South Africa Will Not Pick a Side Between US and Chinese AI Stacks – Malatsi

South Africa will not align itself with either Washington or Beijing on artificial intelligence, communications minister Solly Malatsi has told TechCentral, rejecting the central premise of a new Boston Consulting Group study arguing that the window for countries to hedge between the two AI superpowers is closing.

The Great Divide: How the US and China Are Splitting the AI World, published on 30 June, found that the two countries are no longer building rival products inside one market but separate and increasingly incompatible technology stacks, running from chips through cloud and models to applications. Where BCG’s 2024 assessment found countries could still mix and match across both, it now warns that the two ecosystems have diverged to the point where neutrality may soon become harder to sustain.

South Africa is not named in the report. But nearly every condition BCG identifies as pushing a middle-income country towards one stack or the other is present here — and the country will not have a published national AI policy before next year.

‘Not choosing sides’

Asked by TechCentral whether South Africa’s membership of the China-led World Artificial Intelligence Cooperation Organisation carries any commitment on technology standards or procurement, Malatsi was unequivocal.

“South Africa’s membership, like its participation in any multilateral AI or technology body, carries no binding commitment on technology standards or procurement,” he said in a written response. “South Africa engages with all such organisations, whether led by the US, China, the EU, or any other bloc, on the same basis: to share knowledge and advance global cooperation, not to align our state technology choices with any single geopolitical camp.”

State buying, he said, remains governed by existing procurement law and practice, including value-for-money, interoperability, security and fit-for-purpose assessments — with an approach that has “consistently emphasised open standards, interoperability and avoiding vendor lock-in, principles that apply regardless of a vendor’s country of origin.”

“To be clear: no membership of Waico, or of any comparable body aligned with any other country, will ever determine South Africa’s procurement decisions based on the nationality of an OEM or service provider,” Malatsi said. “South Africa is not choosing sides in the divide the referenced report describes. We intend to remain able to work with the best available technology, wherever it is developed.”

He did not answer a second question on whether the redrafted national AI policy will include guidance on the acquisition of AI hardware and software stacks.

South Africa has signed the agreement establishing Waico, acting deputy director-general Jeanette Morwane told TechCentral last week. The body, proposed by Chinese President Xi Jinping and established in Shanghai in July, will focus on creating native AI innovations and tools rather than simply using tools built elsewhere, she said.

Both ecosystems are selling hard

Huawei has spent 2026 pitching an integrated AI offering to government and enterprise buyers, most recently at its Connect event in July, and its cloud platform offers DeepSeek models to African developers. DeepSeek’s V4, released in April, was built to run on Huawei’s Ascend silicon.

The American side is further along. Microsoft has committed R5.4-billion to AI and cloud infrastructure here on top of more than R20-billion already spent. Amazon Web Services, Google and Oracle all run local cloud regions. Altron’s Nvidia-powered AI factory is live, and Cassava Technologies is building its own.

The lock-in is technical as well as political. Code written for Nvidia’s Cuda software cannot simply be moved onto Huawei’s Ascend chips, BCG noted, because most production AI code would need rewriting for Huawei’s Cann environment.

A test South Africa has already lived through

The sharpest test of the minister’s position is one South Africa — and much of the world — has already experienced.

Anthropic released Claude Fable 5 and Claude Mythos 5 on 9 June, the first models in a tier it says exceeds its Opus line. Fable 5 was the general release, carrying safeguards; Mythos 5, with fewer, went only to a small group of partners in the company’s Project Glasswing for defensive cybersecurity work. Three days later both were gone.

On 12 June, after Amazon researchers found a way to prompt Fable 5 into identifying software vulnerabilities, according to TechCentral’s reporting, the US commerce department directed Anthropic to suspend all access by any foreign national, inside or outside the US. Because the order took effect immediately and the company had no way to verify nationality in real time, it withdrew both models from every user worldwide. South African users lost access overnight.

The controls were lifted on 30 June and Fable 5 returned globally on 1 July — an outage of about three weeks.

BCG read the episode as evidence that national security concerns may in some instances take precedence over the global diffusion of the most capable AI models. The decision was taken in Washington and reversed there, not by any buyer in Pretoria applying value-for-money and fit-for-purpose tests.

Policy vacuum

South Africa is making these choices without a policy to guide them. The draft national AI policy was withdrawn in April over fictitious references, an expert panel chaired by Wits AI researcher Benjamin Rosman is rebuilding it, and Morwane has signalled the redraft will be light-touch — providing national guidelines and leaving sectors to write their own strategies. She has put the redraft at cabinet in November, with January 2027 as the fallback.

TechCentral has previously reported that the withdrawn draft left South Africa too dependent on both superpowers. Malatsi’s answer is that dependency is a false choice, and that open standards and vendor neutrality are enough to keep the country’s options open.

The World Bank’s World Development Report 2026 — which excluded South Africa from its map of countries with national AI strategies precisely because of the April withdrawal — agrees that is the right instinct, recommending that developing economies make interoperability a feature instead of aiming for sovereignty. Where it goes further is on what the instinct requires: interoperability is something a country has to specify and enforce, not something it gets by declining to choose.

BCG’s warning is that the options themselves are being foreclosed. The walls between the US and Chinese AI technology stacks, it said, are going up faster than most companies are reorganising around them.