Authors: Tina Purnat. DrPH(c), Shannon Turner, PhD, Jude Kong, PhD, Ihoghosa Iyamu, MD, PhD
AI is starting to change how health care is delivered in low- and middle-income countries, reading X-rays, guiding community health workers, and helping plan supplies. These tools are built from health data, run on digital infrastructure, and depend on models few countries own. Each layer raises the same question about who controls it, and on whose terms. These operating conditions are impacted by financial dependencies and commercial incentives. An examination of Data Sovereignty in the African context can serve as a vehicle for elaborating this issue.
For four decades, the conditions on health aid to low and middle income countries were financial, set by financial lenders and paid in budget cuts. In 2026, new US bilateral health agreements came with a different condition, expecting access to national health data, from disease surveillance to genomic data and biological samples. This was happening while aid to Africa has dropped nearly 70 percent since 2021 while outbreaks rose more than 40 percent from 2022 to 2024.
The health data these agreements sought represents a valuable strategic resource that can be used to develop, adapt, validate, and deploy AI models for healthcare, while also enabling biomedical research and innovation. Several African governments, among them Ghana, Zimbabwe, and Zambia, refused or stalled deals that hinged on data access, and Kenya's High Court suspended one already signed, on data-protection grounds. Others accepted. Rwanda signed technology-heavy health agreements with the United States, the Gates Foundation and OpenAI, and Anthropic. Each low and middle-income country weighs these agreements through its own national perspective, which are sometimes difficult to reconcile with regional or bilateral relationships.
The interests of higher income countries around AI and data s in LMICs are not uniform either. China's Huawei built much of Africa's data networks, by one estimate 70 percent of its 4G, and now sells data centers branded as sovereignty tools. Media investigations alleged that data from the African Union headquarters were transferred to servers in Shanghai over several years, although both Huawei and African Union officials have disputed these claims. Europe funds digital health systems and health interoperability standards, most recently an EU-WHO push to spread a European-built health certification network across sub-Saharan Africa. Funding from the Gulf is behind new compute capabilities, including the Microsoft and G42 data center in Kenya. However, as of May 2026, the proposed US$1 billion Microsoft–G42 AI data center in Olkaria has encountered significant setbacks and has not progressed as originally announced. Multiple reports indicate that negotiations stalled due to a combination of (i) insufficient national electricity capacity, (ii) disagreements over government commitments to purchase or guarantee minimum cloud-computing capacity, and (iii) broader concerns regarding the commercial viability of the project. President William Ruto publicly acknowledged that a fully scaled facility would require approximately one-third of Kenya's installed electricity generation capacity, remarking that powering the facility could require "switching off half the country." While the Kenyan government subsequently clarified that discussions have not been formally terminated, the project has clearly been delayed and substantially restructured rather than proceeding as initially envisioned. These challenges amplify the importance of fair negotiations and environmental impact studies prior to implementation.
Three "sovereignty" in health data, digital and AI
Three ideas often get lumped together. Separating them helps.
- Data sovereignty is who decides how a population's health data is collected, stored, and shared. Kenya's court blocked a health financing and data sharing agreement on this ground.
- Digital sovereignty is control over the infrastructure the data runs on: connectivity, records systems, data centers. Africa has 18% of the world's people and under 1 percent of its data-center capacity. Africa CDC has asked heads of state to digitalize at least 90 percent of the continent's primary health care systems by 2035.
- AI sovereignty is the capacity to build, adapt, and govern the models. Ai in health has a potential to support transformation and reilience of health systems in LMICs. WHO now recommends AI software to read chest X-rays for tuberculosis, where it matches radiologists and finds cases that would be missed, and AI is already supporting community health workers, triage in clinics, and the planning of supplies and logistics. Yet Africa captures only about 2.5% of the global AI market and 0.3 percent of projected AI investment, and foreign firms can still own models, even if they are trained on a country's health data. Adding to the complexity, African languages barely appear in AI training data for natural-language processing, which is a focus of African research collectives like Masakhane.
Data, digital and AI are one health priority of many
Data, digital infrastructure, and AI are strategic for every country. Lower income countries have less room to negotiate terms around them in international agreements, and pay more when these terms are unfavorable to them.
Africa CDC has put sovereignty at the center of its health agenda and calls data sovereignty a pillar of pandemic preparedness, at the same time emphasizing that sovereignty does not mean isolation and depends on partners who back African priorities.
Doing this well is demanding and it carries costs beyond one-time implementation. An AI tool built elsewhere must be recalibrated to the local population, because its accuracy shifts with age, sex, and setting. That takes local data, local know-how to run and govern the tool, and enough local innovation that a health system can fix and adapt what it uses instead of only licensing it from elsewhere. Implementing AI in health also carries externalities that need to be addressed. For example, compute capacity needs power that African grids struggle to supply, pushing data centers onto costly diesel. If an AI tool carries bias from its training data and is unmonitored as it is implemented across settings, it can erode the trust between people and health systems that that is foundational for introducing new innovations in health. In addition, building a local data center may mean less budget for health services, infrastructure, and health worker salaries. Data-localization rules can protect a local population but this may also slow local research. Withholding pathogen data during an outbreak defends a country's rights but can also delay development of a vaccine that the country’s people need.
Africa has the world's youngest population, and AI could add up to $2.9 trillion to its economy by 2030. That value stays local only if young Africans build and own the AI models, and channel learnings from local implementations back into local research and innovation systems, instead of labeling data cheaply for foreign companies and supporting promotion and implementation of foreign technologies and projects.
Digital, Data Fair Trade - Aligning interests, not only redistributing resources
These arrangements last when they serve both sides, and often they can. African governments want to build and control their own systems, and Africa CDC and the AU have set out priorities to that end, national and continental control of health data, shared regional infrastructure, AI validated and governed locally. High-income countries and companies want reliable data, research access, pandemic security, and commercial return. Those interests are not opposed everywhere. A deal that leaves local engineers trained, models auditable, and benefits shared serves the funder's long-run interest in a stable partner as much as the recipients.
If we explore the principles of Fair Trade and apply them to the digital ecosystem we would expect the following approaches to be applied. Ensuring the Commercial Determinants of Health (CHOH) are health promoting not health harming. The corporate structures, market practices, and supply chains that directly shape health outcomes, often drive health inequities through the commercialization of data, resource extraction, and unequal labor practices. When AI and digital infrastructure are treated purely as corporate commodities, they reinforce systemic health disparities. Applying a Digital Fair Trade lens restructures these market dynamics to protect public health. Under such a model there would be fair compensation for data labour; frameworks would enforce living-wage standards, psychological support, and labor protections for digital pieceworkers/annotators in the Global South, transforming data annotation into a dignified path for local development. If data is extracted from a public system or specific population, a tangible equity return must be guaranteed (e.g., zero-cost licensing for public health agencies, mandatory reinvestment into local digital infrastructure).
Major commercial entities use proprietary algorithms and restrictive intellectual property (IP) laws to create; walled gardens, effectively enclosing data that should belong to the public domain. This restricts public health researchers from analyzing systemic health trends or validating AI safety. Establishing a Managed Commons for public health data, with open- source licensing, shared data repositories, and polycentric stewardship would help to ensure that critical public health insights remain accessible to researchers and local authorities, preventing the monopolization of foundational health intelligence. By mandating Intersectional Algorithmic Impact Assessments (AIAs) and community-led veto power it is possible to halt the deployment of systems that exhibit discriminatory drift. From a Planetary Health and Environmental stewardship model we need to embed strict environmental accountability and resource tracing into procurement. Governance must demand transparent energy provenance and water-use efficiency (WUE) reporting as a prerequisite for institutional or public deployment.
Similar to the unfair debt ratios impacting LMICs we need to ensure that we are not creating deeper poverty by harvesting, extracting, and destabilizing countries seeking to emerge from the information age as fuller partners with unfair digital governance models feeding commercial over public and population health interests. Global and digital equity is directly impacted by these structural determinants.