Less Compute, More Reach: Africa’s Approach to AI

Running AI has a cost, and in African fintech it is paid in a currency the revenue does not come in. Computing is bought abroad; customers pay locally, in small amounts, over slow networks and on basic phones.

These conditions shape what gets built. They push toward smaller models, toward AI that runs on the device, toward services that keep working when the connection drops.

The keynote treats this as a design discipline. Software built where resources are scarce tends to be leaner, cheaper to run and easier to keep in local hands, which decides who can afford to use AI and who can afford to supervise it.

The last part weighs the trade-offs: what to run locally and what to rent, whom to build for, and what is given up in accuracy to gain reach. Constraint is part of the design.