There is a familiar rhythm to technology funding in Africa.
A new technology emerges. Its possibilities become visible. Funders issue calls for innovation. Governments announce pilot programmes. Entrepreneurs are invited to demonstrate solutions in health, agriculture, education, financial inclusion or public administration.
For a period, there is excitement.
There are hackathons. Challenge funds. Innovation labs. Demonstration projects. Perhaps a chatbot for farmers, an early-warning system for disease outbreaks, or an AI tool intended to help teachers manage overcrowded classrooms.
Some of these ideas are useful. Some may even work.
But we need to ask: What must exist underneath these projects for them to become reliable, equitable and sustainable public systems? And: Who is willing to pay for that?
Because public-interest artificial intelligence requires electricity that is reliable enough to run data centres and research facilities. It requires affordable computing capacity. It requires datasets that are relevant, well-governed and representative of the people the system is supposed to serve. It requires African-language resources, public research institutions, regulatory agencies, technical standards, independent auditors and people with the authority and expertise to question how systems are designed.
These are not peripheral investments.
They are the conditions that determine whether Africa becomes a producer and governor of artificial intelligence—or primarily a market in which systems designed elsewhere are deployed.
We are often funding what can be demonstrated
Philanthropy and development finance have long been drawn to visible, time-bound interventions.
A pilot can be funded for twelve or eighteen months. It can have deliverables, beneficiaries and a final report. A funder can point to the number of farmers reached, health workers trained or schools connected.
The backbone ecosystem is more difficult because it may take years to build. Its impact is often indirect. It serves many actors rather than one neatly defined beneficiary group. Its success may be visible precisely because nothing dramatic happens: the data remains secure, the regulator performs its work, researchers gain consistent access to computing resources, or a public institution is able to assess an automated system before it harms people.
This creates an understandable but dangerous bias.
We fund what can be demonstrated before we fund what makes sustained delivery possible.
We fund AI applications without ensuring that the public institutions deploying them can procure, audit or govern them.
We fund data-collection exercises without adequately supporting the long-term stewardship of the data.
We support language-model experiments without building enduring language repositories, standards and community governance structures.
We fund entrepreneurs to create tools, but not always the shared infrastructure that would allow hundreds of entrepreneurs and researchers to build without starting from zero.
The result is an ecosystem full of promising activity, but with weak foundations.
Compute is becoming a development question
Access to computing power is increasingly one of the dividing lines in artificial-intelligence development.
Researchers and small firms need computing resources to train, test and adapt models. Governments need secure infrastructure to host sensitive systems. Universities require high-performance computing to participate meaningfully in research rather than simply consume tools built elsewhere.
Yet Africa continues to hold only a very small share of global data-centre and high-performance-computing capacity. A 2025 IFC investment of $100 million in the Raxio Group was described as the institution’s largest digital-infrastructure investment of its kind in Africa, supporting data-centre expansion across several countries. The investment is important. But its scale also tells us something about how early and underdeveloped this infrastructure landscape remains.
New research mapping Africa’s AI-ready computing environment points to recurring barriers: limited high-performance-computing access, dependence on foreign cloud providers, payment-system incompatibilities, exchange-rate volatility, energy constraints and fragmented approaches to data sovereignty.
These are not just technical problems.
They shape who gets to experiment, whose research is possible and which organisations can afford to build.
A well-funded company may purchase cloud services internationally. A public university, civic-technology organisation or small research lab may not.
When compute is treated only as a commercial commodity, public-interest researchers compete for access in a market designed for actors with substantially greater resources.
That should concern philanthropy as well as development-finance institutions.
There is a legitimate role for commercial investment in data centres and cloud infrastructure. But commercial infrastructure and public-interest infrastructure are not necessarily the same thing.
We should also ask:
- Who can afford to use them?
- What forms of research do they enable?
- Where is the data stored?
- Which legal jurisdiction applies?
- Do African universities and public institutions have preferential access?
- Can civil-society organisations use this capacity to audit systems deployed by governments and companies?
Without these questions, the backbone and ecosystem investment may expand the African AI market while doing much less to expand African agency.
Data cannot remain an invisible subsidy
Artificial intelligence is built on data, but the political economy of data is often absent from funding conversations. African institutions, communities, media organisations, public agencies and citizens continuously produce information that has economic and social value.
Yet the systems for collecting, documenting, digitising, governing and maintaining this information remain severely underfunded.
This is particularly visible in African-language technology.
A 2025 review of large language models found meaningful support for only a fraction of the continent’s languages. It identified publicly available datasets for just 23 African languages and found that support remained concentrated around a small group, including Amharic, Swahili, Afrikaans and Malagasy. More than 98 per cent of African languages remained unsupported within the systems reviewed.
This absence is sometimes presented as a scarcity problem, as though African-language data simply does not exist. But communities speak, write, broadcast, translate, teach and document knowledge every day.
However, systems required to turn this knowledge into responsibly governed digital resources are poorly financed. They forget that;
- Someone must record the language.
- Someone must transcribe it.
- Someone must verify the quality.
- Someone must resolve questions of consent, ownership and community benefit.
- Someone must maintain the corpus after the grant ends.
- Someone must decide whether the data should be open, restricted, licensed or governed through a community trust.
This is the backbone work and it is also labour intensive.
When funders finance a language application but do not finance the people and institutions required to steward language resources over time, they are effectively treating African knowledge as a free input.
Public datasets create a similar challenge.
Governments and development organisations often call for more open data. But high-quality public data does not become available merely because an open-data portal has been created.
It requires statistical agencies, archival systems, interoperability standards, legal safeguards, regular updating, documentation and institutional accountability.
A dataset that was collected once for a donor-funded programme and never maintained is not public infrastructure.
It is a project artefact.
Regulation is also infrastructure
We rarely speak about regulators as part of AI infrastructure.
But perhaps we should.
A country can have data centres, AI start-ups and innovation programmes and still remain structurally dependent if its public institutions cannot understand or govern the technologies being introduced.
Regulators require technical expertise. Data-protection authorities need investigative capacity. Competition agencies must be able to assess digital markets. Procurement teams must know how to evaluate automated systems. Courts and lawmakers need access to independent evidence.
The African Union’s Continental Artificial Intelligence Strategy recognises the need for investment in infrastructure, datasets, skills, research, governance institutions and regional cooperation. It also places AI development within the wider continental ambition for data sovereignty and shared digital capacity.
But strategies do not finance themselves.
Across many countries, the institutions expected to implement these ambitions remain under-resourced. This creates a peculiar imbalance.
- A ministry may receive donor support to test an AI tool, while the regulator responsible for overseeing that tool has no dedicated technical team.
- A government may be encouraged to develop a national AI strategy, but receive little long-term financing to implement the institutions, standards and public accountability mechanisms that the strategy requires.
- Civil society may be invited to consultations, but lack the resources to conduct independent technical research or sustained monitoring.
We then call the resulting process “multi-stakeholder governance,” although the stakeholders enter the room with profoundly unequal capacity. Funding regulatory institutions may seem less innovative than funding an AI pilot. But without capable public institutions, the pilot may eventually become policy without having been adequately tested, challenged or governed.
The risk is a continent of permanent pilots
The danger is not that Africa will fail to adopt artificial intelligence. Adoption is already happening.
The greater danger is that the continent becomes a landscape of permanently dependent adoption.
Systems may be developed elsewhere and adapted locally. Public institutions may rely on proprietary platforms they cannot fully inspect. African data may improve external models without generating equivalent public value. Promising pilots may disappear when grant funding ends. Governments may inherit systems they cannot maintain. Researchers may remain dependent on temporary computing credits issued at the discretion of foreign firms.
This does not create technological sovereignty, but instead subsidised dependency.
And it cannot be solved by simply funding more African AI start-ups, however important entrepreneurship may be. Markets are unlikely to finance every form of infrastructure that the public interest requires.
- A company may invest in the languages with the clearest commercial market. It may not invest in languages spoken by smaller or poorer communities.
- A commercial cloud provider may build where demand and electricity supply are strongest. It may not build the shared research infrastructure needed by public universities.
- An investor may support a scalable health application. It may not finance a regulator, a data trust or an independent audit laboratory.
Yet, these are precisely the spaces in which public finance, philanthropy and development institutions should have a strategic role.
What should funders finance differently?
The first shift is from funding isolated solutions to funding ecosystems of capability.
Every major AI application grant should ask what infrastructure the intervention depends on and whether that infrastructure will remain after the project ends.
The second is to fund shared assets.
This could include regional computing facilities, national research clouds, interoperable public datasets, African-language repositories, public model-evaluation laboratories and secure data-sharing institutions.
The third is to treat governance capacity as a core investment rather than an administrative add-on.
Data-protection authorities, procurement institutions, standards agencies, parliamentary researchers and civil-society watchdogs all require sustained technical capacity.
The fourth is to invest regionally.
Not every African country will be able—or need—to build every part of the AI stack independently. Shared computing resources, language infrastructure, regulatory expertise and research networks could produce far greater public value than fragmented national initiatives.
The fifth is to finance maintenance.
Funders are often interested in creating things. Public infrastructure also requires people who will update, repair, secure and govern what has been created.
Maintenance is not a failure of innovation, but it is the thing that turns innovation into an institution.
Who pays is ultimately a question of who benefits
Public-interest AI infrastructure will require a combination of domestic public investment, regional financing, philanthropy, development finance and responsible private-sector participation.
But the contribution of each actor should be assessed against the value it expects to receive.
Technology companies that benefit from African markets, data and talent cannot be treated simply as benevolent donors to the ecosystem.
Development-finance institutions should evaluate digital infrastructure not only through commercial returns but through public accessibility, research capacity, local ownership and institutional resilience.
Philanthropy should be willing to finance the difficult middle layer between a compelling idea and a functioning public system.
And African governments must recognise that sovereignty cannot be declared in a strategy document. It must be financed through budgets, institutions and long-term capability.
The central question is therefore not whether funders are supporting artificial intelligence in Africa.
Many are.
The question is what kind of AI future their funding is making possible.
Are we financing a sequence of impressive demonstrations?
Or are we building the shared foundations through which African researchers, public institutions, communities and companies can shape, govern and benefit from the systems that will increasingly structure public life?
Because whoever pays for the infrastructure does more than support the technology.
They influence what gets built, who gets access, which forms of knowledge count and whose interests the system is ultimately designed to serve.
Perhaps the next question for governments, funders and technology companies is not how many AI projects we can launch, but what shared infrastructure we are collectively willing to build—and sustain—for the next generation.
