Babacar Seck of Askya Investment Partners has a simple test for an AI startup worth backing: can anyone with a good prompt build the same thing in a weekend? If the answer is yes, the conversation is likely over before it begins. That question sits at the heart of how African investors are rethinking what fundable actually looks like in the age of generative AI.
Over six months, a group of investors shared their thinking with TechCabal on where AI creates real value across the continent. They include Eloho Omame of TLcom Capital, Emmanuel Adegboye of Madica, Magnifique Ishimwe of the Development Bank of Rwanda, Husein Merchant of Village Capital, and Benjamin Toulouze of AXIAN Investment. Their answers keep landing in the same place: the model itself is not the product. What sits around it is.
Voice is the clearest example. Adegboye points to a man who speaks only Wolof and asks what the easiest path into a software-run economy looks like for him. Seck raises the same point about South Africa, where census data shows only 8.7 percent of citizens speak English at home. Products that work in Wolof, in isiZulu, trained on local data, are not things a generic foundational model delivers out of the box.
The number of African AI startups has roughly doubled since 2022, from 104 to 207 across 17 countries, according to TechCabal Insights. Most of the newest entrants work in software development, business intelligence, finance, and education, exactly where falling costs made it easiest to enter. Omame compares the moment to the early days of mobile technology on the continent. Once building gets cheap, she says, companies survive on brand, network effects, and the pain a customer would feel ripping out software woven into how they operate.
That switching cost is the new moat. Licences that take years to earn, integrations too deep to unwind, revenue that proves someone has already made that choice. The investors still writing checks are betting less on the AI and more on everything around it that AI alone cannot replicate.
Originally published by TechCabal.