Half the AI supercycle doesn't trade in New York

The AI supercycle has a US-listed face — $NVDA, $AMD, $AVGO, $MSFT, the hyperscaler cluster — and a non-US backbone that almost nobody trades because their broker can't reach it.

That second half is where most of the supply is constrained. The chips can't be designed without ASML's lithography. The chips can't be fabbed without TSMC's process nodes. The accelerators can't run without HBM stacks from Samsung and SK Hynix. The fabs can't be tooled without Tokyo Electron, Hitachi, Disco, Advantest. None of those companies are listed in New York as primary issues.

If you can only buy what your broker can route to, your "AI supercycle exposure" is by default a US-equities exposure to the demand side of the chain. The supply side — where the genuine bottlenecks are — is offshore. And the bottlenecks are where capital flows price-insensitively when demand spikes.

This article is why we trade that offshore leg on Interactive Brokers, what the operational reality looks like, and where the tradeoffs are.

The four bottleneck stocks that aren't US-listed

Walk the AI supply chain upstream from a hyperscaler's GPU purchase order:

1. EUV lithography — $ASML. Listed primary on Euronext Amsterdam (ticker: ASML.AS). The ADR (also ASML) trades on Nasdaq, but it's the secondary listing. There is no second supplier for high-NA EUV. ASML's order book is the leading indicator for every fab buildout 18 months forward.

2. Leading-edge logic fab — TSMC. Primary listing on Taiwan Stock Exchange (2330.TW). The US ADR ($TSM) tracks well, but the primary is where the Taiwanese institutional flow happens and where dividends and rights issues clear without the depositary friction. TSMC produces ~90% of leading-edge logic globally.

3. HBM memory — Samsung Electronics (005930.KS) and SK Hynix (000660.KS). Both list primary on the Korea Exchange. Neither has a clean US ADR. Samsung trades GDRs in London (SMSN.LN) and some pink-sheet variants that are illiquid and tax-messy. SK Hynix has no major US listing at all. HBM3E and HBM4 supply is the single tightest constraint in the 2025-2027 AI buildout — and most US retail can't trade either name directly.

4. Fab equipment — Tokyo Electron (8035.T), Disco (6146.T), Advantest (6857.T), Hitachi (6501.T). Primary on Tokyo Stock Exchange. A handful have ADRs of varying quality; the rest don't.

Live mapping: these names cluster inside the semi-equipment and memory bubbles in our taxonomy — both of which validate empirically at residualized correlation north of 0.70. The supply-chain bottlenecks trade as a bloc. The US-listed proxies don't always.

The ADR tax: why "just buy the ADR" is a worse trade

The default US-retail answer is "buy the ADR." For names that have a clean ADR — TSM is the canonical example — this is fine for buy-and-hold, ugly for everything else. For names that don't — Samsung, SK Hynix — it isn't an option at all.

What the ADR costs you, in order of magnitude:

• Depositary fees. Typically $0.01–$0.05 per share per year, deducted from dividends. Visible only if you read the prospectus.
• FX spread baked into pricing. ADRs reprice intraday against the primary, but the depositary's FX rate and timing extracts a few bps per round-trip. It compounds on active books.
• Tracking divergence. ADRs and primaries usually trade tight, but they decouple under stress — earnings nights, regional holidays, dividend record dates. The basis can be 0.5-2% on those days. Bad if you're trying to hedge or arb across the pair.
• No options on most of them. Samsung has no listed US options. SK Hynix has no US listing at all, so no options either. If your AI thesis includes hedging memory exposure with puts during a cycle peak, the ADR path can't express it.
• Tax friction. Korean and Taiwanese dividends get a withholding tax that flows through the ADR opaquely. For taxable accounts, the reconciliation at year-end is a chore.

The ADR makes sense when you want passive US-tax-domiciled exposure and don't care about basis or hedging. For everything else — active sizing, options hedging, regional event trading, the names that simply don't have a clean ADR — you need direct access to the primary listing.

What IBKR actually gets you

Interactive Brokers' value proposition for this specific use case is jurisdictional reach without a separate account per exchange. One account, one funding line, one tax statement, direct exchange routing to:

• TSE (Taiwan Stock Exchange) — TSMC primary, plus the Taiwanese semiconductor cluster
• KRX (Korea Exchange) — Samsung, SK Hynix, Hanmi Semiconductor (HBM packaging), the Korean equipment names
• TSE (Tokyo Stock Exchange) — Tokyo Electron, Disco, Advantest, Hitachi, plus the JREITs that own datacenter land
• AEX (Euronext Amsterdam) — ASML primary, plus the European semi cluster
• LSE, ETR, SIX, HKEX, ASX — the rest of the international supply chain when you need it (rare earths, power, networking)

The order entry is the same TWS or API surface you'd use for a US trade — you just append the exchange route (SMART doesn't help here; you specify e.g. TSEJ for Tokyo, KSE for Korea, TWSE for Taiwan). For the Python API, the contract specification carries exchange and primaryExchange fields and the rest behaves as expected.

A concrete example. To buy SK Hynix at IBKR:

Symbol:           000660
Exchange:         KSE
Primary Exchange: KSE
Currency:         KRW
SecType:          STK

Your account holds KRW positions after the fill, marked to USD on your statement at IBKR's wholesale FX rate (which is genuinely close to interbank — that's the part where IBKR's pricing meaningfully beats the consumer broker tier). When you sell, you can hold KRW for the next trade or convert back to USD at your discretion.

There is no second US-retail broker that does this. Schwab's international desk exists but is phone-driven and not algorithmically accessible. Fidelity offers a handful of foreign markets but not the Korean primary and not on the algo path. The crypto-broker-fintech tier (Public, Robinhood, eToro) is US-only or ADR-only across the board.

What we actually trade there

For QuantAbundancia, the IBKR leg carries the international supply chain and a slice of the US tradfi book. The split:

• International AI supply chain — TSMC primary, Samsung Electronics, SK Hynix, Tokyo Electron, Disco. Long-only conviction sizes. Held against the corresponding US-listed counterpart positions to hedge the basis when relevant.
• ASML primary on AEX, paired against US-listed semi-equipment when the basket trade calls for it.
• US-listed equities, ETFs, options for the slice of the AI supercycle that does live in NY — paid for at IBKR margin rates (competitive at retail scale), API-routed via the same gateway as the international leg.
• Paper trading for every tradfi bot in the bot fleet before any of them touches real capital. IBKR's paper account is on the same gateway, the same fills model, and the same data subscriptions as the live account — what passes paper has a real chance of surviving live.

The OKX side of the stack handles the crypto leg of the same playbook. Tradfi exposure to AI supply chain stays at IBKR.

Honest tradeoffs

This isn't a sales pitch. IBKR has real friction:

• The UI is built for institutions. TWS has a learning curve. If you never plan to use the API, the experience is worse than a modern consumer app. The desktop client is the only path to some order types; the mobile app is a triage tool, not a primary surface.
• Data subscriptions are à la carte. Real-time depth, regional bundles, exchange feeds — every line item is opt-in and priced separately. Budget $10-30/month if you want real-time on the international venues; the default at signup is delayed data and ugly defaults.
• Account funding is a multi-day affair. ACH wires clear faster than they used to but international funding still tests your patience. If you need a position by tomorrow, this isn't the broker to open today.
• Margin withdrawal restrictions on non-US-resident personal accounts. Worth checking against your own residency before you fund.
• Customer support is slow and ticket-driven. Fine if you operate the thing yourself; not fine if you want hand-holding.
• IB Gateway has its quirks. Nightly reauth windows, occasional IBC re-login loops, the "unhealthy but Up" container state we monitor for. None are dealbreakers if you're running bots; some are dealbreakers if you're not technical.

None of these change the core fact: for direct exchange access to TSMC, ASML, Samsung, SK Hynix, and the rest of the offshore AI supply chain from a single account, the practical retail option is IBKR. The fintech tier doesn't compete.

How to actually act on this

Three concrete steps if the thesis lands for you:

1. Audit your current AI book. Pull every position you hold in the name of "AI exposure" and tag it as demand-side US (NVDA, AVGO, MSFT, etc.) vs supply-side offshore (TSMC, ASML, Samsung, SK Hynix, Tokyo Electron). If you find the second bucket is empty or all ADRs, you have a single-sided book by accident.

2. Decide whether you need the primary listings or whether ADRs are enough. If you're never going to hedge with options, never going to trade event-driven, and only own TSM and ASML (which have clean ADRs), you're fine staying US-domiciled. If you want HBM exposure (Samsung, SK Hynix), there is no clean US path — direct access is the only path.

3. If you need the access, Interactive Brokers is what we use. The setup is friction-heavy for a week and pays back forever. Full reasoning and the honest list of what we don't like is on the broker page.

The AI supercycle has been the trade of this decade. Half of it ships from places most retail traders can't reach. The retail brokers that own that gap charge a tax on it. The one that closes the gap doesn't.