MTN Group plans to convert its African tower estate into a distributed AI compute fabric, installing open GPU infrastructure at base-station sites so that the same hardware can run both the cellular network and edge AI inference workloads.
The plan was set out by Charles Molapisi, group chief technology and information officer, MTN, at an event hosted by law firm Bowmans in Johannesburg recently— the company’s most detailed explanation yet of how it intends to position itself as the infrastructure layer of Africa’s AI economy.
Every cellular tower today has a baseband unit at its base — single-purpose hardware that exists only to drive the radio access network.
Molapisi said MTN will replace these with open GPU configurations capable of running the radio plus AI inference, in what the company has described as a “distributed AI grid.”
A key pay-off, he argued, is latency. AI workloads that today must be hauled back to a central data centre could instead be processed at or near the tower.
He gave the example of children playing PlayStation on an estate served by a nearby tower: with edge compute installed, the workload could be served locally rather than backhauled to a distant data centre and returned, freeing capacity and cutting round-trip time.
The edge layer sits alongside the centralized half of MTN’s AI infrastructure plan.
The group confirmed in its 2025 financial results in March that it will build two new AI-enabled data centres — one in South Africa and one in Nigeria.
Molapisi described an MTN AI strategy spanning a relatively full stack — procuring silicon, building data centres, running its own cloud platforms, curating models and co-developing applications with partners. The company is also building terrestrial fibre across multiple African markets, including some where it has no GSM licence, to plug what Molapisi called the continent’s missing “rails.”
The investments sit inside MTN’s Ambition 2030 strategy, which reorganized the group around three platforms: connectivity, fintech and digital infrastructure. The tower-to-inference push is the most concrete articulation yet of a thesis MTN has been laying out for more than a year — including an investment in March in U.S. AI-native networking start-up ORAN Development Company alongside NVIDIA, Cisco, Nokia, AT&T and Telecom Italia.
At the time, Mazen Mroué, CEO, Digital Infrastructure CEO, framed the move around “sovereign AI” — the principle that African countries should host AI compute locally rather than relying on offshore infrastructure.
Molapisi said MTN is developing the edge AI grid alongside technology partners, with the ambition for MTN to become “the biggest distributor of edge inference in the continent.”
The strategic case rests on Molapisi’s wider argument that Africa risks repeating its commodity history in the AI era.
With about 1% of global computing power on the continent today, he said, Africa stands to “export raw data” the way it has long exported raw minerals — only to import the intelligence built from it at a premium.
Molapisi conceded that chip generations are turning over quickly enough — NVIDIA’s Hopper to Blackwell inside two years, for example — that procurement decisions made today can be obsolete by deployment. He said MTN is being deliberate about its chip mix and the balance between training and inference silicon, “because if you get that wrong, you’ll get the economics terribly wrong.”
