This article is adapted from the author with permission from Press Technical Policy. Read the book the first article.
South Africa is not just another developing country struggling to manage artificial intelligence (AI); it is something different with average, and the operating window for it is closed. It owns about 88% of the world’s platinum metal reserves, a key input to semiconductor components and the data center supply chains that make AI infrastructure possible. It hosts the largest data center market in the continent. Its existing hyperscaler relationships provide purchasing power that many African countries may not have. And a major geopolitical contest over AI infrastructure is being fought on its own soil right now, between Chinese and American tech companies vying for control of the systems that will underpin the continent’s public sector.
In physics, lift requires three things: a fulcrum, a lever arm and the ability to apply force. The Bushveld Complex, the world’s largest platinum ore deposit, is the fulcrum: a mineral endowment that gives South Africa a position in the semiconductor supply chain that no other African country has. The policy that was drafted has since been withdrawn by the lever arm. Unsettled “OPTIONAL” provisions in the policy are where the power will be exercised. Without a policy that clarifies what South Africa wants in order to reach the market, the lever arm remains unused, and the weight of the two largest technological ecosystems in the world remains where those ecosystems want to be.
This makes South Africa a test case for the rest of the world. Not because its proposed governance methods are exemplary, but because it is the only developing country with enough structural power to negotiate truly different terms, and it is one that chooses, passively, not to do so. The recent announcement of a new panel to review the draft policy is an important opportunity. But the deeper failure is not that the AI policy contains negative indicators. That there is no verification process holding them before the document enters the public domain. That is a programmatic problem, not just a political one. It points to a missing layer in how governments are embracing AI.
The competition is already underway
Last year, Huawei introduced a number of emerging products to tech executives across the continent. Huawei has now bundled access to DeepSeek’s massive language model with its own cloud and storage infrastructure. The price difference was striking: in some cases more than 90%.
At the same time, Microsoft announced plans to spend ZAR 5.4 billion ($300 million) by the end of 2027 on cloud infrastructure and AI in South Africa, building on an earlier investment of ZAR 20.4 billion. Google, AWS and Oracle already have cloud regions in the country. According to some research, the country’s data center market was estimated at $2.16 billion by 2024, the largest in Africa.
This is not a commercially neutral investment. Huawei’s access to infrastructure is clearly linked to China’s strategic goals, including a documented record of providing governments with surveillance infrastructure through the Safe Cities network. The investment of the US hyperscaler comes with its own structure of dependence: closed models, fixed prices and conditions of access that no African government has ever rationally shaped. South Africa is being asked to choose between these types of dependencies without a policy specifying what it wants in return.
The power he has
There is a certain irony in the situation in South Africa. A country whose mines provide the platinum metal essential for semiconductor production, and by using AI compute, it has made a point of being a consumer of AI systems instead of participating in its governance. South Africa celebrates the minerals that make AI possible. No word on the AI built on them.
The three-dimensional framework of AI includes algorithms, computation, and data. South Africa does not have the capacity to develop a border model. South Africa holds significant data assets in financial services, healthcare and agriculture, without a clear framework for their independent management. South Africa holds a PGM of global importance on the compute axis, which is currently being transferred without a meaningful status. It also has very high solar radiance and great renewable energy. A country that can provide both valuable minerals and the infrastructure to provide those minerals helps build is in a bargaining position of unusual power.
The Policy Framework proposes no minimum requirements for hyperscaler investment, no data independence requirements, no technology transfer conditions and no virtual computing infrastructure. Many provisions are clearly left unsettled, marked “OPTION”, including very important options about how governance will work. Infrastructure decisions made now determine what can be renegotiated later, and the answer is: very little.
Three futures, one default
The three infrastructure futures offered each create a different type of dependency structure, and only one creates the potential for independence. The DeepSeek integration hosted by Huawei offers low weights and open sources, but with data stored in infrastructure that may be accessible under Chinese legal frameworks, creating a reliance on surveillance in a pattern already documented across Africa. The second is the dependence of the closed US model: high capacity, very reliable data protection, but complete API dependence on overseas developers. The third is an open-weight infrastructure hosted locally: models governed under South African data sovereignty laws, on infrastructure subject to minimum requirements, developed with South African data. As Nathan Lambert at Interconnects has observed, open-source models are probably the only realistic way to take AI off as a real endeavor, allowing local communities and economies to meaningfully interact with technology. But this requires buying conditions, not favors.
What does binding governance look like?
The GovAI framework “Governing Through the Cloud” identifies four roles that computing providers must adopt as operational conditions at scale: security (protecting model weights and training data), record keepers (keeping logs of infrastructure usage), assurances (ensuring customer compliance with security standards) and enforcers (limiting access in the event of a breach). These are operational requirements, not theoretical categories – specific, enforced, and within the bargaining power of a market of the size and nature of minerals in South Africa.
A detailed policy analysis submitted to the Department of Communications and Digital Technology (DCDT) identifies specific provisions that the final policy should contain: mandatory basic terms for foreign computer infrastructure investment over ZAR 500 million (~$30 million); limitation of electronic reporting; the mandate of the National AI Safety Institute which includes the protection monitoring of the collection of AI capability; and designation of a National AI Champion Sector to create data assets for domestic model development. Each offering turns a property’s profit into a management tool before that profit is actually lost on the market. Just as the security of modern software increasingly depends on knowing what components are inside the system—the model provider, the training data, the computing environment, the test methods, the update cadence, human review points, and failure reporting procedures—the governance of public sector AI requires a clear account of the stack before deployment, not after a problem occurs. A government agency that cannot certify sources in its AI policy is unlikely to be prepared to certify the AI systems it acquires, operates, or controls.
Why is this a test case for the continent
South Africa’s options will set a regional precedent for what can be negotiated commercially in AI infrastructure. When 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 investment and expansion of Huawei’s infrastructure continues on normal commercial terms, as it is now, it makes AI infrastructure rolled out across the country normal. The lesson is not specific to Africa. Governments everywhere are producing AI strategies while lacking the infrastructure to validate AI. South Africa is an early warning, not an isolated story.
The public comment period was closed when the policy was withdrawn. But the same process remains: The Draft Regulations for the National Procurement Agency—the legal instrument that will govern all AI contracts—closes for comment on June 15. Those regulations contain no AI-specific provisions.
South Africa has the highest level of AI of any country on the continent. Others argue, forcefully, that administrative requirements jeopardize investment in infrastructure that South Africa urgently needs: computing capacity, reliable power, start-up capital, and talent retention. That concern needs a specific answer. Minimum procurement criteria, computerized reporting restrictions, and technology transfer conditions are not barriers to investment. They are cases where investment benefits the host country rather than being taken from it. An infrastructure built without minimum requirements produces dependency. The infrastructure they are built on generates energy. To serve the public interest, its AI policy should implement it.
When late last month News24 reported on the AI guidelines set out in the AI policy framework, Communications and Digital Technology Minister Solly Malatsi withdrew the policy framework. That was a mistake that could cost South Africa and the entire continent a program that affects this urgent issue. His recent constitution for an independent body is a long overdue step in the right direction, if it is to turn South African power into policy. The team—led by Professor Benjamin Rosman of the Wits Machine Intelligence and Neural Discovery Institute, including Prof Vukosi Marivate and Alison Gillwald of Research ICT Africa, and Dr Jabu Mtsweni of the CSIR—has the technical and governance credentials to produce a robust document. Not yet produced a timeline. No revised draft is planned. South Africa still does not have a formal AI governance framework yet.
From Your Site Locations
Related Topics on the Web
