AI safety · concentration, fragility, and leverage in the physical stack
Eight layers from materials to cloud. Who holds each chokepoint, what breaks without it, and where a few hands can restrict frontier AI.
Every AI system runs on a physical supply chain: chips, tools, materials, and cloud capacity. A handful of companies and countries control most of it. That concentration is where AI safety meets hardware. Each chokepoint is two things at once: a bottleneck that can disrupt several downstream layers, and a control point where a few hands can restrict or prioritize who builds frontier AI, as export controls already show. Culm maps where that leverage sits: who holds each layer, where the chokepoints are, and what breaks if one is removed. Sourced, neutral, confidence-rated. Culm reports facts; what to do about them is your call.
Frontier AI runs on hardware with extreme concentration: a few firms, countries, and chokepoints sit between raw materials and the compute that trains the largest models. That concentration creates fragility (one removal can stall the stack) and leverage (export controls and supply restrictions already target these layers). Culm maps where both sit, with sourced numbers you can audit.
Explore each layer in the stack, or switch to What if? to simulate removal.
