South Africa’s electricity debate usually begins with Eskom.
Coal stations. Transmission lines. Load shedding schedules. Municipal debt.
But outside the cities, another energy problem has been growing quietly for years. In many rural communities, electricity is either unreliable, too expensive to extend, or still absent altogether. For some regions, the economics no longer make sense. Extending the national grid into sparsely populated areas can cost more than R300,000 per kilometre.
That creates a hard reality for planners. Some communities may wait decades before conventional grid infrastructure reaches them properly.
Researchers presenting at the Southern African Universities Power Engineering Conference believe decentralized solar microgrids controlled by artificial intelligence systems could offer another route.
The proposal is straightforward.
Instead of depending entirely on long-distance transmission infrastructure, rural communities would generate electricity locally through solar photovoltaic systems paired with battery storage. AI systems would then manage how electricity is distributed, stored and conserved throughout the day.
The technology matters because rural energy systems are difficult to balance manually. Solar production changes with weather conditions. Household demand spikes during evenings. Batteries degrade if managed poorly. Diesel generators often fill supply gaps, but fuel costs quickly raise electricity prices.
Traditional rule-based systems react after shortages appear. The researchers tested whether machine learning systems could make decisions earlier.
Using Long Short-Term Memory forecasting models, the system predicted solar generation and electricity demand patterns ahead of time. Reinforcement learning systems then adjusted battery charging, backup generation and electricity distribution continuously.
The operational changes were significant. Diesel runtime dropped by 30%. Electricity costs fell from R7.80 per kilowatt-hour to R5.50. Battery lifespan improved because charging cycles became more controlled. Carbon emissions also declined sharply.
But the study’s most important contribution was political.
The researchers introduced fairness constraints directly into the energy management system after recognizing that electricity shortages are rarely distributed equally across communities. In many off-grid systems, lower-income households absorb repeated outages while better-positioned households remain relatively protected.
Read Also: You can’t digitise a blackout
The AI system tracked how load shedding was shared between households and reduced unequal curtailments substantially. That changes how electrification itself is measured.
For years, governments have treated electricity access largely as a connection problem. Build infrastructure. Link households. Count connections.
But access without reliability, affordability or fairness creates another version of inequality. South Africa’s rural electricity challenge is increasingly exposing the limits of centralized infrastructure planning. Eskom remains essential to the national grid. But rural electrification may eventually depend on smaller localized systems operating closer to communities themselves.
The deeper issue is that electricity systems are becoming software systems. Energy distribution is no longer just about wires and substations. It is increasingly about forecasting, data management and automated decision-making.
That creates new questions. Who owns the operational data collected from households? Who maintains these systems locally?
Who controls the algorithms deciding how electricity shortages are shared? The researchers argue that South Africa will need technical training pipelines through TVET colleges if rural communities are expected to manage increasingly digital energy infrastructure themselves. Because electrification is about institutional capacity.
By Thuita Gatero, Managing Editor, Africa Digest News. He specializes in conversations around AI and energy.
