The AI hardware trade looks investable from the outside. Nvidia, TSMC, ASML, the US equipment makers: liquid, US-listed, covered by everyone. But walk the supply chain underneath those names, stage by stage, and ask the same two questions at each one, who actually controls this stage and can I own them, and the second answer is usually no.
Start with the stage everyone knows. ASML has a monopoly on EUV lithography and trades as a clean US ADR, which I think is part of why people assume the rest of the chain works the same way. It does not, and you don’t even have to leave ASML’s own machine to see it. The EUV optics come from Carl Zeiss SMT and the laser source comes from TRUMPF, both sole-source, both private. Two of the most concentrated chokepoints in the entire chain have no ticker at all, on any exchange, in any currency.
One step over: every EUV mask blank has to pass through inspection tools that Lasertec alone makes. Lasertec trades over the counter as a thin unsponsored line, and the real fundamentals live in Tokyo Stock Exchange filings, on a Japanese fiscal calendar.
Advanced packaging runs on ABF, an insulating film with a near-monopoly supplier: Ajinomoto, a Japanese food seasoning company. Also a thin OTC line. Silicon wafers are effectively three companies, Shin-Etsu, SUMCO, and GlobalWafers. The first two are thin OTC lines and the third trades only in Taiwan.
Then memory. High-bandwidth memory is the input AI accelerators are actually starved for, and the leader is SK Hynix. There is no US listing. The common shares trade in Seoul, and the OTC quotes that exist are thin unsponsored lines. The investable substitute is Micron, and Micron is a follower in HBM, so the tradeable name lags what the leader has already reported in Seoul. Samsung, the other half of the HBM story, is Seoul-only too. Same picture in optics: the largest optical transceiver maker in the world, Zhongji InnoLight, trades only in Shenzhen. A US brokerage account cannot reach it.
I went through the whole chain this way, minerals, gases, wafers, lithography, process equipment, test, foundry, memory, packaging, passives, optics, power, assembly, and listed the foreign companies that matter at each stage. The first pass, just the names big enough to be obvious, comes to a bit over ninety. About ten trade as clean US-listed ADRs. More than half are thin OTC lines. Roughly thirty trade only on their home exchange, with no US line and no SEC filings. A couple have no equity at all.
Then I went deeper, sub-niche by sub-niche: substrates, laminates, probe cards, mask blanks, dicing, thermal, grid equipment. That pass turned up a few hundred more companies, and the ratio gets worse, not better. The overwhelming majority trade only in Tokyo, Seoul, Taipei, Shenzhen, or Shanghai. And the private bucket grows too: SK Siltron in wafers, SCHOTT in specialty glass, Isola in laminates, group-owned or foundation-owned or private-equity-owned, no ticker.
The thin-OTC bucket is also less tradeable than it sounds, and this part I measured. For each OTC line, count the days in a month it actually printed a trade. A healthy line prints every session, about 24 trading days. Horiba’s US line printed one. Anritsu’s, one. Pfeiffer Vacuum’s, two. ULVAC’s, three. These are not obscure companies; they are instrumentation and vacuum suppliers the fabs cannot run without. Some lines turned out to be quote-only, a price on the screen with no trades behind it at all. And the same company often has two US symbols, one of which is quietly dead: Resonac’s commonly cited line traded eight days a month while its other symbol traded every session. An unsponsored F-share gives you a quote, not a market: liquidity that disappears when you want size, no EDGAR filings, financials you reconstruct from Tokyo or Frankfurt disclosures, fiscal years that don’t line up with anything US. It is tradeable in name only.
The pattern that interests me is the direction. The more important a company is to the chain, the less likely you can own it. Ownership access concentrates at the visible ends of the chain, the chip designers and the equipment brands, while control concentrates in the middle: private German optics, Seoul memory, Shenzhen transceivers, Japanese films and chemicals quoted on lines nobody can trade in size. So “investing in the AI supply chain” from a US account collapses into buying the same dozen liquid names as everyone else, while the scarcity everyone is theoretically paying for sits in companies the market can barely price.