Africa should continue developing its own models and research capabilities. But its most immediate commercial opportunity lies in deploying artificial intelligence to solve local problems – not competing immediately in the most expensive segment of the global AI race.
The global artificial intelligence race is often presented as a contest to build the largest data centres, acquire the most graphics processing units and train the biggest models.
For Africa, that framing risks directing limited capital towards the most expensive part of the AI value chain before the continent has established sufficient demand, power and infrastructure to support it.
Africa should not abandon model development, specialised training or long-term ambitions to participate in frontier research. African languages, industries and public institutions require models developed or adapted with local knowledge and relevant data.
However, the continent’s most immediate opportunity may not be training frontier models from the beginning. It lies in AI inference: applying trained or adapted models to real problems in banking, payments, healthcare, agriculture, education, customer service and government.
That distinction shaped the discussion during the “From AI Consumer to AI Producer: Building Africa’s Intelligence Infrastructure” panel at Hyperscalers Convergence Africa 2026.
Moderated by Prof. Nkem Ihenachor, Professor of Strategic Management, Lagos Business School, the session featured Gary Chomse, Sales Director, Vertiv; Wole Abu, Managing Director, West Africa, Equinix; Kanwulia Okafor, Director, Industry Services, GSMA; Lars Johannisson, Chief Executive Officer, Rack Centre; and Guy Zibi, Chief Executive Officer, Xalam Analytics.
Training and inference require different strategies
Training is the process through which an AI model learns patterns from large datasets. Training a frontier model can require thousands of specialised processors operating for extended periods, supported by enormous amounts of electricity, cooling, networking and capital.
Inference happens when a trained model receives information and produces an answer, prediction or action. It is the stage at which AI is used to detect a fraudulent payment, translate a local language, analyse a medical image, answer a customer’s question or advise a farmer.
Both activities require computing capacity, but their commercial and infrastructure requirements can differ.
Large training workloads can often be scheduled in locations offering abundant electricity and specialised computing infrastructure. Many inference workloads need to operate closer to users, businesses and their data because responsiveness, connectivity costs, privacy and regulatory requirements matter.
A fraud-detection system assessing a live payment, for example, may need to respond within milliseconds. A customer-service platform must remain accessible even when international connectivity is disrupted. A system processing sensitive financial or government information may also face requirements concerning where its data is stored and managed.
This makes inference a significant local and regional infrastructure opportunity.
“Africa must choose the market where it can compete,” Wole Abu told the panel. “Inferencing is latency-dependent, it recognises data-residency laws, and it needs to sit closer to the applications and customers it serves.”
This does not mean that every inference workload must remain within national borders. Cloud platforms will continue to distribute workloads according to performance, security, cost and regulatory requirements. It means that a growing category of African applications will benefit from local or regional processing.
The opportunity begins with African use cases
The economic value of AI will not be determined by how many processors Africa acquires, but by what people and businesses do with them.
Across the continent, financial institutions can use AI for fraud detection, credit assessment, compliance and customer service. Healthcare providers can support diagnostic screening and resource allocation. Agricultural applications can provide information on crops, weather, disease and markets. Governments can use AI to improve service delivery, revenue administration and document processing.
The African Union’s Continental Artificial Intelligence Strategy identifies agriculture, education, health, public services, climate, trade, infrastructure and security among the priority areas in which AI can support development. It calls for sustained investment in electricity, broadband, data centres, cloud services, computing capacity, quality data, skills and research.
Research by GSMA has also identified more than 90 AI-enabled use cases across Kenya, Nigeria and South Africa, including applications in agriculture, food security, energy and climate action. The research emphasises that locally relevant solutions must be designed with an understanding of African communities, languages and operating environments.
The foundation already exists. The hurdle is moving from experimentation and isolated applications to repeatable services capable of generating revenue or measurable public value.
“We have to distinguish between interest and what is bankable enough to justify further investment in AI,” said Kanwulia Okafor.
That is the point at which AI consumption becomes effective demand.
Africa’s problem is not only capital
Announcements of new facilities, GPU clusters and proposed investments are increasing across Africa. Yet infrastructure investment cannot be sustained by announcements alone.
Data centre operators and financiers need customers capable of making credible commitments. They need to understand who will use the computing capacity, how much they will consume, what they can afford and for how long.
At present, that demand is fragmented.
A fintech may need computing capacity for fraud detection. A hospital may need it for medical analysis. A university may require resources for research, while a startup needs temporary access to test a specialised model. Individually, those requirements may be too small or unpredictable to justify a dedicated infrastructure project.
Aggregated, they can become a market.
“One fintech workload is too small; 50 fintechs is one project,” Abu said.
Industry associations, governments, telecommunications companies, banks and infrastructure providers could play an important role in combining these requirements. Shared computing platforms could give smaller organisations access to processing power without requiring each of them to purchase and manage expensive hardware.
Government can also provide anchor demand by aggregating requirements across public institutions. The private sector can contribute longer-term contracts, while development finance institutions can help reduce the risks associated with early infrastructure investment.
The objective should be to convert expressions of interest into measurable demand and, ultimately, signed commercial commitments.
Pricing currently excludes African innovators
Much of the continent’s existing data centre, fibre and international connectivity infrastructure was developed for large enterprises, content delivery networks (CDNs) and hyperscale customers.
The associated contracts and pricing structures often assume commitments beyond the reach of local startups, research institutions and smaller businesses.
Africa may therefore possess available infrastructure while its innovators remain unable to use it.
“We have built the brick and mortar, but we have virtually locked African enterprises out because the pricing and contracting models require a scale most of them do not yet have,” Abu said.
A new access model is required.
Providers could offer fractional GPU access, pay-as-you-use computing, shared clusters and sector-specific capacity pools. Governments and development partners could support research access without distorting the wider commercial market. Larger enterprises could underwrite baseline demand, allowing smaller users to access spare capacity through more flexible arrangements.
The objective should not be permanent subsidies. It should be the creation of a market in which infrastructure providers can recover their investments while African businesses obtain computing capacity on terms appropriate to their scale.
There is no AI strategy without an energy strategy
Whether Africa focuses on training, inference or both, reliable power remains essential.
The International Energy Agency estimates that data centres consumed approximately 415 terawatt-hours of electricity in 2024, representing about 1.5 per cent of global electricity consumption. It projects that consumption could rise to approximately 945 terawatt-hours by 2030, driven significantly by AI and other digital services.
AI infrastructure also places unusually concentrated demands on local electricity systems. Specialised processors generate substantial heat, require advanced cooling and must be protected from interruptions and fluctuations.
“No power and no cooling means no revenue,” said Gary Chomse. “The AI engine is an extremely expensive infrastructure asset, and that equipment must be protected.”
African markets that rely heavily on self-generated power can build resilient facilities, but the additional equipment and fuel increase the cost ultimately paid by customers.
The continent’s AI strategy must therefore connect data centre planning with power sector development. Facilities may need to be located close to reliable generation, supported by dedicated energy arrangements or developed alongside renewable energy, gas, storage and improved grid infrastructure.
Cheap and reliable power will influence which African markets can support large-scale training. It will also determine whether local inference can be offered at prices African businesses can afford.
Connectivity is the distribution layer
Inference capacity concentrated in Lagos, Johannesburg, Nairobi or Cairo will have limited developmental effect if businesses elsewhere cannot access it reliably.
Affordable, high-quality broadband is the distribution layer connecting computing infrastructure to enterprises, institutions and users.
“If I had to choose one intervention, it would be affordable, high-quality broadband deployed beyond the major economic centres,” Okafor said. “Connectivity allows fragmented innovation ecosystems to become repeatable and scalable.”
This requires more than mobile coverage. Businesses need consistent service quality, sufficient capacity and affordable access.
Governments can reduce deployment costs by including fibre ducts in roads and connectivity requirements in major buildings. Open-access metropolitan networks and stronger regional interconnection can also make computing resources accessible beyond the cities in which they are physically located.
Edge computing may become increasingly important. Some inference can occur in smaller regional facilities, telecommunications networks or capable devices, reducing dependence on distant cloud regions. GSMA notes that bandwidth costs, connectivity constraints and limited local computing capacity can make edge AI a practical alternative for some applications in developing markets.
Production should be measured by economic impact
Africa will not become an AI producer simply by hosting computing equipment. Production should mean using the technology to improve the productive capabilities of businesses, governments and citizens.
A locally situated GPU cluster with limited utilisation is not evidence of success. Neither is a data centre built without access or a customer pipeline capable of sustaining it.
Guy Zibi proposed measuring AI’s productive contribution to gross domestic product. Other useful indicators could include local computing utilisation, the number of commercially deployed applications, revenue generated by African AI companies, productivity improvements and the share of smaller enterprises able to access computing capacity.
“Productivity is measured in megawatts used, not announcements made,” Abu concluded.
That should become a guiding principle for Africa’s AI infrastructure strategy.
Build from use cases towards capability
Africa should maintain ambitions across the AI value chain. Its universities and companies should continue developing specialised models, adapting existing ones and conducting original research. Countries with suitable power and infrastructure should pursue opportunities in model training.
But the continent does not have to begin by replicating the most capital-intensive strategies being pursued elsewhere.
It can begin with areas in which proximity, local data, languages, regulation and market knowledge create a genuine advantage. It can aggregate demand around those applications, expand access to computing capacity and build the supporting power, connectivity and skills.
As those markets mature, they can justify larger infrastructure investments and more sophisticated model development.
In the end, Africa’s progress will not be measured by the number or size of AI projects announced or computing facilities promised. It will be measured by whether African businesses and institutions can afford to use the technology to solve real problems, and whether those solutions generate enough demand and value to keep the supporting infrastructure commercially viable.
