Launching chokepoints – mapping the bottlenecks in the AI infrastructure stack

- Chokepoints.ai is a new site that maps the key bottlenecks in the global AI infrastructure supply chain — the physical hardware, materials, and manufacturing steps needed to build and run AI systems (not just software). - The AI infrastructure stack spans chip design (Nvidia, AMD), chip fabrication (TSMC, Samsung), memory and packaging (SK Hynix, Samsung, ASE), networking (Broadcom), and raw materials (rare earths, advanced substrates). Each layer has only a handful of companies worldwide that any AI project must pass through. - The core argument: because AI hardware manufacturing is extremely concentrated (some steps have just 1-2 viable suppliers worldwide), these "chokepoints" create single points of failure for the entire industry — vulnerable to geopolitics, export controls (US vs China), natural disasters, or business decisions by a single firm. - The site acts as a public reference and watchdog, tracking who controls each layer and what risks exist, similar to how supply-chain mapping has been done for semiconductors more broadly.