Essay - 2026-06-25

AI's energy bottleneck: the race for delivered power

A GPU reservation is only a promise. Capacity becomes real when turbine slots, transformers, interconnection studies, switchgear, cooling loops and crews converge at the same site.

AI Bottlenecks is a long letter about why the AI buildout rests on physical chokepoints, not just models, clouds, and launch demos.

The thesis tracks the narrow places under the visible AI race: indium phosphide wafers, advanced packaging, memory bandwidth, power equipment, cooling, and the critical minerals that make the stack possible.

Almost everyone can name the hyperscalers. Far fewer people can name the suppliers whose output determines whether a trillion dollars of capex becomes usable compute. That asymmetry of attention is the opportunity.

The project is not a portfolio or a recommendation. It is a public map of the supply chain so curious readers, builders, investors, engineers, and policy people can walk the chain themselves.