Africa’s AI Ambition Has an Electricity Bill. Who Should Pay It?
We have learned to applaud the wrong number.
A minister stands beside an executive. A figure with nine zeros glows on the screen behind them. The photographs are taken. The headline almost writes itself: another billion-dollar technology investment is coming to Africa.
And perhaps we should applaud. Africa needs more data centres, more compute and more of the infrastructure that will allow our businesses and governments to participate seriously in the AI economy.
But I think we are focusing on the wrong electricity question.
The obvious conversation is already happening: AI needs enormous amounts of power, and Africa needs to generate more electricity if it wants to compete.
I am interested in the question underneath that one.
What if the electricity demand created by Africa’s AI boom is not merely another infrastructure cost to finance, but a new economic asset we can use?
A large, credible data-centre operator needing power for the next fifteen or twenty years is not simply a consumer of electricity. Its future demand can potentially help make new generation investable. The power plants, transmission, storage, fibre and other infrastructure built around that demand can themselves become long-lived assets. And if billions of dollars of those assets are going to be created, we should be asking who finances them, who owns them and who receives the returns.
That is why the headline value of the data-centre investment is not enough.
Africa’s data-centre demand is expected to grow from about 0.4GW today to between 1.5GW and 2.2GW by 2030 — roughly three-and-a-half to five-and-a-half times the current level. McKinsey estimates that the data-centre shells alone could require $10 billion to $20 billion of new investment, while creating a potential $20 billion to $30 billion revenue pool across the broader value chain. (McKinsey)
And the pipeline is becoming visible. Nigeria, for example, has a proposed two-site data-centre development beginning at a combined 60–70MW with room to expand to 100MW, including high-density GPU infrastructure intended for AI workloads. (USTDA) This month, ministers and officials from 45 African countries met with Korea and the African Development Bank around an agenda explicitly centred on AI and digital infrastructure; participants themselves identified electricity and data-centre shortages among the constraints to Africa’s ambitions. (African Development Bank)
So the compute is coming.
The question is whether Africa merely supplies it with electricity or uses its arrival to create something larger.
I think three things matter.
First, project economics are not system economics. A billion dollars invested in a data centre may require substantial additional investment in power and infrastructure outside the facility.
Second, electricity demand is not only a cost. A large, credible customer willing to buy electricity for years can itself become an asset that helps make new power projects financeable.
Third, if AI demand creates billions of dollars of generation, storage, fibre and digital infrastructure, we should ask not simply who pays for those assets, but who finances them, who owns them and who earns from them over the next twenty years.
The rest of the discussion follows from those three points.
The Billion-Dollar Investment Is Only Half the Economics
The investor in a data centre has a fairly clear problem to solve. Is the location competitive? Is there connectivity? Can customers be secured? Is sufficient power available at a viable price? Does the project produce the return its investors require?
The country hosting that investment has a bigger equation.
What does making the project possible require from everything around it? Perhaps sufficient power already exists. But perhaps additional generation needs to be developed, transmission capacity expanded, a substation built or storage added. Those investments may never appear in the headline number announced at the launch, but they are still part of the economic reality created by the project.
That is what I mean by project economics versus system economics.
A $1 billion investment tells us something important about the project. It does not tell us what additional investment the country, utility or other investors may need to make around it, whether those assets serve anyone else, who ultimately owns them or what return the wider economy receives.
That distinction matters particularly on a continent where the electricity investment challenge already exists without AI. Nearly 600 million Africans still lack access to electricity, and the International Energy Agency estimates that achieving universal electricity access by 2035 would require around $15 billion of investment every year. Current financing is far below that level. (IEA)
This is not an argument that a megawatt used by a data centre could simply have been redirected to a household. Electricity systems are more complicated than that, and different users require different infrastructure.
It is an argument that Africa is entering an extremely power-intensive technology investment cycle while already facing a significant electricity investment challenge. We should therefore understand the two capital requirements together.
Malaysia provides a useful illustration of why.
The country has been remarkably successful in attracting global data-centre investment. But that success has rapidly become an energy-system issue. Data centres accounted for around 7% of Malaysia’s electricity consumption in 2026 and reached 9.3% during a particularly hot period in August. Their share could reach as much as 31% of Peninsular Malaysia’s electricity consumption by 2035 if the existing pipeline develops as expected. (Reuters)
Malaysia has not made a mistake by attracting data centres. It has attracted precisely the kind of high-value digital investment many countries want.
But success changed the nature of the question.
Malaysia’s investment authorities are now talking not simply about attracting more capacity, but about local supply chains, skilled jobs, innovation, renewable energy and the broader value the industry creates. The conversation has moved from “How much capital can we attract?” to “What economy are we building around that capital?”
Singapore encountered its own version of this problem earlier. As resource constraints tightened, it became more selective about new data-centre growth. Its Green Data Centre Roadmap now aims to release at least 300MW of additional capacity, with further expansion linked to green energy and substantially stronger efficiency expectations. (IMDA Singapore)
Africa does not need to copy Malaysia or Singapore. Our power systems, capital markets and development priorities are different.
But we have one important advantage:
we can see some of their second-generation questions before most of our first-generation investment has arrived.
The Electricity Demand May Be More Valuable Than We Think
This is the part of the argument that interests me most.
We naturally look at a 100MW data centre and see 100MW of additional electricity demand.
A power developer can look at exactly the same number and see something else: a customer.
Imagine trying to raise hundreds of millions of dollars to build new generation. An investor will want to know who will buy the electricity and whether those revenues are sufficiently predictable to justify putting capital into the project.
A large, creditworthy customer prepared to buy substantial amounts of electricity over many years can materially change that conversation.
That means future AI demand potentially has financial value before a single server is switched on.
And it changes the question from:
Where will Africa find enough electricity for AI?
to:
Can AI’s appetite for electricity help make new African electricity infrastructure investable?
South Africa gives us a concrete example.
Teraco, one of Africa’s largest data-centre operators, is building a 120MW utility-scale solar plant in the Free State. It will own the facility and wheel the electricity through Eskom and municipal networks to its data centres. The company explicitly links the investment to the rapidly increasing power requirements of cloud and AI computing and to South Africa’s generation constraints. (Teraco)
Teraco has also contracted wind power to complement that supply.
The lesson is not that every African data centre should arrive carrying its own power station.
It is that future digital demand can help bring new electricity supply into existence rather than simply becoming another claimant on existing supply.
That is an important distinction.
The AI-electricity conversation often treats the data centre as the problem and the electricity system as the thing that must somehow catch up. But a major new electricity customer can also strengthen the commercial case for investment in generation.
Under the right conditions, demand is not simply what infrastructure must serve.
Demand is part of what makes infrastructure financeable.
That seems to me a potentially important opportunity for a continent that has spent decades asking how to mobilise more private capital into electricity infrastructure.
But a Forecast Is Not a Contract
There is an obvious complication.
A company’s announcement that it may eventually require 100MW is not the same thing as that company committing to pay for 100MW for fifteen years.
Someone has to carry the difference.
This is where the conversation stops being a purely technological one.
The proposed Microsoft-G42 investment in Kenya illustrates why. The original announcement envisaged a $1 billion digital investment programme anchored by a geothermal-powered data centre. Reporting this year said negotiations had become caught up in questions around guarantees for contracted capacity, with the project’s size and power requirements potentially being reconsidered. Reuters noted that it could not independently verify Bloomberg’s underlying report, so it would be wrong to present the eventual outcome as settled. (Reuters)
But regardless of how that particular negotiation concludes, it exposes the right question.
If significant infrastructure is built around the expected demand of one large customer, who carries the risk if that demand changes?
Technology evolves. Projects are resized. Locations change. Energy efficiency improves. Capital becomes more expensive. Another market becomes more attractive.
A utility, government or financier cannot build long-lived infrastructure around every announcement as though future demand were guaranteed.
I am not going to pretend to tell a government how to price a power-purchase agreement or capacity guarantee. There are specialists whose professional lives are spent doing exactly that.
But you do not need to be a project-finance specialist to recognise the strategic issue:
projected demand and committed demand are economically different assets.
The greater the infrastructure built around a future customer, the more important it becomes to understand who has actually committed to what, for how long, and who carries the downside if reality changes.
The risk exists whether we discuss it or not.
Someone holds it.
The Bigger Opportunity May Be Ownership
If AI demand does unlock significant new infrastructure, a second opportunity appears.
Someone will have to finance it.
And somebody will own it.
Africa’s AI build-out will not produce only data centres. It can create or accelerate investment in generation, energy storage, transmission, fibre, data-centre real estate and other long-lived assets.
These assets need capital.
They also generate returns.
Sophisticated investors have already begun finding ways to finance African digital infrastructure. IFC’s investment in Raxio, for example, forms part of a $293 million programme for ten data centres across African markets including Uganda, Ethiopia, Tanzania, Mozambique, Angola, Côte d’Ivoire and the DRC. The programme combines $88 million in equity, $165 million in senior debt and $40 million in subordinated debt, with blended-finance capital helping close part of the funding gap. (IFC)
That tells us something useful.
We are already becoming sophisticated about financing the digital asset.
We should become equally interested in financing and owning the assets being created around it.
This is where I think African capital enters the conversation.
A pension fund does not need to speculate on which AI model will dominate five years from now. An insurer does not need to become a venture capitalist betting on GPUs.
But long-duration electricity infrastructure, fibre, data-centre real estate and appropriately structured contracted infrastructure are different propositions.
Africa50’s Infrastructure Acceleration Fund provides one indication that African institutional investors are willing to participate in infrastructure when suitable vehicles exist. Its first close raised $222.5 million, primarily from African institutional investors, against a $500 million target, with investment areas including power, energy and digital infrastructure. (Africa50)
This does not mean African pension funds should automatically finance the AI boom.
It means we should broaden the question.
If AI demand is going to create billions of dollars of long-lived infrastructure assets across Africa, how much of the ownership and long-term return can African capital participate in?
For years we have discussed the continent’s infrastructure gap primarily as a shortage of money.
The AI boom may create an interesting inversion.
Could a new infrastructure demand also create a new class of investable African assets?
Could African savings become African infrastructure ownership?
That is a very different economic proposition from simply hosting somebody else’s servers.
The Headline Number Is Therefore the Beginning, Not the Answer
None of this is an argument for slowing Africa’s AI infrastructure ambitions.
Quite the opposite.
Africa needs more compute. It needs stronger cloud infrastructure, lower latency, greater resilience and the ability for African companies and public institutions to build and run AI services closer to the markets they serve.
But if this build-out is coming, we should become more sophisticated about what constitutes a good outcome.
The billion-dollar announcement tells us how much somebody proposes to invest in a project. It does not tell us whether that project’s demand helps unlock additional electricity supply. It does not tell us how much supporting infrastructure is required outside the facility. It does not tell us who carries the risk if expected demand changes. And it does not tell us who owns the infrastructure and receives its returns twenty years later.
Those questions are not peripheral to the AI investment.
They are part of its economics.
That is why I would judge the coming African AI infrastructure boom through three lenses.
What are the system economics behind the project economics?
Can the electricity demand itself help unlock more infrastructure?
Who will finance, own and earn from the assets that demand creates?
Malaysia, Singapore, South Africa and Kenya do not give Africa one answer.
They are useful because they reveal different parts of the same problem.
Malaysia shows what happens when successful data-centre attraction starts affecting the wider electricity system. Singapore shows how investment attraction can eventually become resource allocation. South Africa shows how digital demand can be linked to additional generation. Kenya reminds us that future demand has financial value only when somebody is prepared to stand behind it.
The point is not to import their answers.
It is to ask their questions earlier.
There Is a Board Question Hidden Here Too
African banks, insurers, pension funds, DFIs and infrastructure investors are likely to see more opportunities connected to this build-out.
Some may finance data centres.
Others may finance generation, fibre, storage or supporting infrastructure.
The AI label should not obscure what those investments really are.
They remain propositions involving long-duration cash flows, customer concentration, construction risk, energy prices, foreign exchange, technology assumptions and views about future demand.
A board or investment committee does not need to understand the engineering of a data centre.
It does need to understand where the durable economic value sits, what has to remain true for that value to materialise, and which risks its institution is being asked to carry.
AI may be new.
Capital allocation is not.
Africa’s Electricity Bill Could Become Economic Leverage
That brings us back to the photograph.
The minister.
The executive.
The billion-dollar number on the screen.
That number matters.
But perhaps it is not the most interesting number in the room.
If Africa can use the electricity demand created by AI to help unlock additional generation, if the infrastructure built around that demand creates investable long-lived assets, and if African capital can participate appropriately in owning some of them, then the electricity bill attached to our AI ambition begins to look very different.
It stops being simply another cost that the continent has to find a way to absorb.
It becomes economic leverage.
So when the next billion-dollar AI infrastructure investment is announced, I would still applaud.
Then I would want to know three more numbers.
How much additional infrastructure will it unlock?
How much of that infrastructure will African capital own?
And how much long-term economic value will still be here after the cameras have gone?