Artificial intelligence is redrawing how business is done around the world, and Africa's AI startups, venture capital firms and technology ecosystems are no exception. The technology is advancing faster than most investors can build evaluation frameworks around it — working out how to judge an AI startup is hard enough; deciding how to fund one is harder still.

Over the past six months, TechCabal has put these questions to African investors spanning different regions, sectors and funding stages: where AI creates genuine value, where it falls short, and what deserves institutional capital.

The group includes Babacar Seck, founder of Askya Investment Partners; Eloho Omame, partner at TLcom Capital; Emmanuel Adegboye, head of Madica; Magnifique Ishimwe, fund manager at the Development Bank of Rwanda; Husein Merchant, investment officer at Village Capital; and Benjamin Toulouze, head of corporate venture capital at AXIAN Investment.

Asked what makes an AI startup fundable, they largely described something the underlying models cannot supply on their own. Anyone can now build on the same foundational models, so what investors are actually paying for sits around them: voice products that work in Wolof and isiZulu, software so deeply woven into a customer's systems that leaving hurts, licences that take years to win, and revenue.

Voice is where they see the clearest opening. Adegboye argues the opportunity lies in building on top of foundational models with local data, and above all in voice. He points to a man who speaks only Wolof: what is the easiest way for him to take part in an economy increasingly run by software? Seck draws the same conclusion from South Africa, where census data shows only 8.7% of citizens speak English at home.

Building software has never been cheaper either, and the count of AI startups has climbed as costs have fallen. TechCabal Insights tallied 207 African AI startups across 17 countries in 2025, roughly double the 104 it tracked in 2022. Most of the newest entrants work in software development, business intelligence, finance and education — precisely the areas where off-the-shelf models lowered the barrier most.

When code is cheap, investors hunt for what rivals cannot replicate. Omame compares the current moment to the arrival of mobile technology in Africa: once building the thing becomes easy, companies defend themselves through other means — brand, network effects, and the cost a customer incurs by pulling out software that runs their business.

The responses below are drawn from a series of interviews TechCabal conducted with investors between April and September 2026, curated by theme and edited for length and clarity.

Why Context Is an Advantage

Local context decides whether an AI tool works in Africa. Founders who possess it can build for specific problems — as Pewbeam AI has with software that projects Bible verses during live church services, and as YarnGPT has with a speech model covering Nigerian-accented English, Yoruba, Igbo and Hausa. They can also produce something as viral as Lagos Life.

Of the 207 African AI startups TechCabal Insights counted across 17 countries in 2025, two-thirds remain early stage and only eight have reached maturity. When that many companies can plug into the same foundation models, describing a product as AI-powered proves very little.

That shifts the questions investors put to founders. If AI keeps making software cheaper to write, software by itself protects nothing. What counts is everything surrounding it: data nobody else holds, local knowledge, a route to customers, regulatory permission, and how deeply the product sits inside the work a customer already does.

Merchant, whose fund has not yet examined AI as a sector, expects most of the African opportunity to sit at the application layer.

Husein Merchant (Village Capital): My personal opinion is that there is going to be a lot of opportunity on the consumer side. Given that Africa may be more of a consumer of AI technology than a creator of it, I think the investments and solutions will lean that way.

Eloho Omame (TLcom Capital): AI does not make everything cheaper to build. It is not a panacea. Some things are cheaper. Testing is cheaper, so you can experiment far more quickly with LLMs than before. Whenever a technology has driven down the cost of delivering something, the shape of defensibility has changed. A company could no longer rely on a high cost of entry. AI has lowered the cost of entry the way mobile technology lowered it before. When we look at a company, we ask how it builds defensibility if testing or delivery is now commoditised. It might be the brand. It might be network effects or switching costs.

Data and Language

Seck sorts AI startups into two camps. One builds the layers of AI itself; the other applies AI to ordinary businesses — and it is the second where he sees Africa's opening.

Speech data for African languages is scarce. A company that gathers it, trains on it and builds a customer base around the results owns an asset rivals cannot easily duplicate. Caantin, a Zambian startup that replaces call-centre staff with AI voice agents for banks and fintechs, told TechCabal in June 2025 that it was approaching $1 million in monthly revenue, billing by the volume of calls it handles.

Nobody has fully cracked the space yet.

Babacar Seck (Askya Investment Partners): Wherever you have structured data at scale, you can build small AI models that improve performance. That is exciting in African context because we are digitising in the age of AI. We are creating structured pools of data at the same time as we apply AI to them, so companies build a competitive advantage much faster as they scale.

Language is the starkest example of data that resists copying. South Africa's Census 2022 found that 8.7% of people speak English most often at home; isiZulu (24.4%), isiXhosa (16.3%), Afrikaans (10.6%) and Sepedi (10%) all rank higher. A product that works only in English serves a small slice of that market.

Babacar Seck (Askya Investment Partners): One very exciting area is voice AI in African languages. Many startups operate in that space, and none have fully cracked it. Today in Africa, most people do not speak English, French, or Portuguese at home. They speak their mother tongue, the language they are most comfortable in. The minute you have an application that lets companies talk to their customers in the customer's own language, you bridge a divide. That applies to companies and their customers, and to governments and their citizens.

Emmanuel Adegboye (Madica): I think daily about people in rural areas. What is the easiest way to take part in the global economy if you only speak Wolof? The fastest way is to build platforms that speak Wolof or deliver information in Wolof, so you can do all of that without needing to learn English. That is why I think voice is an important long-term opportunity in AI, and it is why some of our investments have been in voice AI.

Embedded in the Customer's Operations

Omame's portfolio has surfaced a second form of defence: a company woven into a customer's daily operations. Once its software is wired into thousands of end customers, a rival cannot easily swap it out — and the margin that integration earns is protected along with it.

Eloho Omame (TLcom Capital): We invested in an API infrastructure business. Its API plugs into its customers' software and delivers a fintech product. Before, each time the customer needed the product, it made an API call. Now, because of agentic technology, the system prompts the customer when they are about to run out of the resource they lease. Revenue goes up, and the company also knows its revenue is going up. The product also costs less to deliver, because there is an agentic layer on the production side too.

In principle, more people can now build what this company builds, as cheaply as it can. But once our software is talking to each other and I am delivering value across thousands of your own customers, it is much harder for you to unplug from my system.

I want a clear line of sight to how AI makes the product cheaper to create, and what happens to the margin that creates. The founder can pass it all to the customer or capture it in the business as a higher margin. As an investor, I want to see it captured in the business.

Revenue Comes First

Defensibility answers only half the question. The other half is whether the company earns money while investors are still willing to wait — and on this continent, the wait has grown shorter.

An AI company that spends three years on compute before charging anyone fails that test, however strong the idea.

Magnifique Ishimwe (Development Bank of Rwanda): Basic technology can always be built. I could put together an app like the one I showed you by the next morning. The fundamental question is whether you can execute.

I have done a couple of AI deals, and sometimes we cannot see when the company reaches its horizon. You can absorb two or three equity rounds, and we still do not see a path to profitability. In the United States you can be an Uber that loses money for six, eight, or ten years. In Africa, every investor asks where your revenue is, and if you do not have it, what value you have in its place.

Think of an AI company that builds a chatbot for students, but nobody is willing to pay for it. You spent a lot on compute, we helped you get credits, you paid expensive engineers, and for three years nothing came out of it. Those deals are misaligned with our instrument, because we are not an equity investor.

Some Investors Back AI Infrastructure

A minority do fund the deeper layers, generally for strategic reasons. AXIAN, a Madagascar-based group with telecoms and digital banking operations, frames its infrastructure bets around the question of who controls the data.

Benjamin Toulouze (AXIAN Investment): We want to invest in AI infrastructure, and we have already done so. One example is WideBot AI, a proprietary LLM system for Arabic speakers. We invested in Nucleon Security, a Moroccan company that protects devices using zero trust, EDR, and malware tools.

We are convinced about the sovereignty issue. African countries need autonomy and control over their own data. That is one of the biggest challenges in the African market, because it is fragmented across 54 countries.

Emmanuel Adegboye (Madica): There is also an opportunity in hardware and AI. That is the bet we are making with ChipMango and chip design, and we will see how it evolves.

Madica joined ChipMango's $1.9 million seed round, led by Atlantica Ventures, in September 2026. Even these infrastructure bets obey the same logic as the rest of this discussion. WideBot's edge rests on Arabic dialect data and a book of more than 350 government and enterprise clients. ChipMango recruits engineers through partnerships with Nigerian universities, trains them, and sells chip-design work to global customers. In every case, the difficult thing to copy is the data, the talent pipeline or the customer list.

The Cost of Inaction in AI Investing

If African investors back only the application layer, the deeper layers of AI will be financed, built and owned elsewhere. Global capital is already moving in that direction.

Capital is not the only shortfall. African data centres are customers even as the continent says it needs more of them, and universities are training AI engineers who depart for jobs abroad.

Emmanuel Adegboye (Madica): Capital is concentrating in a handful of funds and a handful of startups. Most funding globally is going to AI infrastructure companies, and only a few of those. On top of that, most of the funding coming into Africa comes from outside Africa, which exposes the continent to that risk. If I am a Silicon Valley investor who is not close to the continent, Africa is not top of mind when I invest.

Babacar Seck (Askya Investment Partners): Most data centres are chasing customers, yet we keep saying Africa does not have enough data centres. One gap is that these operators are not plugged into demand. When you speak to the global cloud players, they will tell you their biggest customers in Africa are startups, with some scale-ups spending upwards of $30 million a year on cloud.

We are training a growing number of AI engineers at Carnegie Mellon in Kigali, at the African Institute for Mathematical Sciences, and at some very good Nigerian universities, and many of them find jobs abroad because the companies on the continent are not yet large enough to keep them.

For now, the investors surveyed here have answered the question within their control: they will pay for the parts of an AI company that grow more valuable as the AI underneath becomes cheaper. Whether the continent also builds that underlying layer is a question their capital alone cannot settle.