The panic over artificial intelligence used to be simple, the machines would take the jobs. That fear has cooled faster than almost anyone predicted, with employers and labour trackers increasingly describing AI as a tool that reshapes tasks rather than deletes roles outright. A newer, more concrete contest has taken its place, not over who does the work, but over who gets the electricity to power the machines doing it. That contest is now shaping how countries compete for AI infrastructure investment.
AI is now the biggest new load on power systems
AI workloads are projected to consume 44 gigawatts of power worldwide in 2026, edging out the 38 gigawatts drawn by non-AI data centre workloads, as GPUs and other AI accelerators replace conventional servers with far more power-hungry hardware. The International Energy Agency projects global data-centre electricity use will more than double by 2030. Unlike most commercial power users, AI data centres need stable, high-density electricity with almost no tolerance for outages or fluctuations, a requirement that is reshaping where investors choose to build, and increasingly rewarding grids that can guarantee reliability over those that simply have generation capacity on paper.
That distinction is playing out most visibly in Africa, where the data-centre market remains small by global standards but is accelerating fast, and where the AI power fight is producing several distinct strategies.
Different grids, different bets
McKinsey projects data-centre capacity across Africa’s five largest markets will grow from around 400 megawatts today to between 1.5 and 2.2 gigawatts by 2030, requiring $10 billion to $20 billion in construction investment. A separate industry report put the continent’s operational capacity above 500 megawatts, with another 890 megawatts in the pipeline, though South Africa still accounts for more than 60% of that total.
South Africa’s advantage is scale and grid maturity. Cape Town alone has approved two hyperscale data centres totalling roughly 174 megawatts, among the largest single new loads any African electricity system has taken on. Africa’s Silicon Valley, Kenya, has instead anchored its pitch around clean, firm power, centring a Microsoft-backed data-centre initiative near Naivasha on geothermal generation, with an initial phase near 100 megawatts and an ambition to scale toward 1 gigawatt. Nigeria has taken a third path, building AI infrastructure largely independent of its national grid. A 20-megawatt data centre under development in Ogun State requires a dedicated 100-megawatt gas plant to run standalone, a choice driven by a grid that has never delivered more than roughly 5,801 megawatts to Nigeria’s 242 million people and that saw available generation fall to about 4,300 megawatts in February amid gas-supply disruptions.
Ghana illustrates a fourth variant of the same problem. A capacity that exists on paper but reliability that lags behind it. Installed generation rose from 2,165 megawatts in 2010 to 5,749 megawatts in 2024, consistently above peak demand, according to grid operator GRIDCo, yet the operator also reports annual demand growth above 10%, alongside persistent voltage and frequency instability.
Digital Realty’s Ghana site, running at about 1.7 megawatts, is a modest load by global standards, but it is still exposed to that instability. Ghana currently hosts about eight data centres competing for a West African pipeline estimated at roughly 440 megawatts, and analysts warn that without stronger transmission and distribution infrastructure, the country risks leaning more heavily on foreign hosting rather than capturing that investment locally.
The common constraint
What links these otherwise different markets is that none of their national power systems were originally designed around AI-scale, always-on industrial loads. Analysts tracking the sector describe African grids broadly as “power-constrained” for the AI era, not for lack of generation, but because few can yet deliver the stable, high-density, low-outage electricity hyperscale operators require. That gap is now a direct factor in where global capital lands: markets that can combine generation with demonstrated reliability, whether through geothermal in Kenya, grid scale in South Africa, standalone gas in Nigeria, or targeted solar and storage investment in Ghana, are the ones best positioned to turn AI’s power hunger into local economic gain rather than into another strain on an already stretched system.
The same tension is visible well beyond Africa. In the United States, utilities have already begun raising household electricity rates to fund grid capacity for data centres, and more than 300 state bills addressing the issue were introduced in 2026 alone, evidence that even mature, high-capacity grids are not immune to AI infrastructure outpacing the systems meant to support it. For emerging markets, the calculation is sharper still: the AI buildout will not wait for grids to catch up, and the countries able to prove reliability now, not just promise favourable investment terms, are the ones set to capture the next wave of digital infrastructure investment.
