Artificial intelligence is often portrayed as a contest of algorithms and model capabilities. This working paper argues that the strategic foundations of AI competition increasingly lie in control over compute infrastructure rather than software alone. Introducing the concept of AI chokepoint geopolitics, the paper examines how advanced semiconductors, cloud computing capacity, fabrication ecosystems, energy infrastructure, and export controls have become central instruments of geopolitical power. Drawing on evidence from organisations including the OECD, Stanford HAI, the World Bank, Reuters, and Nature, it explores how the concentration of AI infrastructure reshapes technological sovereignty, industrial policy, and strategic competition, particularly in the context of the United States, China, and Taiwan. The paper further considers the implications for the Global South, arguing that while AI adoption is becoming increasingly widespread, frontier AI development remains constrained by unequal access to compute resources. It concludes that AI should be understood not merely as a software revolution but as an industrial transformation in which control over computational infrastructure increasingly determines participation in the next generation of intelligent systems.
Karan Aggarwal (Tue,) studied this question.