Most of the debate over the US-China AI split is written from inside the two combatants. A new report from the BCG Institute, "The Great Divide," takes the view from the countries that have to live with the outcome without a vote in it. Its argument is blunt: as America and China build increasingly distinct and incompatible AI stacks, the choice of which model to adopt is quietly becoming a geopolitical one. For a bank in Johannesburg or a health ministry in Nairobi, picking a foundation model is starting to look less like a procurement decision and more like picking a bloc.
The economics, for now, tilt hard one way. "Right now, the Chinese AI stack offers a compelling economic proposition," said BCG's Dawie Scholtz, with some models delivering 80 to 90 percent of the performance at roughly 10 percent of the cost of American alternatives. That is a seductive number for a continent where AI is expected to underpin public services, financial inclusion, health systems and logistics. China's pitch is reinforced by relationships already on the ground: it is the primary trading partner of 78 countries in the Global South and has built, financed or operated more than a third of Africa's commercial ports. The cheaper, faster-to-deploy option also happens to be the one that fits the infrastructure already there.
The catch is what "stack" means. The report's central point is that the race is no longer only about who has the most powerful model. It is about who controls the full chain, from chips and cloud through foundation models, applications, data governance and security protocols. The United States still leads on that supply side, powered by capital and talent: US startups have raised around $380 billion in AI venture funding since 2023, and the capital spending of the top American tech firms passed $400 billion in 2025 against $63 billion in China. China's counter is a cheaper, more self-reliant stack built around open-weight models and domestic chips. The two are drifting toward mutual incompatibility, and, the report warns, "mixing and matching across the two ecosystems will likely become increasingly difficult." Choosing a model may eventually mean choosing the whole stack behind it.
That is where today's bargain can turn into tomorrow's trap. With both Washington and Beijing weighing tighter controls on advanced models, an organisation that has wired itself entirely into one ecosystem is exposed to being cut off from critical tools by a decision made in a foreign capital. The danger the report highlights is not only choosing the "wrong" stack, but building systems too rigid to switch when the geopolitical weather changes.
BCG's prescription is less about which side to pick than about refusing to fully pick one. It names three defences: redundancy, keeping fallback options across models, cloud and compute so no single layer is a single point of failure; modularity, designing systems so one layer can be swapped without rebuilding everything; and heterogeneity, deliberately mixing providers and model types to avoid overexposure to one bloc. It is unglamorous, expensive engineering advice, and it cuts against the pull of the cheapest all-in-one deal. But it reframes the question African leaders actually face. The decision is not simply how fast to adopt AI. It is how to adopt it without handing a distant government a switch it can one day flip.