#method

How this dataset is built and what its tags mean.

• The eleven parts that only show up above hyperscale2026-09-09

  Eleven of this catalog's 130 part categories never appear below campus scale. Ten of them are power. One is optics.

• Half the supplier bench, a sixth of the coverage2026-09-07

  Cooling, power distribution and generation hold 43% of this catalog's supplier placements and drew 16% of three weeks of trade coverage.

• Eight parts of an AI data center have no US-listed supplier at all2026-09-04

  163 parts here have three or more listed suppliers. On eight, not one trades in the US — and six sit either directly under the GPU or directly around it.

• Eighteen companies cross four layers of this stack. 984 never leave one.2026-09-03

  984 of the 1,163 companies on a part here sit in exactly one of the eleven layers. Eighteen reach four or more, and two pairs occupy identical parts.

• The further you get from the chip, the less of the supply chain is listed2026-09-02

  652 of the 1,164 companies on a part here are listed: 87% of the rack layer, 41% of the building shell, and four parts with no listed supplier at all.

• What 'reported' means on this site2026-09-02

  Our three-tier confidence system, why most edges start as leads, and how to send a correction.

• The 17 parts where only one listed company shows up2026-09-01

  A query for parts with exactly one publicly listed supplier returns 17 rows. Every one of those 17 edges is an unsourced lead — which is the actual finding.

• Who depends on whom: reading customer concentration in SEC filings2026-08-31

  197 concentration disclosures from 101 suppliers. Two name the customer. Here is what the other 195 are still worth.