One data centre
Big Tech's AI money passes through three floors before it stops, and you have never heard of the bottom one.
A data centre bill does not go to one place. It goes to a floor of companies, who pay the floor below, who pay the floor below that.
One step from the cheque: Nvidia and Dell for the machines, Arista for the network, Vertiv for power and cooling, GE Vernova for the turbines, Quanta Services to build it. These are direct relationships, not inferences — Microsoft alone was 26% of Arista's 2025 revenue and Meta about 15%.
Two steps: the chip and memory makers those companies buy from. Nvidia pays TSMC to print the logic and Samsung, SK hynix and Micron for the memory stacked beside it.
Three steps: the five firms that sell the fabs their machines — ASML for lithography, Applied Materials, Lam Research and Tokyo Electron for deposition and etch, KLA for inspection — plus Shin-Etsu, which makes the blank silicon they all start from.
What the picture actually shows: it does not fan out. Six names, then four, then six. The narrow floor is the middle one, and that is the whole reason a memory or packaging shortage moves the price of everything above it.
Each floor here is a subset, not a census.
