A $1 billion-plus digital infrastructure initiative by Microsoft and G42 in Kenya highlights the challenges of large-scale investments in regions with constrained power grids and unclear policy frameworks.
The 2024 proposal included a geothermal-powered green data center in Olkaria to establish Microsoft Azure’s East Africa cloud region, aiming to provide scalable cloud and AI services. However, the project was abandoned due to power infrastructure limitations and disagreements over risk allocation.
The initiative faced significant hurdles in scaling.
Power and Policy Obstacles
At GITEX Kenya, Sandy Okoth of Invest Kenya and the Tony Blair Institute highlighted the mismatch between the project’s energy demands and national capacity.
“The initial plan for 100MW expansion to 1GW was misaligned with Kenya’s current grid capacity of ~3,000MW peak demand,” Okoth noted, stressing that the facility would have consumed over half the nation’s power supply.
Risk allocation further complicated matters, as G42 sought government guarantees for full-scale implementation, potentially shifting liabilities to taxpayers.
Africa’s Infrastructure Gap
East Africa lags globally in data center deployment. Kenya leads the region with 19 facilities, while the continent hosts only 259 centers—less than 1% of worldwide capacity—forcing reliance on distant servers.
This dependency hampers digital economy growth and increases latency.
Decentralized Solutions Gain Traction
Snehar Shah of iXAfrica Data Centres advocates for leveraging Africa’s renewable energy potential to develop localized, efficient infrastructure. Traditional enterprise-owned server rooms are deemed costly and energy-intensive.
Modern facilities increasingly adopt advanced technologies, with Shah noting: “Solid floors and high-density power delivery outperform legacy underfloor systems.”
Challenging Megaproject Models
Stanislav Kazanov of Innowise warns that replicating Western hyperscale models risks overwhelming Africa’s grids. In Rwanda, a single 100MW AI facility could consume nearly a third of the country’s 406MW total capacity.
Distributed “edge” data centers near renewable energy sources—such as geothermal or solar—are proposed as alternatives to reduce grid strain and transmission losses.
Energy Costs and AI Workloads
Power scarcity drives up AI service costs in Africa.
Sarah Rees of Signwl reported that in May 2026, renting an NVIDIA H100 chip in Africa cost $13.55/hour—85% above the global average—due to limited supply and energy costs. Training GPUs consume up to 1,200 watts, compared to lighter inference chips.
Rees recommends prioritizing energy-efficient inference workloads to mitigate power constraints.
Building African AI Sovereignty
Experts emphasize that AI sovereignty demands local ecosystems. Trixie LohMirmand of GITEX Global stated: “Creating AI gives you power.”
Sarah Qian of S.I.G.N. urges prioritizing locally relevant datasets and applications, citing Africa’s mobile money innovations as a model for self-reliant systems.
Path Forward
While energy challenges persist, experts stress that Africa’s AI future lies in developing skills, localized datasets, and distributed infrastructure tailored to regional power realities.
The consensus favors scalable, modular systems over centralized megaprojects to ensure sustainable growth.