but as of mid-2025 only about 16 African countries had launched national AI strategies. Rules written after deployment tend to be written by whoever deployed.
I'd add a fifth, less popular one: pick a niche. No African country will win at frontier model training. Several could plausibly win at, say, agricultural AI, or African-language systems, or health diagnostics for tropical disease.
4.What is Africa’s biggest risk with AI?
Not robots taking jobs. It's that the development ladder gets pulled up. Every country that industrialised did it partly by selling cheap labour — textiles, then manufacturing, then business process outsourcing. AI attacks exactly that rung.
Africa's outsourcing sector employs over a million people and is projected to be worth $35 billion by 2028. A 2025 study by Caribou and Genesis Analytics with the Mastercard Foundation found roughly 40% of tasks in that sector could be automated by 2030, with only about 10% fully resilient. Entry-level roles are 68% of the workforce, and tasks performed by women are around 10% more exposed than those performed by men. That is squarely Kenya's growth story, and it is squarely the demographic — young, female, first job — that Colorado audiences usually assume AI will help.
The second-order risk is dependency: buying AI capability the way you'd buy fuel, so that the value, the data and the control all leave the continent. Which brings us to infrastructure.
5.What are the biggest challenges Africa faces in using AI?
Four, and they compound:
- Power. About 600 million Africans have no electricity access, and sub-Saharan Africa now accounts for around 86% of the entire global access deficit — up from 49% in 2010. It isn't closing; it's concentrating.
- Connectivity. The ITU put internet use in Africa at 36% in 2025, against a 74% global average. AI you can't reach is AI you don't have.
- Language and data. Of roughly 2,000 African languages, only about 42 are supported across the major models analysed in a 2025 survey — over 98% unsupported, with four languages (Amharic, Swahili, Afrikaans, Malagasy) absorbing most of the coverage. A model that can't hear you can't serve you.
- Capital. Africa has 20% of the world's people and receives about 3% of global energy investment — roughly $110 billion of $3.4 trillion in 2026. For scale: global energy investment into data-centre infrastructure alone exceeded $105 billion in 2025. The world spent about as much powering data centres as it spent on Africa's entire energy sector.
6.What prospects does AI hold for the african continent?
Large, conditional, and worth stating carefully. The African Development Bank's December 2025 report projects AI could add up to $1 trillion to Africa's GDP by 2035 — close to a third of current output — concentrated in agriculture (20% of the gain), wholesale and retail (14%), manufacturing (9%), finance (8%) and health (7%).
The IMF's 2026 work on sub-Saharan Africa makes the conditionality explicit: with serious investment in power, connectivity and skills, AI could lift regional output by around 4% over the next decade; without it, roughly 0.2%. That's a twentyfold difference driven entirely by choices, not by technology.
Please treat both as scenarios rather than forecasts. Their real value is the ratio: the gap between doing this well and doing it badly is far larger than the gap between adopting AI and not adopting it.
7.Does Africa have enough infrastructure to really welcome the AI age?
No — and I'd rather say that plainly than encourage you.
Africa has roughly 360 MW of live data-centre capacity against a global total of about 55 GW: around 0.6% of the world's capacity for about 20% of its people. The Africa Data Centres Association's own assessment is that even if every announced project is built, Africa's global share will hold steady rather than rise, because hyperscale expansion elsewhere is faster. South Africa alone hosts around 55 facilities; the IMF counts about 160 data centres continent-wide, with nearly half in South Africa, Nigeria and Kenya.
There is real movement — Cassava Technologies and NVIDIA are rolling out around $700 million of AI infrastructure, with 3,000 GPUs delivered in South Africa and 12,000 more planned over three to four years, expanding to Nigeria, Kenya, Egypt and Morocco. That matters, and it's still small against the gap.
But "not enough infrastructure to build AI" and "not enough infrastructure to use AI" are different sentences. Africa can use frontier AI over a phone today, and mobile money already showed that leapfrogging works when the last mile is solved. The strategic question isn't whether Africa can access AI — it's whether it will own any of it. Those require different investments, and only the second one requires data centres.
A closing thought for your Colorado audience
The interesting question about AI and development isn't "will it help or hurt." It's who captures the gain. The technology is genuinely useful, cheap at the point of use, and improving fast. But its benefits flow along existing infrastructure - electricity, connectivity, data, capital - and Africa is short of all four. Left alone, AI will most likely widen the gap rather than close it. That's not a prediction about AI. It's a prediction about infrastructure, and infrastructure is a policy choice.
Thank you Mr. Matija Vidmar for taking your time to answer this questions. It will really help the young people of Colorado,Africa and the world gain insights on AI.
Thank you!
