Artificial Intelligence Data Centres, Electricity Demand and Grid Planning in Africa

Artificial Intelligence Data Centres, Electricity Demand and Grid Planning in Africa Ahmed Odefunso

Abstract Artificial intelligence (AI) is intensifying attention to the electricity requirements of data centres. The International Energy Agency (IEA) projects substantial global growth in data-centre consumption, but such global projections do not specify the load, timing or grid effects in any particular African country. Africa’s planning context includes growing digital infrastructure alongside major electricity-access and financing challenges. The available research data gives a continental estimate of future data-centre demand, but this should be treated as a projection rather than an observed outcome, and it does not establish a uniform African grid condition. This article analyses the mechanisms through which AI-related load may affect electricity systems: concentrated demand, connection queues, local network capacity, uncertainty over workload and efficiency, and the lead times associated with infrastructure. It argues for scenario-based planning that brings digital and energy decisions together, tests projects against local capacity, and distinguishes firm demand from potential flexibility. Policy options include transparent load information, location-sensitive connection planning, demand response where technically feasible, and careful assessment of dedicated generation. These measures must be weighed against electricity access priorities and country-specific evidence rather than justified by global statistics alone.

Introduction Data centres convert digital activity into physical electricity demand. AI adds urgency to this relationship because accelerated computing and the expansion of AI services can increase the power requirements of facilities. The consequence for an electricity system depends not only on total consumption, but also on where a facility connects, how quickly its demand grows, the capacity of nearby networks and the timing and variability of its load.

Global evidence is useful for establishing the scale of the issue, but it cannot be treated as a forecast for Africa. The IEA’s global outlook projects that data-centre electricity consumption could rise from roughly 460 terawatt-hours in 2024 to about 945 terawatt-hours by 2030 in its base case. Those estimates describe a global aggregate. They do not mean that African consumption will grow at the same rate, nor do they identify the distribution of future demand among African countries.

The research data offers a separate Africa-focused estimate: demand associated with data centres is projected to rise from about 0.4 gigawatts in 2026 to between 1.5 and 2.2 gigawatts by 2030. This is a projection, not an observed load measurement, and the range signals uncertainty. It is not appropriate to multiply or otherwise combine it with global estimates to infer a share of worldwide consumption. The analytical task is to consider how planning can respond to plausible local growth while preserving attention to access, reliability and affordability.

Background/Literature Review The IEA’s global analysis attributes rising data-centre electricity demand primarily to increased computing requirements, including AI. Its base-case projection places global consumption at 945 terawatt-hours in 2030, while sensitivity cases vary according to AI adoption, efficiency and other constraints. This range of outcomes is important: projected demand is not a settled fact, and efficiency improvements may be offset by greater use or more demanding workloads. Any infrastructure decision based on one forecast alone risks either inadequate provision or costly excess capacity.

The research data also highlights the geographic concentration of data centres. A relatively modest share of global demand can create substantial pressure where facilities cluster on a particular part of the network. Transmission and distribution constraints are local, whereas global statistics average across regions. In this respect, the relevance of AI load is not determined solely by its percentage of world electricity use. A large, concentrated connection can affect local planning even if its contribution to national or global demand appears small.

The Africa-specific evidence provided in the research data describes a growing data-centre market and identifies grid reliability and infrastructure capacity as important constraints. It discusses different strategies, including renewable supply arrangements in Kenya and South Africa and dedicated gas generation in some contexts. These examples are not evidence that one approach is appropriate across the continent. Nor do they establish the extent of AI-specific load in each country. The data also identifies a risk that dedicated infrastructure could become economically vulnerable if an anchor customer relocates or demand changes.

Electricity access is a necessary part of the context. The IEA data provided states that roughly 600 million people in Africa lacked electricity as of 2024, while the World Bank-associated Tracking SDG 7 material gives a figure above 560 million for sub-Saharan Africa in 2024. These figures cover different geographic scopes and should not be conflated. Both indicate a substantial access deficit, but neither by itself measures the capacity available at a prospective data-centre site. Access and data-centre planning are related policy concerns, not interchangeable indicators of grid strength.

Analysis/Discussion The first planning issue is the character of load growth. A data-centre development can involve a large anticipated connection and a staged increase in demand. AI workloads may add intensity and variability, but the information supplied does not establish that every facility has the same operating profile. Planners therefore need project-specific data on requested capacity, expected commissioning stages and anticipated operating patterns. Treating a proposed maximum connection as a guaranteed, continuous load can distort forecasts; assuming that demand will remain flexible can be equally misleading.

The second issue is geographical concentration. A national supply outlook may appear adequate while a local transmission corridor, substation or distribution network lacks capacity. Conversely, a site with existing network headroom may be better placed to accommodate a new load than one requiring extensive reinforcement. The research data describes the IEA’s concern with localised grid stress and connection queues in global settings. The general mechanism is relevant to African planning, but the scale and location of constraints must be established through country and site-specific analysis.

Third, there is a mismatch in planning time horizons. Digital investment decisions can move rapidly, while transmission, generation and regulatory processes may take longer. The evidence supplied notes that infrastructure and permitting timelines can extend over many years in some global settings. It would be inappropriate to transfer a specific global timeline directly to African jurisdictions. The broader lesson is that planning should begin before demand becomes an urgent connection request, with the time required for local studies, authorisations and network improvements treated as an uncertainty to assess rather than a fixed universal duration.

Fourth, forecasts are sensitive to technology and commercial decisions. Efficiency gains may reduce electricity required for a given computational task, while broader adoption or more intensive applications may increase total consumption. The IEA sensitivity cases in the research data demonstrate that projections can differ materially according to assumptions about AI uptake and efficiency. For African systems, uncertainty is compounded by limited detail in the material supplied about the future geography of facilities and their workloads. A responsible plan should therefore test low, central and high demand pathways without presenting any one scenario as certain.

Grid reliability and access priorities make the policy question especially consequential. The access deficit does not mean that data centres should be excluded from electricity planning, nor does a proposed data-centre investment guarantee improved access for households. The evidence provided describes constraints in access financing and the need for investment in grid and decentralised solutions. Planning should make the opportunity costs and benefits explicit: who pays for network reinforcement, who bears reliability risks, and whether investments can provide wider system value. Claims that a facility will improve public access require evidence about its connection arrangements and any associated infrastructure, not an assumption based on its presence.

Integrated planning is a central option. Energy, digital and relevant transport infrastructure can be assessed together so that authorities understand where power and fibre capacity coincide. The research data reports a recommendation to identify zones with existing power and fibre capacity. Such an approach can direct development towards locations where additional network costs may be lower, but it should not be treated as a substitute for environmental, fiscal or community assessment. A designated zone is only useful if the underlying capacity is verified and the cost allocation is clear.

Connection rules and transparency are another leverage point. Utilities and regulators need timely information about proposed loads, commissioning schedules and expected operating needs. Publicly available aggregate reporting, subject to legitimate commercial confidentiality, can improve demand forecasting and help distinguish committed projects from speculative applications. Queue reform may improve the use of scarce connection capacity, but prioritisation criteria should be explicit and consistently applied. A data centre’s willingness to fund a connection does not by itself establish that the wider system can accommodate it without affecting other users.

Flexibility can contribute to system management, but its potential must be stated carefully. The research data describes temporal load shifting, spatial routing and batteries as possible tools. It also cautions that not all workloads can be shifted: real-time services may have latency constraints, while some training tasks may be more adaptable. These global findings do not quantify the flexible share of African data-centre demand. Any demand-response programme should identify which workloads can be curtailed, for how long, under what notice and with what safeguards for service quality. Planning credit should be given only to flexibility that operators can demonstrate and regulators can verify.

On-site generation presents a further trade-off. Dedicated generation may reduce reliance on an unreliable connection, but can create additional fuel, emissions and asset risks. The research data describes gas generation in some African markets and notes that a dedicated asset may be stranded if the expected customer or demand moves. Renewable supply arrangements also require assessment of their timing, firmness and relationship to the grid. A procurement claim about renewable electricity does not alone establish that a facility’s physical demand will be matched with supply at every hour. Policy should assess the full arrangement and avoid assuming either that on-site generation solves grid constraints or that every private supply arrangement benefits the wider system.

Finally, investment decisions should be compared with access and affordability priorities using transparent criteria. The IEA’s Africa analysis, as summarised in the research data, identifies a large annual investment requirement for universal access, while the World Bank material describes persistent regional access gaps. These figures concern access programmes, not the cost of serving data centres, and should not be combined as if they measured the same policy. They do, however, underline the importance of avoiding opaque subsidies or network commitments whose public costs are not disclosed. Public authorities can require clear responsibility for connection costs and review whether infrastructure investments have benefits beyond a single customer.

Conclusion AI data centres create a planning challenge because projected demand is uncertain, potentially concentrated and connected to infrastructure with local constraints. Global IEA projections describe worldwide growth; Africa-specific estimates in the research data describe a separate, uncertain trajectory and should not be extrapolated beyond their stated scope. Effective planning therefore requires local load information, scenario analysis, coordinated energy and digital infrastructure decisions, and transparent connection and cost rules. Flexibility and dedicated supply may help in particular circumstances, but neither should be credited without evidence about performance, emissions and system effects. The policy task is not to choose between digital investment and electricity access in the abstract. It is to test proposed developments against local capacity, distributional priorities and credible alternatives, while making uncertainties visible before they become infrastructure commitments.

References International Energy Agency. Energy and AI. https://www.iea.org/reports/energy-and-ai International Energy Agency. Africa Energy Outlook 2022. https://www.iea.org/reports/africa-energy-outlook-2022 World Bank and partner agencies. Tracking SDG 7: The Energy Progress Report. https://trackingsdg7.esmap.org/ International Energy Agency. Electricity 2025. https://www.iea.org/reports/electricity-2025 World Bank. Access to electricity, data indicator. https://data.worldbank.org/indicator/EG.ELC.ACCS.ZS