advertisement
The African Challenge Of Data Sovereignty And AI
Africa has made considerable progress in establishing data protection laws, with 44 countries now having legislation governing how personal information is collected, processed and stored. Yet as artificial intelligence adoption accelerates, a different challenge is emerging: much of the continent’s data is either hosted outside its borders or underrepresented in the datasets used to train AI models.
The scale of that gap became evident in January when a University of Pretoria researcher and co-founder of the HausaNLP community highlighted what he described as a record share of African languages in Common Crawl, an open web archive widely used in AI training.
African languages accounted for just 0.057 per cent of the pages captured by the archive. English alone represented approximately 42 per cent, giving it roughly 728 times more representation than the 29 African languages detected combined. Although Igbo pages more than doubled within a month and Swahili reached 294,000 pages, the overall representation remains limited.
advertisement
“There’s still a massive gap, but the direction is right,” the researcher observed.
The imbalance raises important questions about Africa’s digital sovereignty, particularly as organisations increasingly deploy AI tools to serve local markets. While African governments have focused on protecting personal information through legislation, ensuring that the continent’s data is stored, processed and adequately represented in AI systems remains a more complicated undertaking.
The issue featured among discussions at Africa Tech Festival in Cape Town, where data sovereignty and the need for localised computing infrastructure formed part of the broader conversation around Africa’s AI ambitions.
advertisement
On the regulatory front, progress has been significant. According to Strathmore University’s Centre for Intellectual Property and Information Technology Law, 44 African countries had enacted data protection laws by April 2026, while 37 had operational data protection authorities.
The African Union’s Malabo Convention on Cyber Security and Personal Data Protection also came into force in June 2023, nine years after its adoption. However, by early 2026, only 16 of the AU’s 55 member states had ratified it, highlighting the gap between continental policy ambitions and implementation.
Kenya provides an example of how governments are attempting to exercise greater control over sensitive information. Its Data Protection (General) Regulations of 2021 require personal data processed in connection with certain strategic state interests to be handled through servers and data centres in Kenya, or have at least one serving copy maintained locally.
advertisement
The requirements cover areas including revenue administration, civil registration, elections and specified health data. However, they do not extend explicitly to the broader collections of text, images and speech used to train AI models.
Infrastructure remains another constraint. Major cloud providers, including Amazon Web Services, Google Cloud, Microsoft, Oracle, Huawei and Alibaba, operate cloud regions in South Africa. Oracle has also established a public cloud region in Kenya.
Nigeria, despite being Africa’s most populous country, has fewer options for locally hosted hyperscale cloud infrastructure. A review by Lagos-based cloud provider Lineserve in October found that Microsoft Azure, Google Cloud and Oracle did not operate cloud regions in the country. AWS’s Lagos Local Zone, meanwhile, extends its Cape Town region rather than providing an independent full cloud region.
For businesses, this means that data generated in one African market may need to travel considerable distances for processing, potentially introducing additional costs and complicating questions around jurisdiction and data governance.
Guy Zibi, managing partner at Xalam Analytics, argues that governments must help stimulate demand for local infrastructure. “When governments localise workloads, infrastructure follows,” he said.
However, the question of where data is stored is only part of the problem. Its representation in the datasets used to develop AI models is equally important. A study published in the Proceedings of Machine Learning Research examined 20 language models and found that only 41 African languages were represented. Of these, just four, Amharic, Swahili, Afrikaans and Malagasy, were consistently supported. More than 98 per cent of the continent’s languages remained unsupported.
This creates a challenge for organisations deploying AI in markets where customers communicate in languages that global models have limited capacity to understand.
There are also growing questions about whether existing data protection frameworks can effectively protect information once it has been collected by international technology companies.
Kenya’s case against Worldcoin illustrates the issue. In 2023, thousands of Kenyans provided iris and facial scans to the project, co-founded by OpenAI chief executive Sam Altman, in exchange for digital tokens.
In May 2025, Kenya’s High Court ruled that the collection violated the Data Protection Act and ordered the deletion of biometric information belonging to more than 300,000 Kenyans within seven days, under the supervision of the Data Protection Commissioner. The commissioner’s office confirmed in January 2026 that the data had been erased.
Joshua Malidzo Nyawa, counsel for the Katiba Institute, which brought the case, described the ruling as “a win for the right to privacy in Kenya.”
The case demonstrated that African regulators and courts can hold foreign companies accountable for handling citizens’ data. However, the process, which began with Kenya’s suspension of Worldcoin in August 2023 and ended with confirmation of deletion more than two years later, also exposed the challenges of enforcing data protection across international operations.
For CIOs, these developments make data sovereignty an operational concern rather than simply a regulatory consideration. Organisations need to understand where their information is stored and processed, whether their cloud agreements provide sufficient transparency on data location, and whether the AI systems they adopt can adequately serve customers across Africa’s diverse linguistic environments.
Efforts to address the language gap are also emerging from research communities. Since late 2024, the Common Crawl Foundation has committed to expanding language coverage in its web crawls. In June 2025, it partnered with African research network Masakhane on a hackathon to identify and label web pages written in African languages, making them easier to collect and incorporate into future datasets.
These initiatives highlight an important dimension of Africa’s AI ambitions. Data sovereignty is not only about ensuring that information remains within national borders. It is also about ensuring that African data, languages and knowledge are represented in the technologies increasingly shaping the global economy.