LLMs as Instruments of Statecraft: Digital Silk Roads, Export Controls, and the Geopolitical Risks of AI for Africa

Large language models (LLMs) are increasingly deployed as strategic assets rather than neutral technical artefacts, and the consequences of this shift fall unevenly on countries that possess neither sovereign compute nor sovereign training corpora. This paper examines two compounding mechanisms through which frontier AI is becoming an instrument of geopolitics, and asks what each implies for Africa specifically. The first mechanism is upstream: training corpora dominated by the media and cultural output of a small number of states embed those states' epistemologies into model weights, with recent large scale audits showing that government control of media measurably shapes the political valence of LLM outputs. The second mechanism is downstream: access to frontier models can be suspended by executive action in the country where the developer is incorporated, as illustrated by the June 2026 United States export control directive that forced Anthropic to disable its Fable 5 and Mythos 5 models worldwide for eighteen days. We situate both mechanisms within the material infrastructure of China's Digital Silk Road, which now underwrites much of the fibre, data centre, and cloud capacity on which African AI adoption depends, and we contrast this with the more replicable, exportable model of African-led initiatives such as Lelapa AI's InkubaLM. We further argue that the risks of AI for Africa cannot be assessed in isolation from its benefits, particularly job creation and digital public infrastructure, and that recent United Nations reporting on the water and energy footprint of AI data centres sharpens rather than resolves this trade-off. We conclude with governance directions for the African Union and its member states, including sovereign compute investment, multilateral access frameworks, and mandatory training data provenance disclosure.