Africa Has 207 AI Startups. How Many of Them Own Their Intelligence?

The continent's AI boom is overwhelmingly built on foreign models, and investors paying in rands should ask what remains if the supplier changes the terms.

Sign up for my Substack daily AI newsletter here.

Article link: https://johanosteyn.substack.com/p/africa-has-207-ai-startups-how-many?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

Walk into any investor pitch in Johannesburg, Cape Town or Lagos this year, and the deck will carry the letters A and I somewhere near the title slide. These companies raise money in rands or naira and buy their intelligence in dollars, and that mismatch is quietly rewriting the economics of the continent’s fastest-growing technology sector. For South African investors in particular, the gap between a defensible company and a well-marketed layer over someone else’s model is now the single most consequential thing a due diligence process can measure.

CONTEXT AND BACKGROUND

The number of AI-native startups tracked across Africa doubled to two hundred and seven across seventeen countries in three years, according to a landscape brief that places AI firms at the centre of a digital economy projected to reach one hundred and eighty billion dollars. The headline growth is encouraging. The composition of that growth is more complicated. Most of these companies are application-layer businesses that call frontier models through an interface, which means their intelligence is rented, their cost base is set in dollars, and their roadmap is subordinate to decisions taken in San Francisco.

The economics make the problem concrete. A survey of roughly three hundred software executives found AI product gross margins averaging about fifty-two per cent in 2026, well below the seventy-five to eighty-five per cent that software has delivered for decades, with the compute share of total spending rising as these companies mature rather than falling. For a South African founder, that structural compression lands on top of currency exposure. Every token used to serve a Johannesburg customer is bought in a currency that has lost ground to the dollar for most of the past decade.

INSIGHT AND ANALYSIS

The bigger risk is competitive. A Brookings analysis documents how three providers controlled close to ninety per cent of the enterprise model market by the end of 2025, and how one of them launched a coding product that competes directly with Cursor, a company whose business was built on that same supplier’s models. The pattern now extends into legal and design software, where frontier labs are moving into the territory of their own enterprise customers. For an African startup building a thin layer on a foreign model, the question is not whether a competitor will emerge; it is whether that competitor will be the supplier itself.

I have previously written about this when I examined how four Chinese open-source models closed the price gap with the frontier at a fraction of the cost, making self-hosting a Chinese model on local infrastructure a credible alternative to renting intelligence by the token. The strategic point for African founders is the same now as it was then. The companies that will survive the next funding cycle are those that own something the supplier cannot replace with a software update.

IMPLICATIONS

For South African investors, the due diligence conversation needs four questions. What is the gross margin after inference costs, and what happens to it if the model provider doubles prices? What proprietary data trains or steers the system, and is that data legally defensible? What breaks if the underlying model is swapped for a cheaper or open-source alternative? What remains of the product if the supplier releases the same workflow as a native feature in its next version?

For founders, the message is sharper. Some South African builders are already working against the dependency problem. Refiant AI raised five million dollars in April for technology that compresses AI models so they can run more cheaply, led by a California climate fund. That is a company that owns a layer of the stack. It is a template, not an exception.

CLOSING TAKEAWAY

The African AI pitch deck is a confident document. Behind the confidence sits a question that most founders cannot yet answer well and most investors do not yet know to ask. What does this company own when the model provider lowers prices, raises prices, or ships your product? The startups that carry the next ten years of this market will not be the ones with the slickest interface over a borrowed intelligence. They will be the ones that own data, workflows, or technology the supplier cannot delete. In a continent with small funds, expensive compute and a currency that buys less tomorrow than it did yesterday, that is not idealism. It is risk management.

Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net