This text is tailored by the writer with permission from Tech Policy Press. Learn the original article.
South Africa isn’t just one other growing nation struggling to control synthetic intelligence (AI); it’s the exception with leverage, and the window to behave on it’s closing. It holds approximately 88% of global platinum-group metal reserves, crucial inputs to elements of the semiconductor and information middle provide chains that make AI infrastructure doable. It hosts the largest data center market on the continent. Its existing hyperscaler relationships give it procurement leverage that most African states will never have. And a serious geopolitical contest over AI infrastructure is being fought on its soil proper now, between Chinese language and American expertise firms competing for management of the programs that may underpin a complete continent’s public sector.
In physics, leverage requires three issues: a fulcrum, a lever arm and the power to use power. The Bushveld Advanced, the world’s largest platinum-group metal deposit, is the fulcrum: a mineral endowment that provides South Africa a place within the semiconductor provide chain that no different African state holds. The since-withdrawn draft policy is the lever arm. The unresolved “OPTION” provisions within the coverage are the place power can be utilized. With out a coverage that specifies what South Africa desires in return for market entry, the lever arm sits unused, and the load of two of the world’s largest expertise ecosystems settles precisely the place these ecosystems need it to settle.
This makes South Africa a world check case. Not as a result of its proposed technique of governance is exemplary, however as a result of it’s the one growing nation with sufficient structural leverage to barter genuinely different terms, and the one that’s selecting, by means of inaction, to not. The current announcement of a brand new panel to replace the draft coverage is a vital alternative. However the deeper failure isn’t that an AI coverage contained unhealthy references. It’s that no verification course of caught them earlier than the doc entered the general public area. That may be a programs downside, not merely a political one. It factors to a lacking layer in how governments are adopting AI.
The competition already underway
Final yr, Huawei, pitched an emerging product bundle to tech executives throughout the continent. Huawei was now bundling entry to the DeepSeek’s massive language mannequin with its personal cloud and storage infrastructure. The worth differential was stark: in some circumstances by greater than 90%.
On the similar time, Microsoft introduced plans to spend ZAR 5.4 billion ($300 million) by the top of 2027 on cloud and AI infrastructure in South Africa, constructing on a previous ZAR 20.4 billion funding. Google, AWS and Oracle have already got cloud areas within the nation. In keeping with one evaluation, the nation’s information middle market was valued at $2.16 billion in 2024, the largest in Africa.
These are usually not commercially impartial investments. Huawei’s infrastructure attain has been explicitly linked to Chinese strategic objectives, together with a documented track record of offering governments with surveillance infrastructure by means of its Protected Cities community. US hyperscaler funding comes with its personal dependency construction: closed fashions, pricing set unilaterally and phrases of entry that no African authorities has meaningfully formed. South Africa is being requested to decide on between these dependency fashions with out a coverage that specifies what it desires in return.
The leverage it has
There’s a specific irony in South Africa’s place. The nation whose mines provide platinum-group metals important to semiconductor manufacturing, and thru them to AI compute, has drafted a coverage that treats it as a shopper of AI programs moderately than a stakeholder of their governance. South Africa digs up the minerals that make AI doable. It has no say over the AI constructed from them.
The AI triad framework covers algorithms, compute, and information. South Africa has no frontier mannequin growth capability. South Africa holds vital information property in monetary providers, healthcare and agriculture, with no clear framework for his or her sovereign administration. South Africa possesses PGM leverage of world significance on the compute axis, at present being transferred with out significant situation. It additionally has exceptionally high solar irradiance and significant renewable energy potential. A rustic that may supply each crucial mineral inputs and the power to energy the infrastructure these minerals assist construct occupies a negotiating place of bizarre power.
The Draft Coverage proposes no minimal phrases for hyperscaler funding, no information sovereignty necessities, no expertise switch circumstances and no compute visibility mechanism. A number of provisions are explicitly left unresolved, marked “OPTION”, together with essentially the most consequential decisions about how governance will operate. Infrastructure selections made now decide what’s renegotiable later, and the reply is: little or no.
Three futures, one default
The three infrastructure futures on supply every create a structurally totally different type of dependency, and just one creates sovereign functionality. The Huawei-hosted DeepSeek integration gives low value and open-source weights, however with information saved on infrastructure doubtlessly accessible beneath Chinese language authorized frameworks, creating surveillance dependency in a sample already documented throughout Africa. The second is US closed-model dependency: larger functionality, extra dependable information safety, however full API dependency on builders overseas. The third is domestically hosted open-weight infrastructure: fashions ruled beneath South African data sovereignty rules, on infrastructure topic to minimal phrases, developed with South African information. As Nathan Lambert at Interconnects has noticed, open-weight fashions are possible the one sensible approach to get sovereign AI off the bottom as an actual effort, enabling native communities and economies to combine meaningfully with the expertise. However this requires procurement circumstances, not goodwill.
What binding governance seems like
The GovAI “Governing Through the Cloud” framework identifies 4 roles compute suppliers ought to settle for as circumstances of working at scale: securers (defending mannequin weights and coaching information), report keepers (sustaining infrastructure utilization logs), verifiers (confirming buyer compliance with security requirements) and enforcers (proscribing entry when violations happen). These are operational necessities, not theoretical classes — particular, enforceable, and nicely inside the bargaining energy of a market of South Africa’s measurement and mineral place.
A detailed policy analysis submitted to the Division of Communications and Digital Applied sciences (DCDT) identifies the precise provisions the ultimate coverage should comprise: necessary minimal phrases for international compute infrastructure investments above ZAR 500 million (~$30 million); a compute reporting threshold; a Nationwide AI Safety Institute mandate protecting defensive monitoring of AI functionality accumulation; and Nationwide AI Champion Sector designations to create information property for home mannequin growth. Every provision converts a structural benefit right into a governance instrument earlier than that benefit is foreclosed by market actuality. Simply as fashionable software program safety more and more is determined by realizing what parts are inside a system—mannequin supplier, coaching information, compute setting, analysis strategies, replace cadence, human overview factors, and failure-reporting procedures—public-sector AI governance requires a transparent account of the stack earlier than deployment, not after an issue surfaces. A public establishment that can’t confirm the sources in its personal AI coverage is unlikely to be able to confirm the AI programs it procures, deploys, or regulates.
Why that is the continental check case
South Africa’s decisions will set up a regional precedent for what’s commercially negotiable in AI infrastructure. If South Africa negotiates information sovereignty ensures and expertise switch circumstances as necessities for hyperscaler funding, it creates a replicable mannequin. If Microsoft’s $300 million funding and Huawei’s infrastructure enlargement proceed on customary business phrases, as they’re at present, it normalizes extractive AI infrastructure throughout the continent. The lesson isn’t particular to Africa. Governments all over the place are producing AI methods whereas missing AI assurance infrastructure. South Africa is an early warning, not an remoted case.
The general public remark interval closed when the coverage was withdrawn. However a parallel course of stays dwell: the National Treasury’s Draft General Public Procurement Regulations—the authorized instrument that may govern each authorities AI contract—closes for touch upon June 15. These laws comprise no AI-specific provisions.
South Africa has extra AI leverage than any nation on the continent. Some argue, with power, that governance requirements risk deterring the infrastructure investment South Africa urgently wants: compute capability, dependable power, enterprise capital, and expertise retention. That concern deserves a direct reply. Minimal procurement phrases, compute reporting thresholds, and expertise switch circumstances are usually not boundaries to funding. They’re the circumstances beneath which funding serves the host nation moderately than extracting from it. Infrastructure constructed with out minimal phrases produces dependency. Infrastructure constructed with them produces leverage. To serve the general public curiosity, its AI coverage should use it.
When late final month News24 reported AI-hallucinated references within the draft AI coverage, Minister of Communications and Digital Applied sciences Solly Malatsi withdrew the draft policy. That was a mistake that would value South Africa and the remainder of the continent the initiative on this pressing challenge. His newer structure of an independent panel is a belated step in the precise route, if it might probably flip South Africa’s leverage into coverage. The panel—chaired by Prof Benjamin Rosman of the Wits Machine Intelligence and Neural Discovery Institute, and together with Profs Vukosi Marivate and Alison Gillwald of Analysis ICT Africa, and Dr Jabu Mtsweni of the CSIR—has the technical and governance credibility to provide a stronger doc. What it has not but produced is a timeline. No revised draft has been scheduled. South Africa stays with out a formal AI governance framework within the interim.
From Your Web site Articles
Associated Articles Across the Internet
