South Africa: Untapped Potential in AI Governance

A Unique Mineral Potential

South Africa is not just a developing country struggling to regulate artificial intelligence (AI); it represents an exception with leverage, and the window to act is closing. It holds about 88% of the world's platinum group metal reserves, essential elements for the supply chains of semiconductors and data centers that make AI infrastructure possible. The country is home to the largest data center market on the continent. Its existing relationships with hyperscalers give it a purchasing power that most African states will never have. Moreover, a significant geopolitical competition for AI infrastructure is currently unfolding on its soil, between Chinese and American tech companies vying for control of the systems that will support the entire public sector of the continent.

In physics, leverage requires three elements: a fulcrum, a lever arm, and the ability to apply force. The Bushveld Complex, the largest platinum group metal deposit in the world, is the fulcrum: a mineral endowment that gives South Africa a position in the semiconductor supply chain that no other African state possesses. The now-withdrawn policy draft is the lever arm. The unresolved "OPTION" provisions in the policy are where the force would be applied. Without a policy that specifies what South Africa wants in return for market access, the lever arm remains unused, and the weight of two of the world's largest tech ecosystems stabilizes exactly where those ecosystems want it to stabilize.

This makes South Africa a global case study. Not because its proposed governance means are exemplary, but because it is the only developing country with sufficient structural leverage to negotiate truly different terms, and it chooses, through inaction, not to do so. The recent announcement of a new panel to update the policy draft represents a significant opportunity. But the deeper failure is not that the AI policy contained poor references. It is that no verification process detected them before the document entered the public domain. This highlights a systemic issue, not just a political one. It brings to light a missing layer in how governments adopt AI.

The Ongoing Competition

Last year, Huawei presented a suite of emerging products to the continent's tech leaders. Huawei was now offering access to the DeepSeek language model with its own cloud and storage infrastructure. The price gap was striking: in some cases, over 90%.

At the same time, Microsoft announced plans to invest 5.4 billion ZAR (300 million dollars) by the end of 2027 in cloud and AI infrastructure in South Africa, building on a previous investment of 20.4 billion ZAR. Google, AWS, and Oracle already have cloud regions in the country. According to an analysis, the country's data center market was valued at 2.16 billion dollars in 2024, the largest in Africa.

These investments are not commercially neutral. Huawei's infrastructure has been explicitly linked to Chinese strategic objectives, including a documented history of providing surveillance infrastructure to governments through its Safe Cities network. The investment from American hyperscalers comes with its own dependency structure: closed models, unilaterally set prices, and access conditions that no African government has significantly shaped. South Africa is invited to choose between these dependency models without a policy that specifies what it wants in return.

The Leverage It Has

There is a particular irony in South Africa's position. The country whose mines provide essential platinum group metals for semiconductor manufacturing, and through them for AI computing, has drafted a policy that treats it as a consumer of AI systems rather than as a player in their governance. South Africa extracts the minerals that make AI possible. It has no say over the AI built from them.

The triad framework of AI covers algorithms, computing, and data. South Africa lacks the capacity to develop cutting-edge models. It holds significant data assets in financial services, health, and agriculture, without a clear framework for their sovereign management. South Africa possesses globally significant PGM leverage on the computing axis, currently transferred without significant conditions. It also has exceptionally high solar irradiance and significant renewable energy potential. A country capable of offering both critical mineral inputs and the energy to power the infrastructure that these minerals help build occupies a position of unusual negotiating strength.

The policy draft proposes no minimum conditions for investments from hyperscalers, no data sovereignty requirements, no technology transfer conditions, and no visibility mechanisms for computing. Many provisions are explicitly left unresolved, marked "OPTION," including the most consequential choices regarding governance operation. The infrastructure decisions made now will determine what is renegotiable later, and the answer is: very little.

Three Futures, One by Default

The three proposed infrastructure futures each create a structurally different form of dependency, and only one creates sovereign capacity. The integration of DeepSeek hosted by Huawei offers low costs and open-source weights, but with data stored on infrastructure potentially accessible under Chinese legal frameworks, creating a dependency on surveillance in a pattern already documented across Africa. The second is dependency on a closed American model: more capacity, more reliable data protection, but total dependence on foreign developers' APIs. The third is locally hosted open-weight infrastructure: models governed by South African data sovereignty rules, on infrastructure subject to minimum conditions, developed with South African data. As Nathan Lambert at Interconnects observed, open-weight models are likely the only realistic way to launch a sovereign AI as a real effort, allowing local communities and economies to meaningfully integrate with technology. But this requires procurement conditions, not goodwill.

What Binding Governance Looks Like

The GovAI "Governing Through the Cloud" framework identifies four roles that computing providers should accept as conditions for large-scale operation: guardians (protecting model weights and training data), record keepers (maintaining logs of infrastructure usage), verifiers (confirming client compliance with security standards), and enforcers (restricting access in case of violations). These are operational requirements, not mere theoretical categories — specific, applicable, and well within the negotiating power of a market the size and mineral position of South Africa.

A detailed policy analysis submitted to the Department of Communications and Digital Technologies (DCDT) identifies specific provisions that the final policy must contain: mandatory minimum conditions for foreign investments in computing infrastructure exceeding 500 million ZAR (~30 million dollars); a computing reporting threshold; a mandate for a National AI Security Institute covering defensive oversight of AI capability accumulation; and designations of National AI Champion to create data assets for national model development. Each provision converts a structural advantage into a governance instrument before that advantage is closed off by market reality. Just as the security of modern software depends on knowledge of the components inside a system — model provider, training data, computing environment, evaluation methods, update cadence, human revision points, and failure reporting procedures — AI governance in the public sector requires a clear account of the stack before deployment, not after a problem has manifested. A public institution that cannot verify the sources of its own AI policy is unlikely to be ready to verify the AI systems it acquires, deploys, or regulates.

Why This Is the Continental Case Study

South Africa's choices will set a regional precedent for what is commercially negotiable in AI infrastructure. If South Africa negotiates data sovereignty guarantees and technology transfer conditions as requirements for hyperscaler investment, it creates a replicable model. If Microsoft’s 300 million dollars investment and Huawei's infrastructure expansion continue under standard commercial terms, as is currently the case, it normalizes extractive AI infrastructure across the continent. The lesson is not specific to Africa. Governments worldwide are producing AI strategies while lacking AI assurance infrastructure. South Africa is an early warning, not an isolated case.

The public comment period ended when the policy was withdrawn. But a parallel process remains active: the draft general regulations for public procurement from the National Treasury — the legal instrument that will govern every government contract related to AI — will close for comments on June 15. These regulations contain no specific provisions for AI.

South Africa has more leverage in AI than any other country on the continent. Some argue strongly that governance requirements risk discouraging the infrastructure investment that South Africa desperately needs: computing capacity, reliable energy, venture capital, and talent retention. This concern deserves a direct response. Minimum procurement conditions, computing reporting thresholds, and technology transfer conditions are not barriers to investment. They are the conditions under which investment serves the host country rather than extracting from it. Infrastructure built without minimum conditions produces dependency. Infrastructure built with them produces leverage. To serve the public interest, its AI policy must utilize it.