Localized computing infrastructure vital for Africa’s AI future

Microsoft and G42’s planned digital investment in Kenya has become a case study in how large infrastructure deals can run into challenges tied to power availability, policy direction, and public interest. Unveiled in 2024, the proposed $1 billion-plus investment led by G42 and partners centered on a green data centre in Olkaria powered by geothermal energy. The facility was expected to support a new East Africa cloud region for Microsoft Azure, enabling scalable, secure cloud and AI services for businesses across Kenya and the wider region. The project, however, did not proceed. ALSO READ: Rwanda launches Digital Public Infrastructure strategy to power citizen services, AI Power constraints and scale concerns Speaking at GITEX Kenya, Sandy Okoth, director of the Investment Deal Room at Invest Kenya and an adviser with the Tony Blair Institute for Global Change, said the project’s scale quickly became a sticking point. “When you look at the Microsoft G42 deal, initially it was supposed to be piloted with 100 megawatts of power and then scaled up to one gigawatt,” he said. “Kenya is producing around 3,000 megawatts, and peak consumption is about 2,400 megawatts.” He warned that the project could have consumed a disproportionate share of national supply. “This single $1 billion investment was going to consume more than half the country’s power,” Okoth said, noting Kenya’s long-term target of expanding generation capacity to 10,000MW. ALSO READ: Africa’s infrastructure debate must shift The main obstacle, he added, was risk allocation. G42 reportedly sought government guarantees for the full scale-up, including sovereign-backed assurances, an arrangement Okoth said would have made taxpayers “the de facto liability bearers in a private sector investment.” Instead, the government pushed policy reforms such as time-of-use tariffs to shift heavy users to off-peak hours, and an open-access regime allowing private power purchase agreements using state transmission infrastructure. Africa’s limited data centre footprint Despite growing investment interest, East Africa remains a small player in the global data infrastructure map. Kenya leads the region with 19 operational data centres, followed by Tanzania with 11. Rwanda hosts three, while Uganda and DR Congo have four each. ALSO READ: New solar deal to help achieve 900MW off-grid target Across Africa, there are about 259 data centres, less than 1 percent of global capacity leaving much of the continent dependent on servers located thousands of kilometres away. This gap, analysts say, explains why Africa’s digital economy still relies heavily on external infrastructure. A shift toward localized infrastructure Snehar Shah, CEO of iXAfrica Data Centres, said Africa must leverage its renewable energy potential and connectivity growth to build more self-reliant systems. He noted that traditional enterprise-owned server rooms were costly and inefficient, requiring heavy investment in power, cooling, and hardware. ALSO READ: Why Rwanda needs a water policy for data centres now Centralised data centres, he said, offer scalable and more efficient alternatives. Shah added that newer facilities are increasingly better positioned than legacy operators. “Many existing players rely on older technologies like underfloor cabling,” he said. “We use solid floors and can deliver very high-density power.” He said some markets are already approaching capacity, creating opportunities for providers serving hyperscalers. A 2025 industry report estimates East Africa’s total live data centre load at just 30MW. Rethinking hyperscale models Stanislav Kazanov, a data engineer at Innowise, warned that Africa cannot simply replicate Western hyperscale models. He noted that large AI facilities could strain already limited grids. In Rwanda, for example, total generation capacity is about 406MW, meaning a single 100MW AI facility could consume nearly a third of supply. Instead, he advocates for distributed, smaller “edge” data centres located near energy sources such as geothermal or solar plants. “Instead of waiting for a 10-year grid upgrade, you can place modular data centres next to power plants,” he said, arguing this reduces transmission losses and infrastructure delays. High costs of AI compute Power constraints are also driving up the cost of AI services in Africa. Sarah Rees, CEO of Signwl, which tracks global cloud GPU pricing, said that in May 2026, renting an NVIDIA H100 chip in Africa cost about $13.55 (Approx. Rwf19,815) per hour—roughly 85 percent higher than the global average of $7.32. She attributes this price gap to limited supply: there simply aren’t many providers offering GPU servers in the region. Power is a big factor here. A cloud GPU used for training (creating AI models) can draw over 1,200 watts. By contrast, smaller inference GPUs (used for running or “inferring” models) may draw only about 72 W. Rees said African developers should prioritise lower-energy inference workloads where possible. “In a market where energy is tight, inference-focused strategies save substantial power,” she said. Toward African AI sovereignty Experts at the GITEX Kenya discussions stressed that building AI sovereignty requires local ecosystems, not just imported infrastructure. Trixie LohMirmand, CEO of GITEX Global, said: “Using AI makes you a market. Creating AI gives you power.” Sarah Qian, partner at Sign argued that Africa should build foundational layers of the AI stack—such as datasets, models, and applications—starting with locally relevant data. She pointed to Africa’s success with mobile money innovation like M-PESA as proof of what locally driven systems can achieve. Building sovereign datasets in local languages, she said, would allow universities and startups to develop AI tools tailored to African realities. The road ahead While energy constraints remain a major challenge, experts say the bigger opportunity lies in building skills, datasets, and distributed infrastructure now. The consensus is that Africa’s AI future will not be defined by a few massive data centres, but by many smaller, smarter systems designed around local power realities and needs.

News

Localized computing infrastructure vital for Africa’s AI future