Planning for growth: rethinking Africa’s grid assumptions in the age of data centres

The global growth of AI infrastructure is creating a new type of electricity demand. The hyperscale data centre is making decades of grid planning assumptions outdated. Based on insights from a high-level expert webinar organised by the Energy Council of South Africa and the Africa Sustainable Energy Centre (ASEC), this article explores how sudden, concentrated demand is putting strain on grids in North America, and what African power systems in South Africa, Ghana, Kenya, and Egypt need to urgently address before this wave reaches the continent.

The end of predictable demand

For most of the last three decades, electricity demand in developed countries grew slowly and was fairly easy to predict. Power companies could estimate future demand several years ahead and plan investments in power plants, transmission lines, and distribution networks accordingly. However, this situation is changing rapidly.

The rise of artificial intelligence (AI), cloud computing, and digital financial services has created a new type of electricity user: large-scale data centres. A single modern hyperscale data centre can use between 100 megawatts and 1 gigawatt of electricity continuously, about the same amount of power needed by a medium-sized city. These facilities can be built and become operational within 18 to 24 months. In contrast, the transmission infrastructure needed to supply them with electricity often takes 5 to 10 years to plan, approve, finance, and construct.

This difference between the rapid growth in electricity demand and the slower pace of power infrastructure development is creating pressure on electricity grids, increasing investment risks, and challenging regulators in the United States and Europe. Although Africa is still at an earlier stage of data-centre development, it is already attracting interest from major operators, particularly in countries such as South Africa, Ghana, Kenya, and Egypt. The key issue is no longer whether this growth will reach Africa, but whether African power systems will be prepared to support it when it arrives.

The data centre demand shock: what has fundamentally changed

Between 1990 and 2023, approximately, annual load growth in North America averaged a modest 0.5% to 1.5%, enabling conservative yet reliable long-range planning. Since 2024, this pattern has collapsed. The exponential growth of AI workloads, with each large-language model query consuming orders of magnitude more compute than a traditional web search, has driven a step-change in data-centre energy consumption. Hyperscale campuses originally designed for 100 megawatts have subsequently scaled to 500 megawatts or more, presenting grid operators with obligations they had not anticipated in their integrated resource plans.

“The thinking process and assumptions that we are mostly assuming for the grid have totally changed just because of this increment in data centres.”

– Dr Elvis Tuamasi

The challenge is made even greater by where data centres are located. Instead of being spread evenly across a country, they are often concentrated in a few specific areas. In the United States, regions such as the Ashburn–Fairfax–Loudoun County corridor in Northern Virginia (known as “Data Centre Alley”), Columbus in Ohio, and the Dallas–Fort Worth area host most of the country’s hyperscale data centres. Northern Virginia is the world’s largest data-centre market. Because so much computing activity is concentrated in a few locations, the pressure on electricity grids is also concentrated there. This can lead to local problems such as voltage instability, reduced grid reliability, and shortages in network capacity, which can affect larger interconnected power systems.

Many experts initially believed that improvements in server technology would reduce the amount of electricity needed for computing. However, this has not happened. Although individual servers have become more energy-efficient, the number of servers being used has increased rapidly. At the same time, AI applications require far more computing power than traditional business software. For example, a typical desktop computer uses about 150–200 watts of electricity, while a high-performance AI server can consume between 700 watts and 1 kilowatt. Since a hyperscale data centre can operate tens of thousands of these AI servers at the same time, its total electricity demand becomes extremely large.

Why conventional grid planning frameworks are no longer fit for purpose

A central finding of the expert discussion is that hyperscale data centres cannot be treated as ordinary electricity consumers within conventional planning frameworks. Several characteristics distinguish them fundamentally from legacy industrial loads:

Table 1: Distinguishing characteristics of hyperscale data centre loads and their grid planning implications.

There are two additional technical issues that power system planners need to address. The first relates to grid stability. Many data centres are now installing their own renewable energy systems, such as solar power and battery storage, at the same location as their facilities. While this helps reduce carbon emissions and energy costs, it can also make local electricity networks less stable. Traditional power plants, including gas, coal, nuclear, and hydroelectric stations, provide what is known as rotational inertia, which helps keep grid frequency stable when disturbances occur. Solar panels and batteries do not naturally provide this support in the same way. As a result, power system operators must require new technologies, such as synthetic inertia and advanced grid-forming inverters, to ensure that large data-centre developments do not weaken grid reliability.

The second challenge arises when data centres generate more electricity than they need and export the excess power back to the grid. Electricity networks were originally designed with the assumption that power would flow in one direction, from power plants to consumers. However, when data centres with large solar and battery systems begin supplying electricity to the grid, power can flow in both directions. This requires utilities to upgrade protection systems, voltage-control equipment, and grid management technologies so they can safely handle bidirectional power flows. These upgrades can be expensive because existing infrastructure was not designed for this purpose.

Given these changes, traditional planning methods that rely mainly on historical electricity demand data are no longer adequate. Experts recommend the use of digital twins – advanced virtual models of power systems that are continuously updated using real-time sensor data. These tools allow planners to simulate and assess the impact of large new electricity users, such as data centres, before they are connected to the grid. This enables utilities and policymakers to make informed investment decisions in advance, reducing the risk of future grid problems and costly emergency interventions.

Cybersecurity: the hidden dimension of grid digitisation

The cybersecurity vulnerability of modern electricity grids predates the data-centre boom. The shift to digital protection, control, and communication systems began in the 1990s and accelerated through the 2000s. The 2015 and 2016 cyberattacks on Ukrainian distribution utilities were the first publicly confirmed grid cyberattacks in history, demonstrating that a digitised grid is an attackable grid, regardless of data-centre penetration.

“Cyber risk should be treated as operational risk for utilities. Once there is a cybersecurity problem, what is that going to look like for us operationally?”

– Dr Jennifer Appiah Kubi

Large hyperscale data centres create new and more complex cybersecurity risks for electricity grids. A single data-centre campus can consume up to 500 megawatts of electricity at one location while also operating large amounts of on-site energy resources, such as solar panels and battery storage systems. To improve grid reliability, some regulations now require these energy resources to be remotely controlled by grid operators. This allows operators to reduce electricity demand or increase power generation during emergencies. However, this remote control capability also creates a cybersecurity vulnerability. If hackers gain access to the system that manages these energy resources, they could disconnect large amounts of electricity generation and demand at the same time, potentially causing widespread disruptions and threatening grid stability.

Experts also highlighted a major weakness in how many data centres approach resilience planning. Data-centre operators often invest heavily in physical backup systems, such as multiple power supply connections, duplicate transformers, and emergency generators, to ensure continuous operation during equipment failures. However, these physically separate systems are frequently managed through a single cybersecurity control platform. This means that if the control system is compromised by a cyberattack, all the backup systems could fail simultaneously, creating a single point of failure despite the physical redundancy.

To address this risk, cybersecurity protections should be designed to match the level of physical redundancy. This includes using separate control systems for critical assets, maintaining backup systems that can operate independently of network connections, and implementing strong authentication measures for all remote access to grid and data-centre infrastructure. Such measures help ensure that a cyberattack on one system does not compromise the entire operation.

Regulatory and tariff innovation: lessons from North America

Existing tariff and interconnection frameworks were not designed for the scale, speed, or sensitivity of hyperscale data centres. North American regulators are actively developing tailored instruments for this category of consumer. Three innovations stand out as particularly relevant for African policymakers:

• Take-or-pay capacity obligations. Dominion Energy (Virginia) and Georgia Power now require large data centres to pay for a minimum percentage, typically 85 to 100 per cent, of contracted transmission and generation capacity, regardless of actual consumption. This prevents operators from securing capacity commitments and then migrating to other jurisdictions, leaving utilities with stranded assets ultimately recovered from ordinary consumers.
• Cost-causation tariffs. Texas Senate Bill 6 establishes the principle that entities whose demand triggers new transmission infrastructure must pay for it in proportion. This logic prevents cross-subsidisation of large industrial loads by residential ratepayers.
• Bring Your Own Power (BYOP). PJM Interconnection introduced a BYOP framework in early 2026, enabling data-centre operators to co-locate generation on or adjacent to their campuses. In exchange, their load is treated as a lower-priority interruptible supply during grid emergencies, creating a powerful incentive for on-site generation investment while contributing dispatchable capacity to system operators.

The direction of travel is clear: data centres are transitioning from large passive loads to hybrid grid participants, subject to market obligations, emergency dispatch authority, and proportional infrastructure cost recovery.

Africa’s finite window: the leapfrog opportunity and its limits

Africa is not on the periphery of this transition; it is in its path. South Africa, Ghana, Kenya, and Egypt have been identified as priority markets by hyperscale operators seeking to serve growing African cloud demand, and interest from both global and regional digital-infrastructure investors is accelerating. The financial sector’s digitalisation in major African economies adds a further layer of domestic data-centre demand that will manifest over the next decade.

“Not having very well-established grids in Africa is a little bit of an advantage for us, because unlike in the US, where they are now having to change multiple things, we can start with a grid that has been created to be digital from day one.”

– Dr Jennifer Appiah Kubi

African power systems are being asked to accommodate these new load categories at a time when baseload reliability, transmission capacity, and institutional capacity are already under strain. But the continent’s relatively underdeveloped legacy infrastructure also creates a structural advantage. North American and European utilities face the costly, technically complex task of retrofitting digital capabilities onto systems built for a purely analogue paradigm. African utilities and regulators, by contrast, have the opportunity to build grid-enhancing technologies, such as dynamic line rating, wide-area monitoring, advanced distribution management systems, and digital protection, into infrastructure from inception.

This leapfrog opportunity is time-limited. Once legacy infrastructure decisions are locked in, retrofitting becomes exponentially more expensive. Policymakers and utility boards must act decisively within the current investment cycle.

Country-specific contexts vary. South Africa has the continent’s most developed transmission infrastructure but faces acute near-term pressures: the Transmission Development Plan rollout is critical, and the Eskom and NERSA regulatory frameworks require urgent updates to accommodate special-category load tariffs and take-or-pay obligations. Ghana, with its growing technology ecosystem in Accra and integration with the West African Power Pool, is well-positioned to pilot BYOP incentive frameworks. Kenya’s geothermal resources provide reliable, low-carbon baseload ideally suited to the constant, high-intensity demand profile of hyperscale operations. Egypt’s strategic position as a digital gateway between Europe, the Middle East, and Africa makes it a priority market for subsea cable landing stations and associated data-centre infrastructure.

Across all four markets and the continent more broadly, regional power pool integration is essential. Individual African countries will struggle to guarantee the spinning reserve margins and redundancy architectures that data-centre operators require without interconnected grid infrastructure. The Southern African Power Pool and the West African Power Pool are not peripheral considerations; they are prerequisites for credible data-centre investment attraction at scale.

Three priorities for planners, regulators, and investors

The expert webinar converged on eighteen discrete recommendations across three stakeholder groups. Distilled to their core, three priorities stand out:

For system planners

• Reclassify hyperscale loads immediately. Develop dedicated planning categories, load models, and interconnection standards. Remove data centres from generic industrial-load frameworks and incorporate large-load concentration scenarios into N-1 and N-2 contingency analysis.
• Invest in digital twin platforms. Pilot real-time simulation environments in at least one control region per country by 2027, enabling planners to model the dynamic effects of large new loads before connection.
• Design for bidirectionality from the outset. Ensure all new transmission and distribution assets procured for data-centre service corridors can accommodate reverse power flow.

For regulators and policymakers

• Introduce bespoke tariff categories that reflect the unique cost-causation profile of hyperscale data centres, including capacity reservation charges and take-or-pay obligations that protect residential ratepayers.
• Incentivise BYOP through regulatory frameworks that allow co-located generation to participate in capacity markets and emergency dispatch programmes.
• Strengthen cyber governance by mandating utility participation in national and regional information-sharing platforms, and requiring adoption of IEEE 1547-2018 standards as a baseline for all distributed energy resource interconnection.

For data centre investors and operators

• Engage early with national system operators and transmission utilities at the earliest stage of site selection, providing realistic load forecasts and legally binding capacity commitments.
• Build cyber redundancy that mirrors physical redundancy. Separate control domains, air-gapped fallback systems, and layered authentication eliminate single points of cyber control failure.
• Participate proactively in capacity and ancillary service markets rather than waiting for regulatory mandates. Early voluntary participation builds credibility and reduces the risk of punitive retrospective obligations.

Conclusion: anticipate, integrate, collaborate

The intersection of artificial intelligence, cloud computing, and digital finance is creating a new class of electricity demand that challenges foundational assumptions of grid planning, cybersecurity, and regulatory design. Hyperscale data centres are not a future possibility for African power systems – they are an imminent reality.

The North American experience – voltage events, near-miss cascading failures, emergency regulatory legislation, and evolving tariff structures – provides an invaluable roadmap for what to anticipate and, crucially, how to prepare. Three core principles emerge from this analysis.

Anticipation: System planners must incorporate data-centre demand scenarios into all long-range planning exercises today, not when the first 500-megawatt campus applies for grid connection. Integration: cyber resilience, physical resilience, and commercial governance must be designed as a unified system, not separate technical silos. Collaboration: no single actor can manage this transition alone. Structured information-sharing, co-investment frameworks, and joint contingency planning are prerequisites for a managed and equitable transition.

Africa has a finite window to act proactively rather than reactively. The institutions that move now, embedding smart-grid architecture at the foundational level, building bespoke regulatory frameworks, and engaging data-centre investors on terms that protect the broader consumer base, will be the ones that transform digital infrastructure demand from a grid liability into a grid asset.

illuminem Voices is a democratic space presenting the opinions of leading Sustainability Thought Leaders, their views do not necessarily represent those of illuminem.

The world needs sustainability knowledge. At illuminem, no interest group or shareholder can influence our work. Thank you for supporting our mission to make high-quality and independent sustainability information free for all. Every contribution helps. Thank you for donating today.