Reading this catalog like an investor

This site is organized around the physical part, not the company. A part like hbm or a substation large power transformers (345–765 kv) & substations is the fixed unit; companies hang off it through a table that records, for every part-company pair, a role, a source, and a confidence level. Everything else on the site — company pages, layer pages, the cross-cutting views — is a query over that one table. That’s a useful structure, but it rewards being read a particular way. Skim for familiar logos and you’ll miss most of the signal, which lives in the tags sitting next to each name.

Parts first, companies second

Read a part page as the primary document and a company page as a rollup. A part page tells you what the component does, why AI demand changes it (the “AI delta”), whether it currently carries a bottleneck status, and which companies sit on it with what role — designer, manufacturer, ODM, distributor, or in-house. A company page just aggregates every part row that company touches, plus its listing data and ownership notes. If you start from a company and work outward, you’ll tend to see breadth; if you start from a part and work outward, you see the actual competitive set for that specific step in the chain, which is usually the more useful read.

How the confidence tags work

Every company sitting on a part carries one of three tags. Disclosed means the company itself, or a regulatory filing, says so, with a source URL attached. Reported means the placement comes from trade press or a reference-design partner list — credible, but not a company disclosure. Inferred means the placement comes from the taxonomy seed with no citation attached yet; treat it as a lead to verify, not a confirmed fact. Tags are never upgraded without a source behind them, which means an inferred tag sitting next to a name for months isn’t an oversight — it’s an honest statement that nobody has sourced that edge yet.

Sole-supplier flags

Some parts have only one company sitting on them among the ones with public equity. A company page marks that explicitly as a sole listed supplier — a flag computed by counting public companies only, so a part can carry the flag even if private companies also make it; the flag is about what an investor can actually buy, not about total supplier count. The taxonomy itself sometimes goes further and calls a name out as effectively sole-sourced across the whole market: substrate materials notes that the ABF film used in chip substrates is effectively sole-sourced from one supplier, and grain-oriented electrical steel names a sole US producer of the steel that goes into transformer cores. Those are two different, worth-distinguishing claims — “only listed supplier” and “only supplier, period” — and the catalog is careful to keep them separate rather than blur them into one badge.

Why breadth is not revenue exposure

A company can show up on dozens of part rows across several layers without AI being material to what it earns. Placement count is breadth, not revenue exposure — it tells you where a company’s products are technically present in the build-out, not how much of its income comes from being there. passives is a useful example of why that gap matters: the taxonomy notes that AI hardware needs roughly 10–15× the passive component count of a standard server, with high-spec MLCC lead times running around 24 weeks as of mid-2026 trade press. That’s a real, sourced demand signal — and the handful of companies making those parts also sell into phones, cars, and every other electronics category, so the same breadth that makes them show up across this catalog also means AI is one demand driver among several. A company page states its disclosed data-center revenue share where one exists, and says plainly when none is on file. Check that line before treating a placement as an investment thesis on its own.

Three views that cut across parts

Three pages exist specifically to let you slice the catalog a different way than layer-by-layer.

Exchanges groups every company by listing venue. A large share of this supply chain trades outside the US — Taiwan, Korea, and Japan carry an outsized share of names in packaging, substrates, and passives specifically — and this view surfaces tickers a US-only screen would never catch.

Scales is a matrix of parts against facility size, from small builds up through gigawatt-class campuses, with a filter for cooling density. It answers a different question than a layer page does: not “who makes this part” but “does this part even exist at the size of build I care about.” Some parts, like the merchant GPU itself, appear at every tier; others, like a dedicated substation transformer, only show up once a build crosses a size threshold.

Bottlenecks lists every part currently tagged constrained or tight, each with its sourced companies attached. The important caveat sits in the fine print: a part with no tag is simply unassessed, not confirmed unconstrained. Silence on that page is not the same as an all-clear.

A short way to use it

A reasonable path through the catalog starts at a part page, not a company page. Check its bottleneck status and the confidence tags on the companies listed there. Check whether a company carries a sole-supplier flag, and whether that’s a listed-only claim or a stronger, sourced statement from the taxonomy itself. Check the scale-tier matrix to see how broadly the part is actually used, since a part that appears only at gigawatt scale carries different risk than one every tier depends on. Then, before any of that becomes a thesis about a specific company, go read its exposure note — breadth earns a name a spot on this catalog; disclosure is what tells you whether that spot is material.

Where to go next

• hbm — a part tagged constrained, with sourced companies attached.
• Substrate materials — an example of an effectively sole-sourced input.
• Passives — a case where breadth is wide and revenue exposure varies company to company.