MTN Group plans to replace the single-purpose hardware at its African tower sites with GPU configurations capable of running both the cellular radio access network and AI inference workloads from the same equipment, Group CTO Charles Molapisi said Wednesday.
Molapisi laid out the plan at an event hosted by law firm Bowmans in Johannesburg, calling it the most concrete articulation yet of how MTN intends to become “the biggest distributor of edge inference in the continent.” Every cellular tower currently carries a baseband unit dedicated exclusively to driving the radio network. MTN’s plan is to replace these with open GPU setups so the same hardware handles both the radio stack and local AI processing. The practical payoff is latency: instead of routing AI requests to a central data centre and back, the workload gets handled at or near the tower. Molapisi gave a specific example — gaming traffic from a residential estate served by a nearby tower could be processed at the tower site rather than backhauled to a distant server and returned.
The edge layer is one half of MTN’s AI infrastructure build. The group confirmed in its March 2025 annual results that it will also build two new AI-enabled data centres — one in South Africa and one in Nigeria. In March, MTN co-invested in ORAN Development Company, a US AI-native networking startup, alongside Nvidia, Cisco, Nokia, AT&T, and Telecom Italia. The group is also laying terrestrial fibre in markets where it currently holds no GSM licence, targeting what Molapisi called Africa’s missing connectivity “rails.”
The case for the tower strategy rests on a structural concern Molapisi named directly: Africa today holds roughly 1% of global computing power. Without its own AI infrastructure, the continent risks repeating its commodity-extraction pattern — exporting raw data and importing processed intelligence back at a premium. The tower-to-inference push sits inside MTN’s Ambition 2030 strategy, which reorganised the group around connectivity, fintech, and digital infrastructure. MTN operates across more than 17 African markets, giving it a tower footprint no cloud-only player can match; the edge compute plan essentially treats that physical presence as a distributed data centre estate rather than just a mobile network.
Molapisi was candid about the execution risk. GPU generations are cycling fast — Nvidia moved from its Hopper architecture to Blackwell in under two years — meaning procurement decisions made now could be obsolete by deployment. “If you get that wrong, you’ll get the economics terribly wrong,” he said. MTN is being deliberate about its chip mix and the balance between training and inference silicon as a result. The company has not disclosed a deployment timeline or total investment figure for the tower-edge programme.
About Dignited Staff
This account is managed by the in-house team at Dignited, a collective of passionate tech writers, editors, and enthusiasts dedicated to bringing you the latest insights, reviews, and news on consumer technology. For inquiries or feedback, feel free to reach out to us at [email protected].
Discover more from Dignited
Subscribe to get the latest posts sent to your email.
