At a conference in Seoul this month, African officials and Korean financiers kept returning to a number that sounds too large to be useful: $1 trillion. That is the African Development Bank’s estimate of the extra economic output artificial intelligence could add by 2035 if the continent builds the infrastructure, skills and rules needed to use it.
The more revealing figure came in the same room. Africa accounts for less than 2% of global data-centre capacity. The gap between those two numbers is where the next African technology story will be decided. It is not a story about chatbots arriving on smartphones. It is a financing story about electricity, fibre, cloud capacity, data, public procurement and the willingness of investors to fund assets whose returns may take years to appear.
The Korea-Africa Economic Cooperation meeting produced a new set of promises: a five-year cooperation framework between the AfDB and Korea Eximbank; a proposed Korea-Africa AI Hub; an agreement with Korea’s National Information Society Agency on data governance and AI-ready datasets; and a 2027-28 action plan covering digital infrastructure, energy, transport and skills. Those announcements matter because they connect AI to institutions that can prepare and finance projects. They do not, by themselves, create a bankable pipeline.
That is the test now facing the AfDB and its partners. Africa does not need another declaration that AI is important. It needs projects with identifiable customers, dependable power, clear data rules, credible operators and a route to repayment. The first winners may be less glamorous than model developers: fibre companies, tower operators, utilities, data-centre landlords, cloud providers, banks and contractors that can make digital infrastructure work outside a handful of capitals.
The financing story begins with hardware
The AfDB’s Seoul programme is a useful snapshot of where the money is supposed to come from. Korea’s bilateral trust fund has received $132.82 million in cumulative contributions as of May 31, 2026. About $50 million has gone into technical assistance and project preparation, helping support more than 40 operations worth over $6 billion and mobilising at least $4 billion in additional financing. The lesson is not that grant money can pay for Africa’s digital transformation. It is that relatively small preparation budgets can help turn an idea into something lenders can assess.
That conversion is often the missing step. A ministry may know that schools need broadband or that an industrial zone needs reliable data links. A private investor needs a demand forecast, land rights, power contracts, security arrangements, tax treatment, foreign-exchange assumptions and a procurement structure. It also needs to know what happens when a government changes its digital policy or when a cable is cut. Project preparation is not administrative overhead. It is the part of the capital stack that makes larger money possible.
The proposed Korea-Africa AI Hub is intended to bring governments, financial institutions, research centres, universities and businesses together around projects. The AfDB and Korea Eximbank agreement adds loans, guarantees and equity investments to the menu. The National Information Society Agency agreement points to data governance, AI-ready datasets, digital public infrastructure and training. Together, they describe a sensible architecture: technical expertise at the front, risk-sharing in the middle and commercial capital at the back.
But the architecture is still a framework. The AfDB has not published a project-by-project value for the Korea-Africa package, nor a timetable for disbursement. That distinction matters for markets. A memorandum can shape expectations; only signed financing, procurement and construction can change the supply of compute or connectivity.
Power is the constraint investors cannot code around
AI infrastructure is often discussed as if it were mainly a software question. In Africa, it is first an electricity question. Data centres need continuous power, cooling and redundant connections. Fibre networks need electricity at landing stations, exchanges, towers and operating sites. A cloud service can be sold globally, but a server still sits somewhere and draws power every minute.
The AfDB has said the continent’s AI opportunity depends on reliable energy, affordable connectivity, data infrastructure, skills and regulation. That list is more important than the headline GDP estimate. If power is intermittent, operators must spend more on diesel, batteries and backup generation. Those costs either raise prices for customers or reduce the return available to investors. In countries where industrial users already face high energy costs, a new data-centre load can become a political issue as well as a commercial one.
That is why the Korean discussions placed energy beside AI and digital infrastructure. A data-centre project that is financed without a power plan is not an AI investment; it is a stranded-asset risk waiting to be priced. The most credible projects will link compute to new generation, transmission upgrades, renewable power contracts or industrial parks with a dependable grid. They will also show how communities and existing businesses will share scarce capacity.
There is a temptation to assume that solar panels solve the problem. They can reduce fuel use and improve resilience, but high-performance computing requires firm power, storage and cooling. The financial model must account for all three. It must also account for water, land, equipment imports and the replacement cycle for servers. These are ordinary infrastructure concerns, but AI makes each one more expensive and more visible.
Connectivity is improving, but access is not the same as use
Internet access in Africa has expanded sharply, yet the region still has one of the widest gaps between network coverage and actual use. ITU data put Africa’s internet usage at 36% in 2025. The State of Broadband in Africa report says mobile broadband covered about 86% of the population in 2024, while a large usage gap remained among people living within reach of a network. GSMA has described the usage gap in sub-Saharan Africa as about 60%, driven by the cost of smartphones and data, limited skills and concerns about safety.
That gap changes the economics of AI. A public cloud or digital health service cannot be judged only by the number of people technically covered by a tower. It needs users with devices, affordable data, digital confidence and a reason to return. If those conditions are absent, operators carry the cost of capacity without the revenue needed to maintain it.
It also explains why the next phase of investment cannot focus only on flagship data centres. Africa needs backbone fibre, regional internet exchanges, resilient submarine cable routes, rural backhaul and cheaper devices. Nigeria’s D-VIBE project shows the scale involved. The AfDB approved a $200 million loan as part of an $800 million sovereign package for a plan to expand the national fibre backbone from about 30,000 kilometres to roughly 120,000 kilometres and connect all 774 local government areas. Total financing is estimated at $2 billion, including private investment.
That kind of project is less visible than an AI laboratory, but it may have a larger economic effect. It connects schools, hospitals, farms, factories and public agencies to the same digital base. It creates demand for cloud services rather than assuming demand will appear after a server hall is built. It also gives investors a clearer way to measure coverage, traffic and revenue.
The $1 trillion claim needs a balance sheet
The AfDB’s estimate of up to $1 trillion in additional GDP by 2035 is a scenario, not a forecast that governments can book in their budgets. The bank’s underlying report identifies agriculture, wholesale and retail, manufacturing, finance and health as the largest potential areas of gain. Those sectors together are projected to account for 58% of the total potential gain.
The number is useful if it is treated as a map of possible productivity improvements. It becomes misleading when it is used as a substitute for a business case. An agricultural model can reduce crop losses only if farmers receive usable advice, have network access and can act on it. A health system can use machine learning only if its records are digitised, its clinicians trust the tools and the data are governed properly. A factory can automate processes only if it has reliable power and a workforce able to maintain equipment.
The World Bank’s work on AI foundations makes a similar point. It groups the necessary base around connectivity, compute, context and competency. The exact labels matter less than the sequencing. Cheap broadband without usable data is limited. Data without compute is inert. Compute without skills becomes an imported service with little local spillover. Skills without customers produce graduates who leave or work for foreign firms.
This is where capital allocation will separate serious programmes from theatre. Investors should ask which part of the value chain a project serves and who pays. A government procurement contract can support an early digital service. A mobile operator can provide an anchor customer for edge computing. A bank can finance equipment for small firms. A regional data facility can lower latency for several countries. Each structure has a different risk profile. Treating them all as one AI sector will hide the real risks.
Africa can capture value, or simply rent it
Africa’s role in the AI economy is not predetermined. It can be a market for imported models, cloud services and hardware. It can also develop companies that adapt technology to local languages, crops, payment systems, logistics networks and public services. The second path requires more than startup funding. It requires access to data, affordable compute, procurement rules that allow local firms to compete and intellectual-property arrangements that do not strip value from the continent.
The AfDB has framed this as a question of ownership. Its president has said Africa must be able to build, govern, adapt and deploy its own solutions rather than simply consume imported technology. That is a sensible objective, but ownership has costs. Training models in local languages requires clean datasets and researchers. Operating them requires compute. Protecting users requires regulators and courts that can enforce rules. None of these arrives automatically with a foreign data-centre investment.
Critical minerals add another layer. The same Korea-Africa discussions linked AI with mineral processing and industrial value chains. Africa supplies many of the resources that feed modern technology, but extraction alone does not create a technology industry. The stronger proposition is to connect mining regions to power, transport, fibre, technical colleges and manufacturers. That would make digital infrastructure part of industrial policy, rather than a separate technology programme.
There are obvious risks. Foreign capital may build facilities that serve international traffic while local users remain priced out. Governments may offer tax concessions without securing training, local procurement or data access. Public agencies may buy systems that cannot interoperate. Local data may be exported for processing without clear rules about security or benefit sharing. And a rush to announce national AI strategies may produce overlapping regulators and incompatible standards.
What investors will watch next
The first indicator will be whether the AfDB and Korea Eximbank publish a pipeline with named countries, sectors, sponsors and financing instruments. A credible pipeline should show which projects are seeking grants, guarantees, debt or equity. It should disclose the expected private-sector contribution and explain how foreign-exchange and political risks will be managed.
The second indicator will be power. Projects that cannot demonstrate firm electricity, backup capacity and a credible cost path will struggle to attract long-term capital. Investors will also look for evidence that digital infrastructure is being planned alongside transmission, renewable generation and industrial demand.
The third will be procurement. AI projects funded by public money should be evaluated on measurable service outcomes, not on the number of software licences purchased. Open standards, data portability and transparent contracts will determine whether countries can switch providers or build on public systems. Closed platforms may deliver quick demonstrations and expensive dependence.
The fourth will be regional scale. Africa’s fragmented markets make it difficult for a provider to recover infrastructure costs country by country. Shared rules for data, payments, cybersecurity and digital identity could enlarge the customer base. Cross-border fibre and regional compute facilities could reduce duplication. The AfCFTA will matter here less as a slogan than as a mechanism for making digital services tradeable.
The fifth will be who gets paid. If the projected gains appear only in the valuation of foreign technology companies, the continent will have financed consumption rather than transformation. If local firms win contracts, workers gain skills, utilities receive investment and manufacturers use the infrastructure to export higher-value goods, the economics will look different. Those outcomes need to be written into project design and measured after construction.
The market is moving from vision to execution
Africa’s AI financing push is arriving at a useful moment. The continent’s mobile networks are expanding, operators are investing in cloud partnerships and development banks are looking for ways to mobilise private capital. Korea brings industrial experience, technical capacity and export finance. African governments bring markets, labour, data and urgent public-service needs. The AfDB can help connect the two sides and absorb some of the preparation and political risk.
That combination is promising, but it is not a guarantee. The continent has seen many large infrastructure announcements that took years to reach financial close or never reached construction. AI makes the danger greater because the technology changes faster than a conventional infrastructure loan can be repaid.
The sensible approach is to finance the foundations in stages. Start with reliable connectivity, power and public data systems. Build small, useful applications in agriculture, health, finance and logistics. Measure whether they reduce costs or improve services. Then expand the compute and cloud capacity that has proven demand. This may sound less dramatic than a $1 trillion forecast. It is also how the forecast could become more than a headline.
For African markets, the next few years will reveal whether AI becomes another imported cost or a productive asset. The answer will be visible in power contracts, fibre kilometres, loan books, procurement awards, factory output and the earnings of companies serving customers beyond the capital city. The Seoul agreements have opened the financing conversation. The hard part is turning them into projects that can survive contact with a balance sheet.
Banks will have to learn a new kind of infrastructure risk
Commercial banks are likely to sit between development finance and technology companies. They already understand project finance, working capital and public-sector risk. What they often lack is a long record of cash flows from data centres, cloud services and AI applications in African markets. That makes guarantees and subordinated capital especially important during the first wave of projects.
A bank financing a fibre network can estimate subscribers, wholesale prices and maintenance costs. A bank financing a specialised AI facility has to estimate utilisation, equipment obsolescence and the credit quality of customers that may not yet exist. The technology risk is not only whether a model works. It is whether the model remains competitive before the server debt matures.
Local-currency financing can reduce one problem and expose another. Borrowing in naira, shillings, rand or cedi protects a project from a sudden increase in dollar debt service, but revenues may also be in local currency and interest rates can be high. Dollar borrowing can be cheaper for a strong sponsor while leaving the project exposed to devaluation. Guarantees, blended finance and contracts indexed to a stable currency may be needed, but each protection has a cost that should be shown clearly.
The AfDB’s proposed use of guarantees and equity is therefore more consequential than a simple loan announcement. Guarantees can help a local bank lend against a new asset class. Equity can take early construction risk that senior lenders will not. But public risk-sharing should not become a quiet subsidy for weak projects. Sponsors should still carry enough risk to keep costs under control and to maintain the facility after the ribbon-cutting.
Public agencies are the first serious customers
Many AI applications in Africa will begin with government demand. Tax authorities need fraud detection. Health ministries need tools for stock management and diagnosis support. Agricultural agencies need weather and crop information. Courts, schools and municipalities need searchable records. These are practical uses, but they require governments to buy technology in a way that creates durable local capability.
Procurement can either widen the market or lock it up. A contract written around one vendor’s proprietary system may solve a short-term problem and make future changes expensive. A contract requiring interoperable data, local support, training and independent performance testing is harder to write, but it leaves a more useful asset behind. Development banks can make those conditions part of the financing rather than treating them as optional policy language.
Public procurement also creates a way to test whether the $1 trillion claim has substance. If a digital system reduces the time to process a business licence, improves tax collection without raising rates, or cuts medicine stock-outs, the benefit can be measured. If a chatbot produces impressive demonstrations but no reduction in cost or improvement in service, it should not receive another round of financing simply because it is labelled AI.
This is particularly important for smaller economies. They may not be able to justify a large national data centre, but they can share regional facilities and purchase computing as a service. A common procurement framework could allow several governments to buy secure capacity together. The same approach could apply to digital identity, cybersecurity and language datasets. Scale does not always require one enormous building; it can come from common standards and pooled demand.
Energy and data policy will decide where capital lands
Investors will compare African markets on more than tax incentives. They will look at the reliability of the grid, the price and availability of land, the speed of permits, the quality of fibre routes, the independence of regulators and the enforceability of contracts. A country with a smaller market can compete if it offers dependable power and a credible regional strategy. A large market can lose projects if companies cannot predict the rules or move equipment through customs.
Data policy will be just as important. Local storage rules can support domestic capacity, but poorly designed restrictions can raise costs and split markets into small, incompatible silos. Cross-border data flows require safeguards, not blanket barriers. Governments will need to decide which data must remain inside a country, which can be processed regionally and which can be handled by commercial cloud providers under contract.
There is a wider governance question around public datasets. Information collected by governments often has enormous value for AI, but it is also sensitive. Land records, health files, school histories and tax data cannot simply be released to the highest bidder. The rules must set limits on use, protect individuals and give public agencies the ability to audit systems. Trust is not a soft issue in this market. Without it, adoption stalls.
For the private sector, clear rules are a form of infrastructure. So is a regulator that can make decisions on spectrum, data protection, competition and cybersecurity without political uncertainty. The countries that establish those basics will be better placed to attract long-term capital than those that rely on the size of their populations alone.
A narrower opportunity may be a stronger one
There is a risk that Africa’s AI strategy becomes too broad. Every ministry, sector and startup is being told to prepare for an AI future, while basic digital systems remain incomplete. A more disciplined strategy would choose a small number of use cases where African conditions create a real advantage.
Local-language services are one example. A model that handles Swahili, Hausa, Amharic, Yoruba or isiZulu reliably could improve access to public information and financial services. But the opportunity depends on the painstaking work of collecting, cleaning and governing language data. It also requires local researchers who can test whether a system is accurate for different communities. The commercial value will come from trusted deployment, not from the existence of a model alone.
Another opportunity is the link between mobile money and small-business finance. Africa has large volumes of transaction data and a mature mobile payments industry. With consent and strong privacy protections, that data could help lenders assess firms that lack conventional financial statements. The same systems could support inventory finance and cross-border trade. The risk is that weak safeguards turn useful data into a source of exclusion or surveillance.
Industrial applications may offer the clearest route to productivity. Mines can use computer vision for safety and maintenance. Ports can optimise cargo movement. Manufacturers can monitor equipment. Power utilities can predict faults. These businesses already pay for efficiency, so they may become better early customers than consumers asked to pay for another digital subscription. They also offer a way to connect AI investment to exports, jobs and local supply chains.
That is the investment case African policymakers should present. Not a continent waiting for a miracle from software, but a group of markets where specific infrastructure and industrial problems can support specific technology businesses. The difference is subtle in a conference speech and decisive in a loan committee.
