By Ola Adebayo
Every conversation I have about AI in Africa eventually arrives at the same place. Will Nigeria get its own sovereign large language model? Can Lagos afford GPU compute? Why does ChatGPT respond differently to a Yoruba prompt than an English one? These are real questions. Some of them are urgent. But I have come to believe that if we want the next generation of category-defining African technology companies to be built in AI, the model layer is not where most of our attention should sit.
The model layer is capital-intensive, dominated by American and Chinese incumbents, and, more importantly, too far from where the durable opportunity actually lives. The interesting thing happening in AI right now is not what is happening inside the models. It is what is happening around them.
In November 2024, Anthropic released a protocol called the Model Context Protocol, or MCP. It was a quiet release. Most people, including most builders, missed it. The protocol solved a specific and painful problem: every AI model that wanted to use external tools, say, to read a calendar, query a database, or send a Slack message, needed a custom integration with every one of those tools. With ten tools and three AI clients, you were maintaining thirty integrations. MCP collapsed that to a single shared protocol.
