The AI-infrastructure supply chain: an investor's map

From power and cooling to interconnect, the bottlenecks in AI infrastructure are physical. We map the supply chain and where durable value is likely to accrue.

Most of the attention in AI goes to models. Most of the constraints are physical — power, cooling, land, and the components that move data between chips.

Following the bottleneck

Durable value tends to settle where capacity is hardest to add: grid interconnection, high-density cooling, and the specialised equipment with long lead times. We map the chain from substation to server and flag where pricing power is structural rather than cyclical.

This is a placeholder market-view draft. Replace with the finalised article before launch.