- African Governments Emerge as Anchor Customers for Homegrown AI
African governments are beginning to emerge as customers for locally developed artificial intelligence, creating a route to scale for startups while opening difficult questions about privacy, procurement and the state’s reliance on automated decisions.
Public agencies are exploring AI applications across welfare, healthcare, agriculture, education, financial inclusion, climate action, trade and government administration. Their interest reflects a practical need to improve overstretched services, but also a strategic desire to keep sensitive public data and digital systems under greater domestic control.
Yet the state is not an ordinary customer. An AI system used to allocate welfare payments, diagnose illness or identify farmers for assistance can influence citizens’ rights and livelihoods. Public procurement must therefore evaluate not only price and technical performance, but fairness, explainability, cybersecurity, ownership of data and the ability to challenge automated decisions.
Togo provides the clearest example of both the opportunity and the complexity. During the Covid-19 pandemic, the government used machine learning, mobile-phone records and satellite data to identify vulnerable households that conventional welfare records had failed to capture.
The system supported the Novissi emergency cash-transfer programme, helping the government provide monthly payments of about US$20 to 154,238 people between December 2020 and April 2021. Researchers found that the approach improved targeting accuracy by 42 per cent compared with a simpler method based on identifying the poorest districts.
Togo has since expanded the initiative into a permanent AI and data-science programme involving more than 10 government agencies and over 70 projects in agriculture, health and public administration. Google.org announced US$2 million in research support in December 2025, including plans to train more than 100 students and 70 government data scientists.
The progression from emergency experiment to institutional programme is significant. Africa’s AI debate frequently focuses on innovation supply: how many startups exist, how much venture capital they attract and whether local engineers can adapt global models. Togo demonstrates that capable public institutions can create demand, turning technical potential into a functioning market.
Kenya is pursuing a similar approach through an AI Accelerator Programme involving its digital ministry, Konza Technopolis and the United Nations Development Programme. The initiative sought applications capable of improving public services and supporting the country’s green transition.
The Kenyan government is also collaborating with India and Italy on AI systems for African languages, including voice tools designed for areas with poor internet connectivity. This emphasis matters because imported products trained predominantly on foreign languages and contexts may perform badly when confronted with African accents, administrative records or informal economic activity.
Across Kenya, Malawi, Rwanda, South Africa, Togo and Zambia, a United Nations programme has identified 21 potential AI applications while governments invest in computing capacity and skills. Rwanda has identified a broader list of 42 uses across eight sectors.
A related initiative involving the UN, International Telecommunication Union and Research ICT Africa invited startups, universities, public agencies and individual developers to submit working prototypes and products. Selected innovators are expected to present their applications in Accra, where officials and regulators will assess practical benefits alongside cost, privacy and oversight risks.
Liz Orembo of Research ICT Africa argued that developers are too often excluded from policy discussions governing the systems they build.
“Most African AI policy conversations happen without the people actually building and deploying AI in the room,” she said.
The Accra engagement is intended to expose the practical obstacles separating successful pilots from scalable businesses. These include limited access to public data, inadequate computing capacity, unreliable electricity, expensive connectivity, fragmented rules and difficulty securing permission to operate across borders.
“These are the conditions that decide whether an African innovation stays a pilot or reaches scale,” Ms Orembo said.
Government demand alone will not solve those problems. Public procurement systems often favour large established vendors through demanding financial-history requirements, complex tender documents and lengthy payment cycles. A young AI company may possess a superior local product but lack the balance sheet or political connections needed to win a contract.
Officials must avoid replacing foreign technological dependence with domestic patronage. Contracts awarded in the name of supporting local innovation should still be competitive, transparent and subject to independent scrutiny. Local ownership cannot become an excuse for weak value-for-money assessment or undisclosed relationships between suppliers and decision makers.
Data access presents another dilemma. Developers need representative information to train and test systems, but governments hold records containing sensitive details about health, income, identity and location. Opening those datasets without robust safeguards could expose citizens to surveillance, commercial exploitation or security breaches.
The Togo welfare model illustrates the tension. Analysing mobile-phone behaviour allowed assistance to reach people quickly where income records were incomplete. But using personal communications data to estimate wealth raises questions about consent, accuracy and whether individuals wrongly excluded can obtain an explanation or appeal.
African governments must consequently treat AI procurement as a continuing governance responsibility rather than a one-time technology purchase. Contracts should define data ownership, audit rights, performance standards, bias testing, human oversight, cybersecurity obligations and exit arrangements if a provider fails.
Lesotho’s approach offers a useful lesson in sequencing. As the first country to apply the United Nations Blueprint for National AI and Data Strategies, it has begun a national computing plan running from 2026 to 2029 after assessing electricity, connectivity, data infrastructure and skills.
“We will build what fits our resources and our needs. Infrastructure must follow demand,” Prime Minister Samuel Matekane said.
The opportunity is substantial. Government can become the anchor customer that helps African AI companies prove their technology, employ skilled workers and expand across borders. But scale achieved through the state also magnifies every design error and governance failure.
Africa’s public-sector AI market will therefore be judged by more than the number of applications deployed. The decisive measure will be whether locally built systems improve services while keeping citizens’ data secure, public decisions contestable and procurement accountable. If governments can achieve that balance, they will not merely become customers for African AI; they will help create an industry grounded in local needs and public trust.
