advertisement
Africa’s AI Runs On Rented Infrastructure
As Africa positions itself to become a producer rather than a consumer of artificial intelligence, questions around ownership of the infrastructure, technology and data powering that ambition are becoming harder to ignore. Much of the continent’s AI ecosystem still depends on foreign-owned data centres, imported computing power and externally developed models, raising questions about what AI sovereignty will mean in practice.
In Nairobi, Joan Kinyua represents a group of workers whose contribution to the global artificial intelligence industry rarely features in conversations about technological advancement. As president of the Data Labelers Association, she represents approximately 1,100 Kenyan workers involved in tagging images, transcribing audio and moderating content used to train AI systems.
These workers help build the technology increasingly shaping global economies, yet their share of the value generated remains limited. “The workers power all these technological advancements, but they’re paid peanuts,” Kinyua told Computer Weekly when the association was launched.
advertisement
Her observation highlights a contradiction at the centre of Africa’s AI ambitions. While governments and technology leaders are increasingly calling for greater digital and technological sovereignty, the continent continues to depend heavily on external players for the infrastructure, computing resources and platforms needed to develop and deploy AI.
The issue featured prominently as Africa Tech Festival opened in Cape Town, where sovereign AI emerged as one of the central themes of discussions. The question facing African economies is no longer simply how to adopt artificial intelligence, but how to develop the capacity to produce, own and control the technology.
Addressing delegates remotely from Pretoria, South Africa’s Deputy President Paul Mashatile emphasised the importance of regional cooperation, saying that “no African country will realise its digital potential in isolation.”
advertisement
However, collaboration alone will not resolve the ownership question. When an African government, bank or start-up deploys an AI model, who owns the chips processing the information, the data centre housing the infrastructure, the model itself and the data used to train it? Examining each layer reveals how much of the continent’s AI value chain remains dependent on foreign ownership.
The most immediate constraint is computing capacity. Africa accounts for approximately 18 per cent of the world’s population but holds less than one per cent of global data centre capacity, according to industry estimates cited by the World Economic Forum. Guy Zibi, managing partner at Xalam Analytics, estimates that the continent hosts only about 0.2 per cent of global AI computing capacity.
This shortage has direct financial implications for organisations looking to develop or run AI applications locally. According to Sarah Rees, chief executive of GPU price tracker Signwl, renting an Nvidia H100 GPU in Africa cost approximately $13.55 per hour in May, compared with a global average of $7.32. The difference, which puts African pricing about 85 per cent higher, reflects limited supply and the high cost of energy.
advertisement
The ownership of the facilities housing this computing infrastructure presents another challenge. Some of the continent’s largest data centre operators are now controlled by international companies, meaning that infrastructure located in Africa does not necessarily translate into African ownership.
South Africa’s Teraco, the continent’s largest data centre platform, has been majority-owned by US-based Digital Realty since 2022, following a transaction that valued the business at approximately $3.5 billion. Digital Realty had also acquired Kenya’s iColo and Nigeria’s Medallion, while US-based Equinix expanded its African footprint through the $320 million acquisition of MainOne’s West African data centre operations.
Consequently, a significant portion of the infrastructure supporting Africa’s digital economy is ultimately owned by companies headquartered outside the continent. This distinction matters as governments increasingly seek greater control over where their data is stored, processed and governed.
There are, however, efforts to build African-owned capacity. Cassava Technologies, founded by Zimbabwean entrepreneur Strive Masiyiwa, launched its Nvidia-powered AI factory in South Africa in March, with plans to expand into Nigeria, Kenya, Egypt and Morocco.
The company has indicated that it intends to deploy between 12,000 and 13,000 GPUs across its markets, alongside plans for a 20-megawatt facility in Johannesburg. The investment represents an important step towards developing locally accessible AI infrastructure under African ownership. Yet even here, the underlying hardware remains dependent on Nvidia, the US-based manufacturer that dominates the global market for high-end AI processors.
The challenge of securing infrastructure at scale is also evident in Kenya, where plans for a major geothermal-powered data centre have exposed the complexities surrounding foreign investment and national energy priorities.
In 2024, Abu Dhabi-based G42, in partnership with Microsoft, announced plans for a $1 billion data centre at Olkaria, intended to support a new Azure cloud region for East Africa. The project was expected to leverage Kenya’s geothermal resources to provide power for energy-intensive computing infrastructure.
Speaking at GITEX Kenya in May, Sandy Okoth, director of the Investment Deal Room at Invest Kenya, said the proposal had expanded from an initial 100-megawatt pilot to a potential one-gigawatt development. Kenya generates approximately 3,000 megawatts of electricity and has peak demand of about 2,400 megawatts.
According to Okoth, the proposed development would have consumed more than half of the country’s available power generation capacity. He also raised concerns over government guarantees sought by G42, arguing that the proposed arrangements could have transferred significant private investment risks to Kenyan taxpayers. Kenyan officials subsequently told Al Jazeera in June that discussions over the project were continuing.
Beyond physical infrastructure, Africa’s position in the AI value chain is equally complicated at the model development layer. Local developers are increasingly building AI systems that account for African languages and contexts, but many still rely on computing infrastructure provided by international technology companies.
Nigeria, for instance, unveiled N-ATLAS, an open-source language model supporting Yoruba, Hausa, Igbo and Nigerian-accented English, at the United Nations General Assembly in September 2025. Developed with Lagos-based start-up Awarri, the project represents an effort to address the underrepresentation of African languages in mainstream AI systems. However, TechCabal reported that Awarri relies on Amazon Web Services and Google Cloud for its computing requirements.
South Africa’s Lelapa AI has taken a similar approach to developing models suited to African languages and resource constraints. Its InkubaLM model, described as the continent’s first multilingual small language model, was developed using computing credits from Microsoft’s AI for Good Lab.
Rather than attempting to compete directly with the resource-intensive models developed by global technology companies, Lelapa AI has focused on building smaller, more efficient systems. The company reduced the model’s size by up to 75 per cent without sacrificing performance, demonstrating how local developers can adapt AI development to the infrastructure limitations they face.
Chief executive Pelonomi Moiloa has described this approach as a necessity, noting that the company builds lean from the outset “because we must make things work under constraint.”
The dependence on external infrastructure becomes even more pronounced when examining the data underpinning AI systems and the people responsible for preparing it. This is one area where Africa already plays a significant role in the global AI supply chain, although the economic benefits remain unevenly distributed.
Kenyan content moderators working on projects for companies such as Meta and OpenAI through outsourcing firms have reportedly earned between $1.46 and $3.74 per hour, according to Business Daily. In April, outsourcing company Sama announced more than 1,000 job cuts in Nairobi following the end of its contract with Meta.
The developments have brought renewed attention to the working conditions and compensation of African data workers, whose contributions are essential to training and evaluating AI systems used globally.
Kenya has since published a draft policy proposing that wages for data annotation, content moderation and AI evaluation work be benchmarked against international rates. It also proposes requiring employers to disclose their remuneration structures.
For African technology leaders, particularly CIOs, these developments raise questions that extend beyond the economics of AI development. Decisions about where AI workloads are hosted, which cloud providers organisations depend on and who controls the underlying infrastructure have implications for data governance, regulatory compliance, operational costs and business continuity.
An organisation running its AI workloads on infrastructure located in Africa may still be dependent on a foreign company for ownership, hardware, software and operational control. Similarly, developing an African language model does not necessarily mean that the technology is independent if its training and deployment depend entirely on external computing platforms.
The distinction between infrastructure being hosted in Africa and being owned and controlled in Africa is therefore central to the continent’s pursuit of AI sovereignty. Local data centres, African-developed models and access to computing resources are important, but they do not individually guarantee independence across the technology value chain.
Joseph Asunka, chief executive of Afrobarometer, has cautioned against AI infrastructure agreements being “conducted at the elite level and dumped on citizens.” His warning points to another dimension of sovereignty: ensuring that the economic and social benefits of AI investments extend beyond governments and large technology companies.
Ultimately, Africa’s AI ambitions will depend on more than its ability to attract foreign investment or adopt emerging technologies. The continent must also build the capacity to own and operate the infrastructure, develop competitive models, retain value from its data and ensure that the people contributing to the AI economy benefit from its growth.
As discussions at Africa Tech Festival continue to focus on the future of sovereign AI, the question remains whether Africa can move beyond supplying the labour, languages and data that power the global AI industry to owning a greater share of the technology and infrastructure behind it.