Why Africa doesn’t need to build frontier AI models first—Experts

In recent years, the cost of using cutting-edge AI models has plummeted, making them accessible to startups and businesses in Africa. Mustafa Ehsan, founder of Convly, a research firm that tracks AI model performance, has observed this trend. A workload that would cost a Nigerian startup around $1,750 a month on a premium American model can now be run for about $21 on a cheaper alternative.

However, Ehsan argues that while model access has become less of a barrier, the real challenge lies in making AI useful in Africa's unique context.

Jephte Ioudom, founder of FoubsLabs, explored this challenge through a classroom project in Benin, where an AI mathematics tutoring platform was developed for pre-service mathematics teachers. The platform aimed to generate data that could help understand user patterns, shape AI products, and discover new use cases. Ioudom believes that owning the data layer is crucial for Africa's AI sovereignty, and he advocates for stronger data policy and infrastructure to prevent valuable data from being provided for free to foreign companies.

Both Ehsan and Ioudom agree that building a frontier AI model is not the first step for Africa. Instead, they suggest focusing on building the layers around it, such as data, talent, and infrastructure. Ehsan emphasizes the need for local financial, legal, and health data, as well as evaluation sets that test how well a model works for African languages and dialects.

Ioudom, on the other hand, stresses the importance of building the digital infrastructure and talent pipeline first, allowing the development of AI models to follow. He suggests that Africa should prioritize policies, education, and infrastructure investment before attempting to build its own frontier AI model.